Hugging Face's logo

Spaces:
RaynWu2002
/
DORNet
Sleeping

App Files Community
1
RaynWu2002 commited on
Commit
68c537d
·
verified ·
1 Parent(s): 7d27cb0

Upload 49 files

Browse files
Files changed (50) hide show
  1. .gitattributes +5 -0
  2. Figs/Params_Time.png +3 -0
  3. Figs/Pipeline.png +3 -0
  4. Figs/RGBDD.png +3 -0
  5. LICENSE +201 -0
  6. README.md +137 -14
  7. app.py +156 -0
  8. checkpoints/NYU_T_X16.pth +3 -0
  9. checkpoints/NYU_T_X4.pth +3 -0
  10. checkpoints/NYU_T_X8.pth +3 -0
  11. checkpoints/NYU_X16.pth +3 -0
  12. checkpoints/NYU_X4.pth +3 -0
  13. checkpoints/NYU_X8.pth +3 -0
  14. checkpoints/RGBDD.pth +3 -0
  15. checkpoints/RGBDD_Noisy.pth +3 -0
  16. checkpoints/RGBDD_Noisy_T.pth +3 -0
  17. checkpoints/RGBDD_T.pth +3 -0
  18. checkpoints/TOFDSR.pth +3 -0
  19. checkpoints/TOFDSR_Noisy.pth +3 -0
  20. checkpoints/TOFDSR_Noisy_T.pth +3 -0
  21. checkpoints/TOFDSR_T.pth +3 -0
  22. data/TOFDSR_Test.txt +560 -0
  23. data/TOFDSR_Train.txt +0 -0
  24. data/__pycache__/nyu_dataloader.cpython-311.pyc +0 -0
  25. data/__pycache__/rgbdd_dataloader.cpython-311.pyc +0 -0
  26. data/nyu_dataloader.py +47 -0
  27. data/rgbdd_dataloader.py +118 -0
  28. data/tofdc_dataloader.py +90 -0
  29. examples/RGB-D-D/20200518160957_LR_fill_depth.png +0 -0
  30. examples/RGB-D-D/20200518160957_RGB.jpg +0 -0
  31. examples/TOFDSR/2020_09_08_13_59_59_435_rgb_depth_crop_fill.png +0 -0
  32. examples/TOFDSR/2020_09_08_13_59_59_435_rgb_rgb_crop.png +3 -0
  33. net/CR.py +63 -0
  34. net/__pycache__/CR.cpython-311.pyc +0 -0
  35. net/__pycache__/deform_conv.cpython-311.pyc +0 -0
  36. net/__pycache__/dornet.cpython-311.pyc +0 -0
  37. net/__pycache__/dornet_ddp.cpython-311.pyc +0 -0
  38. net/deform_conv.py +75 -0
  39. net/dornet.py +586 -0
  40. net/dornet_ddp.py +600 -0
  41. test_img.py +17 -0
  42. test_img/RGB-D-D/20200518160957_LR_fill_depth.png +0 -0
  43. test_img/RGB-D-D/20200518160957_RGB.jpg +0 -0
  44. test_img/TOFDSR/2020_09_08_13_59_59_435_rgb_depth_crop_fill.png +0 -0
  45. test_img/TOFDSR/2020_09_08_13_59_59_435_rgb_rgb_crop.png +3 -0
  46. test_nyu_rgbdd.py +103 -0
  47. test_tofdsr.py +66 -0
  48. train_nyu_rgbdd.py +158 -0
  49. train_tofdsr.py +167 -0
  50. utils.py +37 -0
.gitattributes CHANGED
@@ -33,3 +33,8 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ examples/TOFDSR/2020_09_08_13_59_59_435_rgb_rgb_crop.png filter=lfs diff=lfs merge=lfs -text
37
+ Figs/Params_Time.png filter=lfs diff=lfs merge=lfs -text
38
+ Figs/Pipeline.png filter=lfs diff=lfs merge=lfs -text
39
+ Figs/RGBDD.png filter=lfs diff=lfs merge=lfs -text
40
+ test_img/TOFDSR/2020_09_08_13_59_59_435_rgb_rgb_crop.png filter=lfs diff=lfs merge=lfs -text
Figs/Params_Time.png ADDED

Git LFS Details

  • SHA256: 62889c8bdc8f548508fd388b9a2fd1bba2e48c3aea8915ff278cbcccb123f62b
  • Pointer size: 132 Bytes
  • Size of remote file: 5.55 MB
Figs/Pipeline.png ADDED

Git LFS Details

  • SHA256: 228f461e4556a94f6bf73a01b8aa9419f014a1e8b771d11fda965b51fed6c8b0
  • Pointer size: 132 Bytes
  • Size of remote file: 3.08 MB
Figs/RGBDD.png ADDED

Git LFS Details

  • SHA256: c357585db78aeba83e448119206c8b1337432a4d43653b657c05dafaee2b2b2f
  • Pointer size: 133 Bytes
  • Size of remote file: 11 MB
LICENSE ADDED
@@ -0,0 +1,201 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Apache License
2
+ Version 2.0, January 2004
3
+ http://www.apache.org/licenses/
4
+
5
+ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
6
+
7
+ 1. Definitions.
8
+
9
+ "License" shall mean the terms and conditions for use, reproduction,
10
+ and distribution as defined by Sections 1 through 9 of this document.
11
+
12
+ "Licensor" shall mean the copyright owner or entity authorized by
13
+ the copyright owner that is granting the License.
14
+
15
+ "Legal Entity" shall mean the union of the acting entity and all
16
+ other entities that control, are controlled by, or are under common
17
+ control with that entity. For the purposes of this definition,
18
+ "control" means (i) the power, direct or indirect, to cause the
19
+ direction or management of such entity, whether by contract or
20
+ otherwise, or (ii) ownership of fifty percent (50%) or more of the
21
+ outstanding shares, or (iii) beneficial ownership of such entity.
22
+
23
+ "You" (or "Your") shall mean an individual or Legal Entity
24
+ exercising permissions granted by this License.
25
+
26
+ "Source" form shall mean the preferred form for making modifications,
27
+ including but not limited to software source code, documentation
28
+ source, and configuration files.
29
+
30
+ "Object" form shall mean any form resulting from mechanical
31
+ transformation or translation of a Source form, including but
32
+ not limited to compiled object code, generated documentation,
33
+ and conversions to other media types.
34
+
35
+ "Work" shall mean the work of authorship, whether in Source or
36
+ Object form, made available under the License, as indicated by a
37
+ copyright notice that is included in or attached to the work
38
+ (an example is provided in the Appendix below).
39
+
40
+ "Derivative Works" shall mean any work, whether in Source or Object
41
+ form, that is based on (or derived from) the Work and for which the
42
+ editorial revisions, annotations, elaborations, or other modifications
43
+ represent, as a whole, an original work of authorship. For the purposes
44
+ of this License, Derivative Works shall not include works that remain
45
+ separable from, or merely link (or bind by name) to the interfaces of,
46
+ the Work and Derivative Works thereof.
47
+
48
+ "Contribution" shall mean any work of authorship, including
49
+ the original version of the Work and any modifications or additions
50
+ to that Work or Derivative Works thereof, that is intentionally
51
+ submitted to Licensor for inclusion in the Work by the copyright owner
52
+ or by an individual or Legal Entity authorized to submit on behalf of
53
+ the copyright owner. For the purposes of this definition, "submitted"
54
+ means any form of electronic, verbal, or written communication sent
55
+ to the Licensor or its representatives, including but not limited to
56
+ communication on electronic mailing lists, source code control systems,
57
+ and issue tracking systems that are managed by, or on behalf of, the
58
+ Licensor for the purpose of discussing and improving the Work, but
59
+ excluding communication that is conspicuously marked or otherwise
60
+ designated in writing by the copyright owner as "Not a Contribution."
61
+
62
+ "Contributor" shall mean Licensor and any individual or Legal Entity
63
+ on behalf of whom a Contribution has been received by Licensor and
64
+ subsequently incorporated within the Work.
65
+
66
+ 2. Grant of Copyright License. Subject to the terms and conditions of
67
+ this License, each Contributor hereby grants to You a perpetual,
68
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
69
+ copyright license to reproduce, prepare Derivative Works of,
70
+ publicly display, publicly perform, sublicense, and distribute the
71
+ Work and such Derivative Works in Source or Object form.
72
+
73
+ 3. Grant of Patent License. Subject to the terms and conditions of
74
+ this License, each Contributor hereby grants to You a perpetual,
75
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
76
+ (except as stated in this section) patent license to make, have made,
77
+ use, offer to sell, sell, import, and otherwise transfer the Work,
78
+ where such license applies only to those patent claims licensable
79
+ by such Contributor that are necessarily infringed by their
80
+ Contribution(s) alone or by combination of their Contribution(s)
81
+ with the Work to which such Contribution(s) was submitted. If You
82
+ institute patent litigation against any entity (including a
83
+ cross-claim or counterclaim in a lawsuit) alleging that the Work
84
+ or a Contribution incorporated within the Work constitutes direct
85
+ or contributory patent infringement, then any patent licenses
86
+ granted to You under this License for that Work shall terminate
87
+ as of the date such litigation is filed.
88
+
89
+ 4. Redistribution. You may reproduce and distribute copies of the
90
+ Work or Derivative Works thereof in any medium, with or without
91
+ modifications, and in Source or Object form, provided that You
92
+ meet the following conditions:
93
+
94
+ (a) You must give any other recipients of the Work or
95
+ Derivative Works a copy of this License; and
96
+
97
+ (b) You must cause any modified files to carry prominent notices
98
+ stating that You changed the files; and
99
+
100
+ (c) You must retain, in the Source form of any Derivative Works
101
+ that You distribute, all copyright, patent, trademark, and
102
+ attribution notices from the Source form of the Work,
103
+ excluding those notices that do not pertain to any part of
104
+ the Derivative Works; and
105
+
106
+ (d) If the Work includes a "NOTICE" text file as part of its
107
+ distribution, then any Derivative Works that You distribute must
108
+ include a readable copy of the attribution notices contained
109
+ within such NOTICE file, excluding those notices that do not
110
+ pertain to any part of the Derivative Works, in at least one
111
+ of the following places: within a NOTICE text file distributed
112
+ as part of the Derivative Works; within the Source form or
113
+ documentation, if provided along with the Derivative Works; or,
114
+ within a display generated by the Derivative Works, if and
115
+ wherever such third-party notices normally appear. The contents
116
+ of the NOTICE file are for informational purposes only and
117
+ do not modify the License. You may add Your own attribution
118
+ notices within Derivative Works that You distribute, alongside
119
+ or as an addendum to the NOTICE text from the Work, provided
120
+ that such additional attribution notices cannot be construed
121
+ as modifying the License.
122
+
123
+ You may add Your own copyright statement to Your modifications and
124
+ may provide additional or different license terms and conditions
125
+ for use, reproduction, or distribution of Your modifications, or
126
+ for any such Derivative Works as a whole, provided Your use,
127
+ reproduction, and distribution of the Work otherwise complies with
128
+ the conditions stated in this License.
129
+
130
+ 5. Submission of Contributions. Unless You explicitly state otherwise,
131
+ any Contribution intentionally submitted for inclusion in the Work
132
+ by You to the Licensor shall be under the terms and conditions of
133
+ this License, without any additional terms or conditions.
134
+ Notwithstanding the above, nothing herein shall supersede or modify
135
+ the terms of any separate license agreement you may have executed
136
+ with Licensor regarding such Contributions.
137
+
138
+ 6. Trademarks. This License does not grant permission to use the trade
139
+ names, trademarks, service marks, or product names of the Licensor,
140
+ except as required for reasonable and customary use in describing the
141
+ origin of the Work and reproducing the content of the NOTICE file.
142
+
143
+ 7. Disclaimer of Warranty. Unless required by applicable law or
144
+ agreed to in writing, Licensor provides the Work (and each
145
+ Contributor provides its Contributions) on an "AS IS" BASIS,
146
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
147
+ implied, including, without limitation, any warranties or conditions
148
+ of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
149
+ PARTICULAR PURPOSE. You are solely responsible for determining the
150
+ appropriateness of using or redistributing the Work and assume any
151
+ risks associated with Your exercise of permissions under this License.
152
+
153
+ 8. Limitation of Liability. In no event and under no legal theory,
154
+ whether in tort (including negligence), contract, or otherwise,
155
+ unless required by applicable law (such as deliberate and grossly
156
+ negligent acts) or agreed to in writing, shall any Contributor be
157
+ liable to You for damages, including any direct, indirect, special,
158
+ incidental, or consequential damages of any character arising as a
159
+ result of this License or out of the use or inability to use the
160
+ Work (including but not limited to damages for loss of goodwill,
161
+ work stoppage, computer failure or malfunction, or any and all
162
+ other commercial damages or losses), even if such Contributor
163
+ has been advised of the possibility of such damages.
164
+
165
+ 9. Accepting Warranty or Additional Liability. While redistributing
166
+ the Work or Derivative Works thereof, You may choose to offer,
167
+ and charge a fee for, acceptance of support, warranty, indemnity,
168
+ or other liability obligations and/or rights consistent with this
169
+ License. However, in accepting such obligations, You may act only
170
+ on Your own behalf and on Your sole responsibility, not on behalf
171
+ of any other Contributor, and only if You agree to indemnify,
172
+ defend, and hold each Contributor harmless for any liability
173
+ incurred by, or claims asserted against, such Contributor by reason
174
+ of your accepting any such warranty or additional liability.
175
+
176
+ END OF TERMS AND CONDITIONS
177
+
178
+ APPENDIX: How to apply the Apache License to your work.
179
+
180
+ To apply the Apache License to your work, attach the following
181
+ boilerplate notice, with the fields enclosed by brackets "[]"
182
+ replaced with your own identifying information. (Don't include
183
+ the brackets!) The text should be enclosed in the appropriate
184
+ comment syntax for the file format. We also recommend that a
185
+ file or class name and description of purpose be included on the
186
+ same "printed page" as the copyright notice for easier
187
+ identification within third-party archives.
188
+
189
+ Copyright [yyyy] [name of copyright owner]
190
+
191
+ Licensed under the Apache License, Version 2.0 (the "License");
192
+ you may not use this file except in compliance with the License.
193
+ You may obtain a copy of the License at
194
+
195
+ http://www.apache.org/licenses/LICENSE-2.0
196
+
197
+ Unless required by applicable law or agreed to in writing, software
198
+ distributed under the License is distributed on an "AS IS" BASIS,
199
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
200
+ See the License for the specific language governing permissions and
201
+ limitations under the License.
README.md CHANGED
@@ -1,14 +1,137 @@
1
- ---
2
- title: DORNet
3
- emoji: 🚀
4
- colorFrom: indigo
5
- colorTo: gray
6
- sdk: gradio
7
- sdk_version: 5.36.2
8
- app_file: app.py
9
- pinned: false
10
- license: apache-2.0
11
- short_description: Demo for DORNet
12
- ---
13
-
14
- Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <p align="center">
2
+ <h3 align="center"> DORNet: A Degradation Oriented and Regularized Network for <br> Blind Depth Super-Resolution
3
+ <br>
4
+ :star2: CVPR 2025 (Oral Presentation) :star2:
5
+ </h3>
6
+
7
+ <p align="center"><a href="https://scholar.google.com/citations?user=VogTuQkAAAAJ&hl=zh-CN">Zhengxue Wang</a><sup>1*</sup>,
8
+ <a href="https://yanzq95.github.io/">Zhiqiang Yan✉</a><sup>1*</sup>,
9
+ <a href="https://jspan.github.io/">Jinshan Pan</a><sup>1</sup>,
10
+ <a href="https://guangweigao.github.io/">Guangwei Gao</a><sup>2</sup>,
11
+ <a href="https://cszn.github.io/">Kai Zhang</a><sup>3</sup>,
12
+ <a href="https://scholar.google.com/citations?user=6CIDtZQAAAAJ&hl=zh-CN">Jian Yang✉</a><sup>1</sup> <!--&Dagger;-->
13
+ </p>
14
+
15
+ <p align="center">
16
+ <sup>*</sup>Equal contribution&nbsp;&nbsp;&nbsp;
17
+ <sup>✉</sup>Corresponding author&nbsp;&nbsp;&nbsp;<br>
18
+ <sup>1</sup>Nanjing University of Science and Technology&nbsp;&nbsp;&nbsp;
19
+ <br>
20
+ <sup>2</sup>Nanjing University of Posts and Telecommunications&nbsp;&nbsp;&nbsp;
21
+ <sup>3</sup>Nanjing University&nbsp;&nbsp;&nbsp;
22
+ </p>
23
+
24
+ <p align="center">
25
+ <img src="Figs/Pipeline.png", width="800"/>
26
+ </p>
27
+
28
+
29
+ Overview of DORNet. Given $\boldsymbol D_{up}$ as input, the degradation learning first encodes it to produce degradation representations $\boldsymbol {\tilde{D}}$ and $\boldsymbol D $. Then, $\boldsymbol {\tilde{D}}$, $\boldsymbol D $, $\boldsymbol D_{lr} $, and $\boldsymbol I_{r}$ are fed into multiple degradation-oriented feature transformation (DOFT) modules, generating the HR depth $\boldsymbol D_{hr}$. Finally, $\boldsymbol D$ and $\boldsymbol D_{hr}$ are sent to the degradation regularization to obtain $\boldsymbol D_{d}$, which is used as input for the degradation loss $\mathcal L_{deg}$ and the contrastive loss $\mathcal L_{cont}$. The degradation regularization only applies during training and adds no extra overhead in testing.
30
+
31
+ ## Dependencies
32
+
33
+ ```bash
34
+ Python==3.11.5
35
+ PyTorch==2.1.0
36
+ numpy==1.23.5
37
+ torchvision==0.16.0
38
+ scipy==1.11.3
39
+ Pillow==10.0.1
40
+ tqdm==4.65.0
41
+ scikit-image==0.21.0
42
+ mmcv-full==1.7.2
43
+ ```
44
+
45
+ ## Datasets
46
+
47
+ [RGB-D-D](https://github.com/lingzhi96/RGB-D-D-Dataset)
48
+
49
+ [TOFDSR](https://yanzq95.github.io/projectpage/TOFDC/index.html)
50
+
51
+ [NYU-v2](https://cs.nyu.edu/~fergus/datasets/nyu_depth_v2.html)
52
+
53
+ ## Models
54
+
55
+ Pretrained models can be found in <a href="https://github.com/yanzq95/DORNet/tree/main/checkpoints">checkpoints</a>.
56
+
57
+
58
+ ## Training
59
+
60
+ For the RGB-D-D and NYU-v2 datasets, we use a single GPU to train our DORNet. For the larger TOFDC dataset, we employ multiple GPUs to accelerate training.
61
+
62
+ ### DORNet
63
+ ```
64
+ Train on real-world RGB-D-D
65
+ > python train_nyu_rgbdd.py
66
+ Train on real-world TOFDSR
67
+ > python -m torch.distributed.launch --nproc_per_node 2 train_tofdsr.py --result_root 'experiment/TOFDSR'
68
+ Train on synthetic NYU-v2
69
+ > python train_nyu_rgbdd.py
70
+ ```
71
+
72
+ ### DORNet-T
73
+ ```
74
+ Train on real-world RGB-D-D
75
+ > python train_nyu_rgbdd.py --tiny_model
76
+ Train on real-world TOFDSR
77
+ > python -m torch.distributed.launch --nproc_per_node 2 train_tofdsr.py --result_root 'experiment/TOFDSR_T' --tiny_model
78
+ Train on synthetic NYU-v2
79
+ > python train_nyu_rgbdd.py --tiny_model
80
+ ```
81
+
82
+ ## Testing
83
+
84
+ ### DORNet
85
+ ```
86
+ Test on real-world RGB-D-D
87
+ > python test_nyu_rgbdd.py
88
+ Test on real-world TOFDSR
89
+ > python test_tofdsr.py
90
+ Test on synthetic NYU-v2
91
+ > python test_nyu_rgbdd.py
92
+ ```
93
+
94
+ ### DORNet-T
95
+ ```
96
+ Test on real-world RGB-D-D
97
+ > python test_nyu_rgbdd.py --tiny_model
98
+ Test on real-world TOFDSR
99
+ > python test_tofdsr.py --tiny_model
100
+ Test on synthetic NYU-v2
101
+ > python test_nyu_rgbdd.py --tiny_model
102
+ ```
103
+
104
+ ## Experiments
105
+
106
+ ### Quantitative comparison
107
+
108
+ <p align="center">
109
+ <img src="Figs/Params_Time.png", width="500"/>
110
+ <br>
111
+ Complexity on RGB-D-D (w/o Noisy) tested by a 4090 GPU. A larger circle diameter indicates a higher inference time.
112
+ </p>
113
+
114
+
115
+
116
+ ### Visual comparison
117
+
118
+ <p align="center">
119
+ <img src="Figs/RGBDD.png", width="1000"/>
120
+ <br>
121
+ Visual results on the real-world RGB-D-D dataset (w/o Noise).
122
+ </p>
123
+
124
+
125
+ ## Citation
126
+
127
+ If our method proves to be of any assistance, please consider citing:
128
+ ```
129
+ @inproceedings{wang2025dornet,
130
+ title={DORNet: A Degradation Oriented and Regularized Network for Blind Depth Super-Resolution},
131
+ author={Wang, Zhengxue and Yan, Zhiqiang and Pan, Jinshan and Gao, Guangwei and Zhang, Kai and Yang, Jian},
132
+ booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
133
+ pages={15813--15822},
134
+ year={2025}
135
+ }
136
+ ```
137
+
app.py ADDED
@@ -0,0 +1,156 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import gradio as gr
2
+ import numpy as np
3
+ import torch
4
+ import os
5
+ import cv2
6
+ from PIL import Image
7
+ import torchvision.transforms as transforms
8
+ from net.dornet import Net
9
+ from net.dornet_ddp import Net_ddp
10
+
11
+ # init
12
+ device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
13
+ net = Net(tiny_model=False).to(device)
14
+ model_ckpt_map = {
15
+ "RGB-D-D": "./checkpoints/RGBDD.pth",
16
+ "TOFDSR": "./checkpoints/TOFDSR.pth"
17
+ }
18
+
19
+ # load model
20
+ def load_model(model_type: str):
21
+ global net
22
+ ckpt_path = model_ckpt_map[model_type]
23
+ print(f"Loading weights from: {ckpt_path}")
24
+ if model_type == "RGB-D-D":
25
+ net = Net(tiny_model=False).to(device)
26
+ elif model_type == "TOFDSR":
27
+ net = Net_ddp(tiny_model=False).srn.to(device)
28
+ else:
29
+ raise ValueError(f"Unknown model_type: {model_type}")
30
+
31
+ net.load_state_dict(torch.load(ckpt_path, map_location=device))
32
+ net.eval()
33
+
34
+ load_model("RGB-D-D")
35
+
36
+
37
+ # data process
38
+ def preprocess_inputs(rgb_image: Image.Image, lr_depth: Image.Image):
39
+ image = np.array(rgb_image.convert("RGB")).astype(np.float32)
40
+ h, w, _ = image.shape
41
+ lr = np.array(lr_depth.resize((w, h), Image.BICUBIC)).astype(np.float32)
42
+ # Normalize depth
43
+ max_out, min_out = 5000.0, 0.0
44
+ lr = (lr - min_out) / (max_out - min_out)
45
+ # Normalize RGB
46
+ maxx, minn = np.max(image), np.min(image)
47
+ image = (image - minn) / (maxx - minn)
48
+ # To tensor
49
+ data_transform = transforms.Compose([transforms.ToTensor()])
50
+ image = data_transform(image).float()
51
+ lr = data_transform(np.expand_dims(lr, 2)).float()
52
+ # Add batch dimension
53
+ lr = lr.unsqueeze(0).to(device)
54
+ image = image.unsqueeze(0).to(device)
55
+ return image, lr, min_out, max_out
56
+
57
+
58
+ # model inference
59
+ @torch.no_grad()
60
+ def infer(rgb_image: Image.Image, lr_depth: Image.Image, model_type: str):
61
+ load_model(model_type) # reset weight
62
+
63
+ image, lr, min_out, max_out = preprocess_inputs(rgb_image, lr_depth)
64
+
65
+ if model_type == "RGB-D-D":
66
+ out = net(x_query=lr, rgb=image)
67
+ elif model_type == "TOFDSR":
68
+ out, _ = net(x_query=lr, rgb=image)
69
+
70
+ pred = out[0, 0] * (max_out - min_out) + min_out
71
+ pred = pred.cpu().numpy().astype(np.uint16)
72
+ # raw
73
+ pred_gray = Image.fromarray(pred)
74
+
75
+ # heat
76
+ pred_norm = (pred - np.min(pred)) / (np.max(pred) - np.min(pred)) * 255
77
+ pred_vis = pred_norm.astype(np.uint8)
78
+ pred_heat = cv2.applyColorMap(pred_vis, cv2.COLORMAP_PLASMA)
79
+ pred_heat = cv2.cvtColor(pred_heat, cv2.COLOR_BGR2RGB)
80
+ return pred_gray, Image.fromarray(pred_heat)
81
+
82
+
83
+ # Gradio
84
+ # demo = gr.Interface(
85
+ # fn=infer,
86
+ # inputs=[
87
+ # gr.Image(label="RGB Image", type="pil"),
88
+ # gr.Image(label="Low-res Depth", type="pil", image_mode="I"),
89
+ # gr.Dropdown(choices=["RGB-D-D", "TOFDSR"], label="Model Type", value="RGB-D-D")
90
+ # ],
91
+ # outputs=[
92
+ # gr.Image(label="DORNet Output", type="pil", elem_classes=["output-image"]),
93
+ # gr.Image(label="Normalized Output", type="pil", elem_classes=["output-image"])
94
+ # ],
95
+ # examples=[
96
+ # ["examples/RGB-D-D/20200518160957_RGB.jpg", "examples/RGB-D-D/20200518160957_LR_fill_depth.png", "RGB-D-D"],
97
+ # ["examples/TOFDSR/2020_09_08_13_59_59_435_rgb_rgb_crop.png", "examples/TOFDSR/2020_09_08_13_59_59_435_rgb_depth_crop_fill.png", "TOFDSR"],
98
+ # ],
99
+ # allow_flagging="never",
100
+ # title="DORNet: A Degradation Oriented and Regularized Network for Blind Depth Super-Resolution \n CVPR 2025 (Oral Presentation)",
101
+ # css="""
102
+ # .output-image {
103
+ # display: flex;
104
+ # justify-content: center;
105
+ # align-items: center;
106
+ # }
107
+ # .output-image img {
108
+ # margin: auto;
109
+ # display: block;
110
+ # }
111
+ # """
112
+ # )
113
+ #
114
+ # demo.launch(share=True)
115
+ Intro = """
116
+ ## DORNet: A Degradation Oriented and Regularized Network for Blind Depth Super-Resolution
117
+ [📄 Paper](https://arxiv.org/pdf/2410.11666) • [💻 Code](https://github.com/yanzq95/DORNet) • [📦 Model](https://huggingface.co/wzxwyx/DORNet/tree/main)
118
+ """
119
+
120
+ with gr.Blocks(css="""
121
+ .output-image {
122
+ display: flex;
123
+ justify-content: center;
124
+ align-items: center;
125
+ }
126
+ .output-image img {
127
+ margin: auto;
128
+ display: block;
129
+ }
130
+ """) as demo:
131
+ gr.Markdown(Intro)
132
+
133
+ with gr.Row():
134
+ with gr.Column():
135
+ rgb_input = gr.Image(label="RGB Image", type="pil")
136
+ lr_input = gr.Image(label="Low-res Depth", type="pil", image_mode="I")
137
+ with gr.Column():
138
+ output1 = gr.Image(label="DORNet Output", type="pil", elem_classes=["output-image"])
139
+ output2 = gr.Image(label="Normalized Output", type="pil", elem_classes=["output-image"])
140
+
141
+ model_selector = gr.Dropdown(choices=["RGB-D-D", "TOFDSR"], label="Model Type", value="RGB-D-D")
142
+ run_button = gr.Button("Run Inference")
143
+
144
+ gr.Examples(
145
+ examples=[
146
+ ["examples/RGB-D-D/20200518160957_RGB.jpg", "examples/RGB-D-D/20200518160957_LR_fill_depth.png", "RGB-D-D"],
147
+ ["examples/TOFDSR/2020_09_08_13_59_59_435_rgb_rgb_crop.png", "examples/TOFDSR/2020_09_08_13_59_59_435_rgb_depth_crop_fill.png", "TOFDSR"],
148
+ ],
149
+ inputs=[rgb_input, lr_input, model_selector],
150
+ outputs=[output1, output2],
151
+ label="Try Examples ↓"
152
+ )
153
+
154
+ run_button.click(fn=infer, inputs=[rgb_input, lr_input, model_selector], outputs=[output1, output2])
155
+
156
+ demo.launch(share=True)
checkpoints/NYU_T_X16.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3613d2cbd652990525d915972e1c24a885832d3196a929e13aa40833a13f1df8
3
+ size 1957834
checkpoints/NYU_T_X4.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f91a4b63870363483c189201196d1625b313fd45b8ed4935d3a55696fdbe7234
3
+ size 1957834
checkpoints/NYU_T_X8.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6cbd299ec3a589c23ded86a9c956cea7d70e7e3593ca3593337836b1e84b36f8
3
+ size 1958239
checkpoints/NYU_X16.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:00b68485b244b7579e0b471b8a2d7946493c320eedf000e1cbd334ca8ad55d79
3
+ size 12359114
checkpoints/NYU_X4.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9be0f4a053b724dbbbd140db4a32f32973cb01e97edd47c1100053797bebe262
3
+ size 12359114
checkpoints/NYU_X8.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:26b1d4f972aefc56e74d427345a7ed9a6b61af8319ac4a635af3d7d687a2c1e6
3
+ size 12359519
checkpoints/RGBDD.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c4c8aed8c42a292dc5a852f760cab4d4ecb7d5aceb548671a2a8bc31d7b57aa5
3
+ size 12359519
checkpoints/RGBDD_Noisy.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d1e6994abbf02b8a03de789511580bd0a5e7219f4d9ffa9e84a5a3f6ffefd79f
3
+ size 12359519
checkpoints/RGBDD_Noisy_T.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:626c55ff559f9d6ce6e3f4229c421de467b5d03c9c27f8bb871c40597a21c878
3
+ size 1958239
checkpoints/RGBDD_T.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:38bd20f150d138e395a66cf0a981f6e0e840a995dc4866a6c8b1cadaf5494180
3
+ size 1958239
checkpoints/TOFDSR.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:687f072ac89e88c47c1248db0b3e52c487fe0d198be546b5348038b6b3369e49
3
+ size 11806363
checkpoints/TOFDSR_Noisy.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cabe16fe9593dede63883af64678496ab92a83d3271ef0e23e8471574ce4e1cf
3
+ size 11806363
checkpoints/TOFDSR_Noisy_T.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7ac5209fa1e013ab14cc4ad973cc01264549612b5f3361f31274efb9aade88be
3
+ size 1849078
checkpoints/TOFDSR_T.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:68e1955b37405cb209ce1b7a8decca81169f7280e58c151c44525c6da902cdbd
3
+ size 1850267
data/TOFDSR_Test.txt ADDED
@@ -0,0 +1,560 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ TOFDC_split/Test/RGB/20200716_104505_rgb_crop.png,TOFDC_split/Test/GT/20200716_104505_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200716_104505_depth_crop_fill.png
2
+ TOFDC_split/Test/RGB/20200717_100634_rgb_crop.png,TOFDC_split/Test/GT/20200717_100634_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200717_100634_depth_crop_fill.png
3
+ TOFDC_split/Test/RGB/20200818_120711_rgb_crop.png,TOFDC_split/Test/GT/20200818_120711_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_120711_depth_crop_fill.png
4
+ TOFDC_split/Test/RGB/20200603_123554_rgb_crop.png,TOFDC_split/Test/GT/20200603_123554_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200603_123554_depth_crop_fill.png
5
+ TOFDC_split/Test/RGB/2020_09_21_21_29_46_695_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_21_21_29_46_695_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_21_21_29_46_695_rgb_depth_crop_fill.png
6
+ TOFDC_split/Test/RGB/20200818_105105_rgb_crop.png,TOFDC_split/Test/GT/20200818_105105_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_105105_depth_crop_fill.png
7
+ TOFDC_split/Test/RGB/20200818_152310_rgb_crop.png,TOFDC_split/Test/GT/20200818_152310_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_152310_depth_crop_fill.png
8
+ TOFDC_split/Test/RGB/20200722_103518_rgb_crop.png,TOFDC_split/Test/GT/20200722_103518_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200722_103518_depth_crop_fill.png
9
+ TOFDC_split/Test/RGB/20200721_100558_rgb_crop.png,TOFDC_split/Test/GT/20200721_100558_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200721_100558_depth_crop_fill.png
10
+ TOFDC_split/Test/RGB/20200722_102852_rgb_crop.png,TOFDC_split/Test/GT/20200722_102852_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200722_102852_depth_crop_fill.png
11
+ TOFDC_split/Test/RGB/20200920_112932_rgb_crop.png,TOFDC_split/Test/GT/20200920_112932_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_112932_depth_crop_fill.png
12
+ TOFDC_split/Test/RGB/20200831_111743_rgb_crop.png,TOFDC_split/Test/GT/20200831_111743_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200831_111743_depth_crop_fill.png
13
+ TOFDC_split/Test/RGB/20200722_153225_rgb_crop.png,TOFDC_split/Test/GT/20200722_153225_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200722_153225_depth_crop_fill.png
14
+ TOFDC_split/Test/RGB/20200921_140327_rgb_crop.png,TOFDC_split/Test/GT/20200921_140327_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200921_140327_depth_crop_fill.png
15
+ TOFDC_split/Test/RGB/20200726_102650_rgb_crop.png,TOFDC_split/Test/GT/20200726_102650_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200726_102650_depth_crop_fill.png
16
+ TOFDC_split/Test/RGB/20200604_152722_rgb_crop.png,TOFDC_split/Test/GT/20200604_152722_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200604_152722_depth_crop_fill.png
17
+ TOFDC_split/Test/RGB/20200927_202052_rgb_crop.png,TOFDC_split/Test/GT/20200927_202052_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_202052_depth_crop_fill.png
18
+ TOFDC_split/Test/RGB/20200818_114303_rgb_crop.png,TOFDC_split/Test/GT/20200818_114303_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_114303_depth_crop_fill.png
19
+ TOFDC_split/Test/RGB/20200820_170902_rgb_crop.png,TOFDC_split/Test/GT/20200820_170902_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200820_170902_depth_crop_fill.png
20
+ TOFDC_split/Test/RGB/20200927_194454_rgb_crop.png,TOFDC_split/Test/GT/20200927_194454_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_194454_depth_crop_fill.png
21
+ TOFDC_split/Test/RGB/2020_09_21_21_50_54_065_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_21_21_50_54_065_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_21_21_50_54_065_rgb_depth_crop_fill.png
22
+ TOFDC_split/Test/RGB/20200927_191907_rgb_crop.png,TOFDC_split/Test/GT/20200927_191907_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_191907_depth_crop_fill.png
23
+ TOFDC_split/Test/RGB/20200919_190817_rgb_crop.png,TOFDC_split/Test/GT/20200919_190817_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_190817_depth_crop_fill.png
24
+ TOFDC_split/Test/RGB/20200719_194950_rgb_crop.png,TOFDC_split/Test/GT/20200719_194950_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200719_194950_depth_crop_fill.png
25
+ TOFDC_split/Test/RGB/2020_09_11_22_16_01_836_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_11_22_16_01_836_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_11_22_16_01_836_rgb_depth_crop_fill.png
26
+ TOFDC_split/Test/RGB/20200818_155908_rgb_crop.png,TOFDC_split/Test/GT/20200818_155908_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_155908_depth_crop_fill.png
27
+ TOFDC_split/Test/RGB/20200721_201000_rgb_crop.png,TOFDC_split/Test/GT/20200721_201000_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200721_201000_depth_crop_fill.png
28
+ TOFDC_split/Test/RGB/2020_09_09_11_19_02_530_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_11_19_02_530_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_11_19_02_530_rgb_depth_crop_fill.png
29
+ TOFDC_split/Test/RGB/2020_09_08_18_00_27_005_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_18_00_27_005_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_18_00_27_005_rgb_depth_crop_fill.png
30
+ TOFDC_split/Test/RGB/2020_09_22_15_03_30_825_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_22_15_03_30_825_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_22_15_03_30_825_rgb_depth_crop_fill.png
31
+ TOFDC_split/Test/RGB/2020_09_10_11_09_19_261_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_10_11_09_19_261_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_10_11_09_19_261_rgb_depth_crop_fill.png
32
+ TOFDC_split/Test/RGB/20200719_175258_rgb_crop.png,TOFDC_split/Test/GT/20200719_175258_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200719_175258_depth_crop_fill.png
33
+ TOFDC_split/Test/RGB/2020_09_11_22_28_43_143_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_11_22_28_43_143_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_11_22_28_43_143_rgb_depth_crop_fill.png
34
+ TOFDC_split/Test/RGB/20200819_183942_rgb_crop.png,TOFDC_split/Test/GT/20200819_183942_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_183942_depth_crop_fill.png
35
+ TOFDC_split/Test/RGB/2020_09_11_10_41_16_143_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_11_10_41_16_143_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_11_10_41_16_143_rgb_depth_crop_fill.png
36
+ TOFDC_split/Test/RGB/20200725_154850_rgb_crop.png,TOFDC_split/Test/GT/20200725_154850_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200725_154850_depth_crop_fill.png
37
+ TOFDC_split/Test/RGB/20200818_155243_rgb_crop.png,TOFDC_split/Test/GT/20200818_155243_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_155243_depth_crop_fill.png
38
+ TOFDC_split/Test/RGB/20200824_143942_rgb_crop.png,TOFDC_split/Test/GT/20200824_143942_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_143942_depth_crop_fill.png
39
+ TOFDC_split/Test/RGB/2020_09_14_19_53_27_896_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_14_19_53_27_896_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_14_19_53_27_896_rgb_depth_crop_fill.png
40
+ TOFDC_split/Test/RGB/20200719_154136_rgb_crop.png,TOFDC_split/Test/GT/20200719_154136_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200719_154136_depth_crop_fill.png
41
+ TOFDC_split/Test/RGB/20200919_111906_rgb_crop.png,TOFDC_split/Test/GT/20200919_111906_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_111906_depth_crop_fill.png
42
+ TOFDC_split/Test/RGB/20200606_145744_rgb_crop.png,TOFDC_split/Test/GT/20200606_145744_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200606_145744_depth_crop_fill.png
43
+ TOFDC_split/Test/RGB/20200928_152622_rgb_crop.png,TOFDC_split/Test/GT/20200928_152622_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200928_152622_depth_crop_fill.png
44
+ TOFDC_split/Test/RGB/20200928_152109_rgb_crop.png,TOFDC_split/Test/GT/20200928_152109_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200928_152109_depth_crop_fill.png
45
+ TOFDC_split/Test/RGB/20200718_103458_rgb_crop.png,TOFDC_split/Test/GT/20200718_103458_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200718_103458_depth_crop_fill.png
46
+ TOFDC_split/Test/RGB/20200819_141832_rgb_crop.png,TOFDC_split/Test/GT/20200819_141832_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_141832_depth_crop_fill.png
47
+ TOFDC_split/Test/RGB/20200719_152930_rgb_crop.png,TOFDC_split/Test/GT/20200719_152930_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200719_152930_depth_crop_fill.png
48
+ TOFDC_split/Test/RGB/20200818_114851_rgb_crop.png,TOFDC_split/Test/GT/20200818_114851_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_114851_depth_crop_fill.png
49
+ TOFDC_split/Test/RGB/20200719_105752_rgb_crop.png,TOFDC_split/Test/GT/20200719_105752_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200719_105752_depth_crop_fill.png
50
+ TOFDC_split/Test/RGB/20200818_123419_rgb_crop.png,TOFDC_split/Test/GT/20200818_123419_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_123419_depth_crop_fill.png
51
+ TOFDC_split/Test/RGB/2020_09_11_11_11_15_515_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_11_11_11_15_515_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_11_11_11_15_515_rgb_depth_crop_fill.png
52
+ TOFDC_split/Test/RGB/20200725_161537_rgb_crop.png,TOFDC_split/Test/GT/20200725_161537_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200725_161537_depth_crop_fill.png
53
+ TOFDC_split/Test/RGB/20200820_164628_rgb_crop.png,TOFDC_split/Test/GT/20200820_164628_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200820_164628_depth_crop_fill.png
54
+ TOFDC_split/Test/RGB/20200927_152821_rgb_crop.png,TOFDC_split/Test/GT/20200927_152821_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_152821_depth_crop_fill.png
55
+ TOFDC_split/Test/RGB/20200824_121012_rgb_crop.png,TOFDC_split/Test/GT/20200824_121012_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_121012_depth_crop_fill.png
56
+ TOFDC_split/Test/RGB/2020_09_21_20_42_08_839_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_21_20_42_08_839_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_21_20_42_08_839_rgb_depth_crop_fill.png
57
+ TOFDC_split/Test/RGB/2020_09_09_11_26_16_634_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_11_26_16_634_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_11_26_16_634_rgb_depth_crop_fill.png
58
+ TOFDC_split/Test/RGB/2020_09_09_11_57_42_136_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_11_57_42_136_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_11_57_42_136_rgb_depth_crop_fill.png
59
+ TOFDC_split/Test/RGB/2020_09_26_11_40_36_466_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_26_11_40_36_466_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_26_11_40_36_466_rgb_depth_crop_fill.png
60
+ TOFDC_split/Test/RGB/2020_09_08_15_55_37_230_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_15_55_37_230_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_15_55_37_230_rgb_depth_crop_fill.png
61
+ TOFDC_split/Test/RGB/2020_09_12_20_59_31_547_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_20_59_31_547_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_20_59_31_547_rgb_depth_crop_fill.png
62
+ TOFDC_split/Test/RGB/20200603_134220_rgb_crop.png,TOFDC_split/Test/GT/20200603_134220_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200603_134220_depth_crop_fill.png
63
+ TOFDC_split/Test/RGB/2020_09_21_20_13_06_978_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_21_20_13_06_978_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_21_20_13_06_978_rgb_depth_crop_fill.png
64
+ TOFDC_split/Test/RGB/2020_09_12_16_08_56_136_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_16_08_56_136_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_16_08_56_136_rgb_depth_crop_fill.png
65
+ TOFDC_split/Test/RGB/20200820_202007_rgb_crop.png,TOFDC_split/Test/GT/20200820_202007_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200820_202007_depth_crop_fill.png
66
+ TOFDC_split/Test/RGB/2020_09_08_10_39_34_461_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_10_39_34_461_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_10_39_34_461_rgb_depth_crop_fill.png
67
+ TOFDC_split/Test/RGB/2020_09_08_11_34_21_487_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_11_34_21_487_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_11_34_21_487_rgb_depth_crop_fill.png
68
+ TOFDC_split/Test/RGB/20200819_105135_rgb_crop.png,TOFDC_split/Test/GT/20200819_105135_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_105135_depth_crop_fill.png
69
+ TOFDC_split/Test/RGB/20200601_194309_rgb_crop.png,TOFDC_split/Test/GT/20200601_194309_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200601_194309_depth_crop_fill.png
70
+ TOFDC_split/Test/RGB/20200719_145024_rgb_crop.png,TOFDC_split/Test/GT/20200719_145024_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200719_145024_depth_crop_fill.png
71
+ TOFDC_split/Test/RGB/20200819_122505_rgb_crop.png,TOFDC_split/Test/GT/20200819_122505_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_122505_depth_crop_fill.png
72
+ TOFDC_split/Test/RGB/20200818_160711_rgb_crop.png,TOFDC_split/Test/GT/20200818_160711_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_160711_depth_crop_fill.png
73
+ TOFDC_split/Test/RGB/20200819_170025_rgb_crop.png,TOFDC_split/Test/GT/20200819_170025_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_170025_depth_crop_fill.png
74
+ TOFDC_split/Test/RGB/20200920_113836_rgb_crop.png,TOFDC_split/Test/GT/20200920_113836_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_113836_depth_crop_fill.png
75
+ TOFDC_split/Test/RGB/20200721_162455_rgb_crop.png,TOFDC_split/Test/GT/20200721_162455_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200721_162455_depth_crop_fill.png
76
+ TOFDC_split/Test/RGB/20200717_162540_rgb_crop.png,TOFDC_split/Test/GT/20200717_162540_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200717_162540_depth_crop_fill.png
77
+ TOFDC_split/Test/RGB/20200818_160441_rgb_crop.png,TOFDC_split/Test/GT/20200818_160441_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_160441_depth_crop_fill.png
78
+ TOFDC_split/Test/RGB/20200824_164044_rgb_crop.png,TOFDC_split/Test/GT/20200824_164044_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_164044_depth_crop_fill.png
79
+ TOFDC_split/Test/RGB/20200824_161533_rgb_crop.png,TOFDC_split/Test/GT/20200824_161533_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_161533_depth_crop_fill.png
80
+ TOFDC_split/Test/RGB/20200919_110444_rgb_crop.png,TOFDC_split/Test/GT/20200919_110444_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_110444_depth_crop_fill.png
81
+ TOFDC_split/Test/RGB/20200723_205806_rgb_crop.png,TOFDC_split/Test/GT/20200723_205806_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200723_205806_depth_crop_fill.png
82
+ TOFDC_split/Test/RGB/20200718_150137_rgb_crop.png,TOFDC_split/Test/GT/20200718_150137_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200718_150137_depth_crop_fill.png
83
+ TOFDC_split/Test/RGB/20200927_202911_rgb_crop.png,TOFDC_split/Test/GT/20200927_202911_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_202911_depth_crop_fill.png
84
+ TOFDC_split/Test/RGB/20200723_102945_rgb_crop.png,TOFDC_split/Test/GT/20200723_102945_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200723_102945_depth_crop_fill.png
85
+ TOFDC_split/Test/RGB/20200726_105624_rgb_crop.png,TOFDC_split/Test/GT/20200726_105624_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200726_105624_depth_crop_fill.png
86
+ TOFDC_split/Test/RGB/2020_09_13_19_31_37_869_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_13_19_31_37_869_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_13_19_31_37_869_rgb_depth_crop_fill.png
87
+ TOFDC_split/Test/RGB/20200920_170617_rgb_crop.png,TOFDC_split/Test/GT/20200920_170617_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_170617_depth_crop_fill.png
88
+ TOFDC_split/Test/RGB/20200927_193927_rgb_crop.png,TOFDC_split/Test/GT/20200927_193927_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_193927_depth_crop_fill.png
89
+ TOFDC_split/Test/RGB/2020_09_08_16_07_41_800_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_16_07_41_800_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_16_07_41_800_rgb_depth_crop_fill.png
90
+ TOFDC_split/Test/RGB/20200603_131120_rgb_crop.png,TOFDC_split/Test/GT/20200603_131120_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200603_131120_depth_crop_fill.png
91
+ TOFDC_split/Test/RGB/20200721_111013_rgb_crop.png,TOFDC_split/Test/GT/20200721_111013_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200721_111013_depth_crop_fill.png
92
+ TOFDC_split/Test/RGB/20200601_110530_rgb_crop.png,TOFDC_split/Test/GT/20200601_110530_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200601_110530_depth_crop_fill.png
93
+ TOFDC_split/Test/RGB/2020_09_10_22_49_02_519_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_10_22_49_02_519_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_10_22_49_02_519_rgb_depth_crop_fill.png
94
+ TOFDC_split/Test/RGB/20200927_155929_rgb_crop.png,TOFDC_split/Test/GT/20200927_155929_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_155929_depth_crop_fill.png
95
+ TOFDC_split/Test/RGB/20200820_100501_rgb_crop.png,TOFDC_split/Test/GT/20200820_100501_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200820_100501_depth_crop_fill.png
96
+ TOFDC_split/Test/RGB/2020_09_09_11_45_50_523_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_11_45_50_523_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_11_45_50_523_rgb_depth_crop_fill.png
97
+ TOFDC_split/Test/RGB/2020_09_23_22_33_24_383_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_23_22_33_24_383_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_23_22_33_24_383_rgb_depth_crop_fill.png
98
+ TOFDC_split/Test/RGB/2020_09_13_21_28_52_074_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_13_21_28_52_074_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_13_21_28_52_074_rgb_depth_crop_fill.png
99
+ TOFDC_split/Test/RGB/2020_09_26_10_32_39_351_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_26_10_32_39_351_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_26_10_32_39_351_rgb_depth_crop_fill.png
100
+ TOFDC_split/Test/RGB/20200927_153446_rgb_crop.png,TOFDC_split/Test/GT/20200927_153446_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_153446_depth_crop_fill.png
101
+ TOFDC_split/Test/RGB/20200818_213819_rgb_crop.png,TOFDC_split/Test/GT/20200818_213819_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_213819_depth_crop_fill.png
102
+ TOFDC_split/Test/RGB/2020_09_22_10_05_08_603_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_22_10_05_08_603_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_22_10_05_08_603_rgb_depth_crop_fill.png
103
+ TOFDC_split/Test/RGB/20200919_170833_rgb_crop.png,TOFDC_split/Test/GT/20200919_170833_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_170833_depth_crop_fill.png
104
+ TOFDC_split/Test/RGB/20200727_165805_rgb_crop.png,TOFDC_split/Test/GT/20200727_165805_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200727_165805_depth_crop_fill.png
105
+ TOFDC_split/Test/RGB/20200819_145319_rgb_crop.png,TOFDC_split/Test/GT/20200819_145319_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_145319_depth_crop_fill.png
106
+ TOFDC_split/Test/RGB/20200818_151827_rgb_crop.png,TOFDC_split/Test/GT/20200818_151827_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_151827_depth_crop_fill.png
107
+ TOFDC_split/Test/RGB/20200819_153248_rgb_crop.png,TOFDC_split/Test/GT/20200819_153248_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_153248_depth_crop_fill.png
108
+ TOFDC_split/Test/RGB/2020_09_26_14_21_00_533_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_26_14_21_00_533_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_26_14_21_00_533_rgb_depth_crop_fill.png
109
+ TOFDC_split/Test/RGB/20200927_203648_rgb_crop.png,TOFDC_split/Test/GT/20200927_203648_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_203648_depth_crop_fill.png
110
+ TOFDC_split/Test/RGB/2020_09_13_21_18_21_224_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_13_21_18_21_224_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_13_21_18_21_224_rgb_depth_crop_fill.png
111
+ TOFDC_split/Test/RGB/20200718_153745_rgb_crop.png,TOFDC_split/Test/GT/20200718_153745_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200718_153745_depth_crop_fill.png
112
+ TOFDC_split/Test/RGB/20200818_154707_rgb_crop.png,TOFDC_split/Test/GT/20200818_154707_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_154707_depth_crop_fill.png
113
+ TOFDC_split/Test/RGB/20200927_142826_rgb_crop.png,TOFDC_split/Test/GT/20200927_142826_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_142826_depth_crop_fill.png
114
+ TOFDC_split/Test/RGB/20200820_111453_rgb_crop.png,TOFDC_split/Test/GT/20200820_111453_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200820_111453_depth_crop_fill.png
115
+ TOFDC_split/Test/RGB/20200601_151641_rgb_crop.png,TOFDC_split/Test/GT/20200601_151641_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200601_151641_depth_crop_fill.png
116
+ TOFDC_split/Test/RGB/20200726_103105_rgb_crop.png,TOFDC_split/Test/GT/20200726_103105_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200726_103105_depth_crop_fill.png
117
+ TOFDC_split/Test/RGB/20200820_200355_rgb_crop.png,TOFDC_split/Test/GT/20200820_200355_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200820_200355_depth_crop_fill.png
118
+ TOFDC_split/Test/RGB/20200717_111412_rgb_crop.png,TOFDC_split/Test/GT/20200717_111412_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200717_111412_depth_crop_fill.png
119
+ TOFDC_split/Test/RGB/20200726_110251_rgb_crop.png,TOFDC_split/Test/GT/20200726_110251_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200726_110251_depth_crop_fill.png
120
+ TOFDC_split/Test/RGB/2020_09_10_16_26_44_874_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_10_16_26_44_874_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_10_16_26_44_874_rgb_depth_crop_fill.png
121
+ TOFDC_split/Test/RGB/20200722_201920_rgb_crop.png,TOFDC_split/Test/GT/20200722_201920_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200722_201920_depth_crop_fill.png
122
+ TOFDC_split/Test/RGB/2020_09_09_17_02_43_395_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_17_02_43_395_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_17_02_43_395_rgb_depth_crop_fill.png
123
+ TOFDC_split/Test/RGB/2020_09_13_19_47_04_644_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_13_19_47_04_644_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_13_19_47_04_644_rgb_depth_crop_fill.png
124
+ TOFDC_split/Test/RGB/2020_09_08_11_22_28_946_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_11_22_28_946_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_11_22_28_946_rgb_depth_crop_fill.png
125
+ TOFDC_split/Test/RGB/2020_09_21_19_57_19_431_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_21_19_57_19_431_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_21_19_57_19_431_rgb_depth_crop_fill.png
126
+ TOFDC_split/Test/RGB/20200921_140059_rgb_crop.png,TOFDC_split/Test/GT/20200921_140059_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200921_140059_depth_crop_fill.png
127
+ TOFDC_split/Test/RGB/20200725_185201_rgb_crop.png,TOFDC_split/Test/GT/20200725_185201_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200725_185201_depth_crop_fill.png
128
+ TOFDC_split/Test/RGB/20200818_215051_rgb_crop.png,TOFDC_split/Test/GT/20200818_215051_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_215051_depth_crop_fill.png
129
+ TOFDC_split/Test/RGB/20200831_104743_rgb_crop.png,TOFDC_split/Test/GT/20200831_104743_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200831_104743_depth_crop_fill.png
130
+ TOFDC_split/Test/RGB/20200607_170307_rgb_crop.png,TOFDC_split/Test/GT/20200607_170307_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200607_170307_depth_crop_fill.png
131
+ TOFDC_split/Test/RGB/20200604_150753_rgb_crop.png,TOFDC_split/Test/GT/20200604_150753_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200604_150753_depth_crop_fill.png
132
+ TOFDC_split/Test/RGB/20200720_151834_rgb_crop.png,TOFDC_split/Test/GT/20200720_151834_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200720_151834_depth_crop_fill.png
133
+ TOFDC_split/Test/RGB/20200818_121442_rgb_crop.png,TOFDC_split/Test/GT/20200818_121442_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_121442_depth_crop_fill.png
134
+ TOFDC_split/Test/RGB/20200725_160055_rgb_crop.png,TOFDC_split/Test/GT/20200725_160055_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200725_160055_depth_crop_fill.png
135
+ TOFDC_split/Test/RGB/20200601_095208_rgb_crop.png,TOFDC_split/Test/GT/20200601_095208_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200601_095208_depth_crop_fill.png
136
+ TOFDC_split/Test/RGB/20200920_152246_rgb_crop.png,TOFDC_split/Test/GT/20200920_152246_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_152246_depth_crop_fill.png
137
+ TOFDC_split/Test/RGB/20200927_134410_rgb_crop.png,TOFDC_split/Test/GT/20200927_134410_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_134410_depth_crop_fill.png
138
+ TOFDC_split/Test/RGB/20200819_122215_rgb_crop.png,TOFDC_split/Test/GT/20200819_122215_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_122215_depth_crop_fill.png
139
+ TOFDC_split/Test/RGB/20200824_151316_rgb_crop.png,TOFDC_split/Test/GT/20200824_151316_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_151316_depth_crop_fill.png
140
+ TOFDC_split/Test/RGB/20200921_142741_rgb_crop.png,TOFDC_split/Test/GT/20200921_142741_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200921_142741_depth_crop_fill.png
141
+ TOFDC_split/Test/RGB/20200920_190514_rgb_crop.png,TOFDC_split/Test/GT/20200920_190514_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_190514_depth_crop_fill.png
142
+ TOFDC_split/Test/RGB/20200818_150602_rgb_crop.png,TOFDC_split/Test/GT/20200818_150602_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_150602_depth_crop_fill.png
143
+ TOFDC_split/Test/RGB/20200927_155110_rgb_crop.png,TOFDC_split/Test/GT/20200927_155110_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_155110_depth_crop_fill.png
144
+ TOFDC_split/Test/RGB/20200928_151401_rgb_crop.png,TOFDC_split/Test/GT/20200928_151401_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200928_151401_depth_crop_fill.png
145
+ TOFDC_split/Test/RGB/20200920_172019_rgb_crop.png,TOFDC_split/Test/GT/20200920_172019_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_172019_depth_crop_fill.png
146
+ TOFDC_split/Test/RGB/20200602_154724_rgb_crop.png,TOFDC_split/Test/GT/20200602_154724_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200602_154724_depth_crop_fill.png
147
+ TOFDC_split/Test/RGB/20200824_171017_rgb_crop.png,TOFDC_split/Test/GT/20200824_171017_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_171017_depth_crop_fill.png
148
+ TOFDC_split/Test/RGB/20200820_103917_rgb_crop.png,TOFDC_split/Test/GT/20200820_103917_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200820_103917_depth_crop_fill.png
149
+ TOFDC_split/Test/RGB/20200531_170527_rgb_crop.png,TOFDC_split/Test/GT/20200531_170527_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200531_170527_depth_crop_fill.png
150
+ TOFDC_split/Test/RGB/2020_09_09_11_35_41_090_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_11_35_41_090_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_11_35_41_090_rgb_depth_crop_fill.png
151
+ TOFDC_split/Test/RGB/2020_09_11_22_09_17_091_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_11_22_09_17_091_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_11_22_09_17_091_rgb_depth_crop_fill.png
152
+ TOFDC_split/Test/RGB/2020_09_21_21_12_25_296_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_21_21_12_25_296_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_21_21_12_25_296_rgb_depth_crop_fill.png
153
+ TOFDC_split/Test/RGB/2020_09_09_20_50_10_664_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_20_50_10_664_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_20_50_10_664_rgb_depth_crop_fill.png
154
+ TOFDC_split/Test/RGB/2020_09_22_09_57_03_131_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_22_09_57_03_131_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_22_09_57_03_131_rgb_depth_crop_fill.png
155
+ TOFDC_split/Test/RGB/20200818_144858_rgb_crop.png,TOFDC_split/Test/GT/20200818_144858_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_144858_depth_crop_fill.png
156
+ TOFDC_split/Test/RGB/20200819_184628_rgb_crop.png,TOFDC_split/Test/GT/20200819_184628_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_184628_depth_crop_fill.png
157
+ TOFDC_split/Test/RGB/20200919_195247_rgb_crop.png,TOFDC_split/Test/GT/20200919_195247_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_195247_depth_crop_fill.png
158
+ TOFDC_split/Test/RGB/2020_09_11_11_54_40_367_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_11_11_54_40_367_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_11_11_54_40_367_rgb_depth_crop_fill.png
159
+ TOFDC_split/Test/RGB/2020_09_22_10_35_16_382_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_22_10_35_16_382_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_22_10_35_16_382_rgb_depth_crop_fill.png
160
+ TOFDC_split/Test/RGB/2020_09_26_11_23_20_457_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_26_11_23_20_457_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_26_11_23_20_457_rgb_depth_crop_fill.png
161
+ TOFDC_split/Test/RGB/2020_09_12_11_42_17_374_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_11_42_17_374_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_11_42_17_374_rgb_depth_crop_fill.png
162
+ TOFDC_split/Test/RGB/2020_09_26_15_17_14_600_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_26_15_17_14_600_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_26_15_17_14_600_rgb_depth_crop_fill.png
163
+ TOFDC_split/Test/RGB/2020_09_10_21_53_39_630_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_10_21_53_39_630_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_10_21_53_39_630_rgb_depth_crop_fill.png
164
+ TOFDC_split/Test/RGB/20200717_194846_rgb_crop.png,TOFDC_split/Test/GT/20200717_194846_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200717_194846_depth_crop_fill.png
165
+ TOFDC_split/Test/RGB/20200919_094258_rgb_crop.png,TOFDC_split/Test/GT/20200919_094258_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_094258_depth_crop_fill.png
166
+ TOFDC_split/Test/RGB/20200819_180228_rgb_crop.png,TOFDC_split/Test/GT/20200819_180228_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_180228_depth_crop_fill.png
167
+ TOFDC_split/Test/RGB/20200920_152749_rgb_crop.png,TOFDC_split/Test/GT/20200920_152749_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_152749_depth_crop_fill.png
168
+ TOFDC_split/Test/RGB/20200725_163942_rgb_crop.png,TOFDC_split/Test/GT/20200725_163942_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200725_163942_depth_crop_fill.png
169
+ TOFDC_split/Test/RGB/20200920_164908_rgb_crop.png,TOFDC_split/Test/GT/20200920_164908_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_164908_depth_crop_fill.png
170
+ TOFDC_split/Test/RGB/20200819_204159_rgb_crop.png,TOFDC_split/Test/GT/20200819_204159_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_204159_depth_crop_fill.png
171
+ TOFDC_split/Test/RGB/20200928_150743_rgb_crop.png,TOFDC_split/Test/GT/20200928_150743_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200928_150743_depth_crop_fill.png
172
+ TOFDC_split/Test/RGB/20200725_094624_rgb_crop.png,TOFDC_split/Test/GT/20200725_094624_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200725_094624_depth_crop_fill.png
173
+ TOFDC_split/Test/RGB/20200606_155808_rgb_crop.png,TOFDC_split/Test/GT/20200606_155808_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200606_155808_depth_crop_fill.png
174
+ TOFDC_split/Test/RGB/20200927_153237_rgb_crop.png,TOFDC_split/Test/GT/20200927_153237_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_153237_depth_crop_fill.png
175
+ TOFDC_split/Test/RGB/20200717_160548_rgb_crop.png,TOFDC_split/Test/GT/20200717_160548_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200717_160548_depth_crop_fill.png
176
+ TOFDC_split/Test/RGB/20200928_151544_rgb_crop.png,TOFDC_split/Test/GT/20200928_151544_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200928_151544_depth_crop_fill.png
177
+ TOFDC_split/Test/RGB/20200819_181407_rgb_crop.png,TOFDC_split/Test/GT/20200819_181407_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_181407_depth_crop_fill.png
178
+ TOFDC_split/Test/RGB/20200824_163933_rgb_crop.png,TOFDC_split/Test/GT/20200824_163933_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_163933_depth_crop_fill.png
179
+ TOFDC_split/Test/RGB/20200819_202647_rgb_crop.png,TOFDC_split/Test/GT/20200819_202647_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_202647_depth_crop_fill.png
180
+ TOFDC_split/Test/RGB/20200719_145453_rgb_crop.png,TOFDC_split/Test/GT/20200719_145453_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200719_145453_depth_crop_fill.png
181
+ TOFDC_split/Test/RGB/2020_09_21_21_00_09_208_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_21_21_00_09_208_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_21_21_00_09_208_rgb_depth_crop_fill.png
182
+ TOFDC_split/Test/RGB/20200819_145821_rgb_crop.png,TOFDC_split/Test/GT/20200819_145821_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_145821_depth_crop_fill.png
183
+ TOFDC_split/Test/RGB/2020_09_11_22_04_04_209_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_11_22_04_04_209_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_11_22_04_04_209_rgb_depth_crop_fill.png
184
+ TOFDC_split/Test/RGB/20200818_213907_rgb_crop.png,TOFDC_split/Test/GT/20200818_213907_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_213907_depth_crop_fill.png
185
+ TOFDC_split/Test/RGB/20200819_134535_rgb_crop.png,TOFDC_split/Test/GT/20200819_134535_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_134535_depth_crop_fill.png
186
+ TOFDC_split/Test/RGB/20200819_203952_rgb_crop.png,TOFDC_split/Test/GT/20200819_203952_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_203952_depth_crop_fill.png
187
+ TOFDC_split/Test/RGB/20200927_140000_rgb_crop.png,TOFDC_split/Test/GT/20200927_140000_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_140000_depth_crop_fill.png
188
+ TOFDC_split/Test/RGB/2020_09_09_11_40_52_856_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_11_40_52_856_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_11_40_52_856_rgb_depth_crop_fill.png
189
+ TOFDC_split/Test/RGB/2020_09_08_17_08_24_083_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_17_08_24_083_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_17_08_24_083_rgb_depth_crop_fill.png
190
+ TOFDC_split/Test/RGB/20200927_201011_rgb_crop.png,TOFDC_split/Test/GT/20200927_201011_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_201011_depth_crop_fill.png
191
+ TOFDC_split/Test/RGB/2020_09_12_21_34_27_739_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_21_34_27_739_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_21_34_27_739_rgb_depth_crop_fill.png
192
+ TOFDC_split/Test/RGB/2020_09_14_23_36_46_156_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_14_23_36_46_156_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_14_23_36_46_156_rgb_depth_crop_fill.png
193
+ TOFDC_split/Test/RGB/20200721_200407_rgb_crop.png,TOFDC_split/Test/GT/20200721_200407_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200721_200407_depth_crop_fill.png
194
+ TOFDC_split/Test/RGB/20200717_110213_rgb_crop.png,TOFDC_split/Test/GT/20200717_110213_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200717_110213_depth_crop_fill.png
195
+ TOFDC_split/Test/RGB/20200920_170202_rgb_crop.png,TOFDC_split/Test/GT/20200920_170202_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_170202_depth_crop_fill.png
196
+ TOFDC_split/Test/RGB/2020_09_13_19_05_26_941_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_13_19_05_26_941_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_13_19_05_26_941_rgb_depth_crop_fill.png
197
+ TOFDC_split/Test/RGB/20200722_095703_rgb_crop.png,TOFDC_split/Test/GT/20200722_095703_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200722_095703_depth_crop_fill.png
198
+ TOFDC_split/Test/RGB/20200819_134306_rgb_crop.png,TOFDC_split/Test/GT/20200819_134306_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_134306_depth_crop_fill.png
199
+ TOFDC_split/Test/RGB/2020_09_12_11_25_02_700_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_11_25_02_700_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_11_25_02_700_rgb_depth_crop_fill.png
200
+ TOFDC_split/Test/RGB/20200818_153720_rgb_crop.png,TOFDC_split/Test/GT/20200818_153720_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_153720_depth_crop_fill.png
201
+ TOFDC_split/Test/RGB/20200819_104804_rgb_crop.png,TOFDC_split/Test/GT/20200819_104804_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_104804_depth_crop_fill.png
202
+ TOFDC_split/Test/RGB/20200927_203912_rgb_crop.png,TOFDC_split/Test/GT/20200927_203912_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_203912_depth_crop_fill.png
203
+ TOFDC_split/Test/RGB/2020_09_10_11_31_09_140_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_10_11_31_09_140_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_10_11_31_09_140_rgb_depth_crop_fill.png
204
+ TOFDC_split/Test/RGB/20200818_151128_rgb_crop.png,TOFDC_split/Test/GT/20200818_151128_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_151128_depth_crop_fill.png
205
+ TOFDC_split/Test/RGB/20200820_144718_rgb_crop.png,TOFDC_split/Test/GT/20200820_144718_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200820_144718_depth_crop_fill.png
206
+ TOFDC_split/Test/RGB/20200920_154626_rgb_crop.png,TOFDC_split/Test/GT/20200920_154626_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_154626_depth_crop_fill.png
207
+ TOFDC_split/Test/RGB/20200927_203237_rgb_crop.png,TOFDC_split/Test/GT/20200927_203237_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_203237_depth_crop_fill.png
208
+ TOFDC_split/Test/RGB/20200818_123732_rgb_crop.png,TOFDC_split/Test/GT/20200818_123732_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_123732_depth_crop_fill.png
209
+ TOFDC_split/Test/RGB/20200927_141218_rgb_crop.png,TOFDC_split/Test/GT/20200927_141218_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_141218_depth_crop_fill.png
210
+ TOFDC_split/Test/RGB/20200927_140327_rgb_crop.png,TOFDC_split/Test/GT/20200927_140327_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_140327_depth_crop_fill.png
211
+ TOFDC_split/Test/RGB/20200531_165554_rgb_crop.png,TOFDC_split/Test/GT/20200531_165554_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200531_165554_depth_crop_fill.png
212
+ TOFDC_split/Test/RGB/20200921_143357_rgb_crop.png,TOFDC_split/Test/GT/20200921_143357_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200921_143357_depth_crop_fill.png
213
+ TOFDC_split/Test/RGB/20200818_212139_rgb_crop.png,TOFDC_split/Test/GT/20200818_212139_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_212139_depth_crop_fill.png
214
+ TOFDC_split/Test/RGB/20200903_173732_rgb_crop.png,TOFDC_split/Test/GT/20200903_173732_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200903_173732_depth_crop_fill.png
215
+ TOFDC_split/Test/RGB/20200723_175217_rgb_crop.png,TOFDC_split/Test/GT/20200723_175217_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200723_175217_depth_crop_fill.png
216
+ TOFDC_split/Test/RGB/20200903_173336_rgb_crop.png,TOFDC_split/Test/GT/20200903_173336_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200903_173336_depth_crop_fill.png
217
+ TOFDC_split/Test/RGB/20200819_121346_rgb_crop.png,TOFDC_split/Test/GT/20200819_121346_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_121346_depth_crop_fill.png
218
+ TOFDC_split/Test/RGB/20200824_175053_rgb_crop.png,TOFDC_split/Test/GT/20200824_175053_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_175053_depth_crop_fill.png
219
+ TOFDC_split/Test/RGB/20200603_205658_rgb_crop.png,TOFDC_split/Test/GT/20200603_205658_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200603_205658_depth_crop_fill.png
220
+ TOFDC_split/Test/RGB/2020_09_08_16_52_41_055_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_16_52_41_055_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_16_52_41_055_rgb_depth_crop_fill.png
221
+ TOFDC_split/Test/RGB/2020_09_22_10_23_50_598_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_22_10_23_50_598_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_22_10_23_50_598_rgb_depth_crop_fill.png
222
+ TOFDC_split/Test/RGB/2020_09_22_21_26_23_881_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_22_21_26_23_881_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_22_21_26_23_881_rgb_depth_crop_fill.png
223
+ TOFDC_split/Test/RGB/2020_09_08_14_08_15_244_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_14_08_15_244_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_14_08_15_244_rgb_depth_crop_fill.png
224
+ TOFDC_split/Test/RGB/2020_09_23_21_47_53_262_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_23_21_47_53_262_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_23_21_47_53_262_rgb_depth_crop_fill.png
225
+ TOFDC_split/Test/RGB/2020_09_26_11_11_14_488_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_26_11_11_14_488_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_26_11_11_14_488_rgb_depth_crop_fill.png
226
+ TOFDC_split/Test/RGB/2020_09_08_17_52_29_102_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_17_52_29_102_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_17_52_29_102_rgb_depth_crop_fill.png
227
+ TOFDC_split/Test/RGB/2020_09_11_17_39_08_025_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_11_17_39_08_025_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_11_17_39_08_025_rgb_depth_crop_fill.png
228
+ TOFDC_split/Test/RGB/2020_09_14_23_21_39_930_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_14_23_21_39_930_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_14_23_21_39_930_rgb_depth_crop_fill.png
229
+ TOFDC_split/Test/RGB/20200920_112301_rgb_crop.png,TOFDC_split/Test/GT/20200920_112301_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_112301_depth_crop_fill.png
230
+ TOFDC_split/Test/RGB/20200919_094704_rgb_crop.png,TOFDC_split/Test/GT/20200919_094704_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_094704_depth_crop_fill.png
231
+ TOFDC_split/Test/RGB/2020_09_12_21_23_54_045_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_21_23_54_045_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_21_23_54_045_rgb_depth_crop_fill.png
232
+ TOFDC_split/Test/RGB/2020_09_12_10_49_10_074_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_10_49_10_074_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_10_49_10_074_rgb_depth_crop_fill.png
233
+ TOFDC_split/Test/RGB/2020_09_11_11_37_50_280_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_11_11_37_50_280_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_11_11_37_50_280_rgb_depth_crop_fill.png
234
+ TOFDC_split/Test/RGB/20200919_165543_rgb_crop.png,TOFDC_split/Test/GT/20200919_165543_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_165543_depth_crop_fill.png
235
+ TOFDC_split/Test/RGB/2020_09_11_22_23_15_039_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_11_22_23_15_039_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_11_22_23_15_039_rgb_depth_crop_fill.png
236
+ TOFDC_split/Test/RGB/2020_09_11_11_48_08_203_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_11_11_48_08_203_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_11_11_48_08_203_rgb_depth_crop_fill.png
237
+ TOFDC_split/Test/RGB/2020_09_26_11_31_05_109_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_26_11_31_05_109_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_26_11_31_05_109_rgb_depth_crop_fill.png
238
+ TOFDC_split/Test/RGB/2020_09_06_21_26_30_171_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_06_21_26_30_171_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_06_21_26_30_171_rgb_depth_crop_fill.png
239
+ TOFDC_split/Test/RGB/20200717_095653_rgb_crop.png,TOFDC_split/Test/GT/20200717_095653_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200717_095653_depth_crop_fill.png
240
+ TOFDC_split/Test/RGB/2020_09_22_16_01_03_086_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_22_16_01_03_086_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_22_16_01_03_086_rgb_depth_crop_fill.png
241
+ TOFDC_split/Test/RGB/20200919_152619_rgb_crop.png,TOFDC_split/Test/GT/20200919_152619_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_152619_depth_crop_fill.png
242
+ TOFDC_split/Test/RGB/20200927_143926_rgb_crop.png,TOFDC_split/Test/GT/20200927_143926_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_143926_depth_crop_fill.png
243
+ TOFDC_split/Test/RGB/20200927_193552_rgb_crop.png,TOFDC_split/Test/GT/20200927_193552_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_193552_depth_crop_fill.png
244
+ TOFDC_split/Test/RGB/20200819_175048_rgb_crop.png,TOFDC_split/Test/GT/20200819_175048_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_175048_depth_crop_fill.png
245
+ TOFDC_split/Test/RGB/2020_09_10_11_19_33_244_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_10_11_19_33_244_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_10_11_19_33_244_rgb_depth_crop_fill.png
246
+ TOFDC_split/Test/RGB/20200727_172827_rgb_crop.png,TOFDC_split/Test/GT/20200727_172827_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200727_172827_depth_crop_fill.png
247
+ TOFDC_split/Test/RGB/2020_09_12_15_02_15_336_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_15_02_15_336_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_15_02_15_336_rgb_depth_crop_fill.png
248
+ TOFDC_split/Test/RGB/20200819_175543_rgb_crop.png,TOFDC_split/Test/GT/20200819_175543_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_175543_depth_crop_fill.png
249
+ TOFDC_split/Test/RGB/2020_09_21_20_50_54_019_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_21_20_50_54_019_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_21_20_50_54_019_rgb_depth_crop_fill.png
250
+ TOFDC_split/Test/RGB/2020_09_10_22_11_37_156_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_10_22_11_37_156_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_10_22_11_37_156_rgb_depth_crop_fill.png
251
+ TOFDC_split/Test/RGB/20200819_163645_rgb_crop.png,TOFDC_split/Test/GT/20200819_163645_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_163645_depth_crop_fill.png
252
+ TOFDC_split/Test/RGB/2020_09_09_14_58_41_346_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_14_58_41_346_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_14_58_41_346_rgb_depth_crop_fill.png
253
+ TOFDC_split/Test/RGB/20200919_100421_rgb_crop.png,TOFDC_split/Test/GT/20200919_100421_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_100421_depth_crop_fill.png
254
+ TOFDC_split/Test/RGB/20200720_100836_rgb_crop.png,TOFDC_split/Test/GT/20200720_100836_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200720_100836_depth_crop_fill.png
255
+ TOFDC_split/Test/RGB/20200603_125406_rgb_crop.png,TOFDC_split/Test/GT/20200603_125406_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200603_125406_depth_crop_fill.png
256
+ TOFDC_split/Test/RGB/20200605_100852_rgb_crop.png,TOFDC_split/Test/GT/20200605_100852_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200605_100852_depth_crop_fill.png
257
+ TOFDC_split/Test/RGB/20200831_103209_rgb_crop.png,TOFDC_split/Test/GT/20200831_103209_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200831_103209_depth_crop_fill.png
258
+ TOFDC_split/Test/RGB/20200820_161810_rgb_crop.png,TOFDC_split/Test/GT/20200820_161810_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200820_161810_depth_crop_fill.png
259
+ TOFDC_split/Test/RGB/2020_09_09_16_41_42_774_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_16_41_42_774_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_16_41_42_774_rgb_depth_crop_fill.png
260
+ TOFDC_split/Test/RGB/20200601_141248_rgb_crop.png,TOFDC_split/Test/GT/20200601_141248_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200601_141248_depth_crop_fill.png
261
+ TOFDC_split/Test/RGB/20200820_114434_rgb_crop.png,TOFDC_split/Test/GT/20200820_114434_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200820_114434_depth_crop_fill.png
262
+ TOFDC_split/Test/RGB/2020_09_23_22_39_23_658_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_23_22_39_23_658_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_23_22_39_23_658_rgb_depth_crop_fill.png
263
+ TOFDC_split/Test/RGB/20200820_160235_rgb_crop.png,TOFDC_split/Test/GT/20200820_160235_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200820_160235_depth_crop_fill.png
264
+ TOFDC_split/Test/RGB/20200919_104030_rgb_crop.png,TOFDC_split/Test/GT/20200919_104030_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_104030_depth_crop_fill.png
265
+ TOFDC_split/Test/RGB/20200604_155501_rgb_crop.png,TOFDC_split/Test/GT/20200604_155501_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200604_155501_depth_crop_fill.png
266
+ TOFDC_split/Test/RGB/20200818_214102_rgb_crop.png,TOFDC_split/Test/GT/20200818_214102_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_214102_depth_crop_fill.png
267
+ TOFDC_split/Test/RGB/20200725_161320_rgb_crop.png,TOFDC_split/Test/GT/20200725_161320_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200725_161320_depth_crop_fill.png
268
+ TOFDC_split/Test/RGB/20200903_155040_rgb_crop.png,TOFDC_split/Test/GT/20200903_155040_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200903_155040_depth_crop_fill.png
269
+ TOFDC_split/Test/RGB/20200717_104214_rgb_crop.png,TOFDC_split/Test/GT/20200717_104214_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200717_104214_depth_crop_fill.png
270
+ TOFDC_split/Test/RGB/20200820_142910_rgb_crop.png,TOFDC_split/Test/GT/20200820_142910_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200820_142910_depth_crop_fill.png
271
+ TOFDC_split/Test/RGB/20200725_160150_rgb_crop.png,TOFDC_split/Test/GT/20200725_160150_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200725_160150_depth_crop_fill.png
272
+ TOFDC_split/Test/RGB/20200824_152521_rgb_crop.png,TOFDC_split/Test/GT/20200824_152521_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_152521_depth_crop_fill.png
273
+ TOFDC_split/Test/RGB/20200819_200541_rgb_crop.png,TOFDC_split/Test/GT/20200819_200541_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_200541_depth_crop_fill.png
274
+ TOFDC_split/Test/RGB/20200722_143443_rgb_crop.png,TOFDC_split/Test/GT/20200722_143443_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200722_143443_depth_crop_fill.png
275
+ TOFDC_split/Test/RGB/20200819_160230_rgb_crop.png,TOFDC_split/Test/GT/20200819_160230_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_160230_depth_crop_fill.png
276
+ TOFDC_split/Test/RGB/20200604_161732_rgb_crop.png,TOFDC_split/Test/GT/20200604_161732_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200604_161732_depth_crop_fill.png
277
+ TOFDC_split/Test/RGB/2020_09_23_22_47_07_966_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_23_22_47_07_966_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_23_22_47_07_966_rgb_depth_crop_fill.png
278
+ TOFDC_split/Test/RGB/20200608_084013_rgb_crop.png,TOFDC_split/Test/GT/20200608_084013_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200608_084013_depth_crop_fill.png
279
+ TOFDC_split/Test/RGB/20200920_104950_rgb_crop.png,TOFDC_split/Test/GT/20200920_104950_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_104950_depth_crop_fill.png
280
+ TOFDC_split/Test/RGB/20200927_203124_rgb_crop.png,TOFDC_split/Test/GT/20200927_203124_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_203124_depth_crop_fill.png
281
+ TOFDC_split/Test/RGB/20200717_201642_rgb_crop.png,TOFDC_split/Test/GT/20200717_201642_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200717_201642_depth_crop_fill.png
282
+ TOFDC_split/Test/RGB/20200719_103252_rgb_crop.png,TOFDC_split/Test/GT/20200719_103252_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200719_103252_depth_crop_fill.png
283
+ TOFDC_split/Test/RGB/20200920_191008_rgb_crop.png,TOFDC_split/Test/GT/20200920_191008_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_191008_depth_crop_fill.png
284
+ TOFDC_split/Test/RGB/20200718_201047_rgb_crop.png,TOFDC_split/Test/GT/20200718_201047_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200718_201047_depth_crop_fill.png
285
+ TOFDC_split/Test/RGB/20200722_155035_rgb_crop.png,TOFDC_split/Test/GT/20200722_155035_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200722_155035_depth_crop_fill.png
286
+ TOFDC_split/Test/RGB/20200603_122349_rgb_crop.png,TOFDC_split/Test/GT/20200603_122349_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200603_122349_depth_crop_fill.png
287
+ TOFDC_split/Test/RGB/20200824_154114_rgb_crop.png,TOFDC_split/Test/GT/20200824_154114_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_154114_depth_crop_fill.png
288
+ TOFDC_split/Test/RGB/20200725_160612_rgb_crop.png,TOFDC_split/Test/GT/20200725_160612_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200725_160612_depth_crop_fill.png
289
+ TOFDC_split/Test/RGB/20200606_162252_rgb_crop.png,TOFDC_split/Test/GT/20200606_162252_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200606_162252_depth_crop_fill.png
290
+ TOFDC_split/Test/RGB/20200920_151138_rgb_crop.png,TOFDC_split/Test/GT/20200920_151138_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_151138_depth_crop_fill.png
291
+ TOFDC_split/Test/RGB/20200725_163501_rgb_crop.png,TOFDC_split/Test/GT/20200725_163501_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200725_163501_depth_crop_fill.png
292
+ TOFDC_split/Test/RGB/20200531_171903_rgb_crop.png,TOFDC_split/Test/GT/20200531_171903_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200531_171903_depth_crop_fill.png
293
+ TOFDC_split/Test/RGB/20200920_104725_rgb_crop.png,TOFDC_split/Test/GT/20200920_104725_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_104725_depth_crop_fill.png
294
+ TOFDC_split/Test/RGB/20200820_104816_rgb_crop.png,TOFDC_split/Test/GT/20200820_104816_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200820_104816_depth_crop_fill.png
295
+ TOFDC_split/Test/RGB/20200819_182001_rgb_crop.png,TOFDC_split/Test/GT/20200819_182001_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_182001_depth_crop_fill.png
296
+ TOFDC_split/Test/RGB/20200718_100742_rgb_crop.png,TOFDC_split/Test/GT/20200718_100742_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200718_100742_depth_crop_fill.png
297
+ TOFDC_split/Test/RGB/20200602_162319_rgb_crop.png,TOFDC_split/Test/GT/20200602_162319_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200602_162319_depth_crop_fill.png
298
+ TOFDC_split/Test/RGB/20200927_134848_rgb_crop.png,TOFDC_split/Test/GT/20200927_134848_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_134848_depth_crop_fill.png
299
+ TOFDC_split/Test/RGB/20200726_095539_rgb_crop.png,TOFDC_split/Test/GT/20200726_095539_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200726_095539_depth_crop_fill.png
300
+ TOFDC_split/Test/RGB/2020_09_11_21_50_06_664_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_11_21_50_06_664_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_11_21_50_06_664_rgb_depth_crop_fill.png
301
+ TOFDC_split/Test/RGB/20200819_183727_rgb_crop.png,TOFDC_split/Test/GT/20200819_183727_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_183727_depth_crop_fill.png
302
+ TOFDC_split/Test/RGB/2020_09_23_21_34_14_417_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_23_21_34_14_417_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_23_21_34_14_417_rgb_depth_crop_fill.png
303
+ TOFDC_split/Test/RGB/2020_09_10_16_04_24_266_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_10_16_04_24_266_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_10_16_04_24_266_rgb_depth_crop_fill.png
304
+ TOFDC_split/Test/RGB/2020_09_10_10_48_18_225_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_10_10_48_18_225_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_10_10_48_18_225_rgb_depth_crop_fill.png
305
+ TOFDC_split/Test/RGB/2020_09_22_16_36_21_153_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_22_16_36_21_153_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_22_16_36_21_153_rgb_depth_crop_fill.png
306
+ TOFDC_split/Test/RGB/20200819_163855_rgb_crop.png,TOFDC_split/Test/GT/20200819_163855_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_163855_depth_crop_fill.png
307
+ TOFDC_split/Test/RGB/20200928_153558_rgb_crop.png,TOFDC_split/Test/GT/20200928_153558_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200928_153558_depth_crop_fill.png
308
+ TOFDC_split/Test/RGB/2020_09_10_16_17_50_551_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_10_16_17_50_551_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_10_16_17_50_551_rgb_depth_crop_fill.png
309
+ TOFDC_split/Test/RGB/20200727_165514_rgb_crop.png,TOFDC_split/Test/GT/20200727_165514_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200727_165514_depth_crop_fill.png
310
+ TOFDC_split/Test/RGB/20200818_150942_rgb_crop.png,TOFDC_split/Test/GT/20200818_150942_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_150942_depth_crop_fill.png
311
+ TOFDC_split/Test/RGB/2020_09_10_22_22_02_175_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_10_22_22_02_175_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_10_22_22_02_175_rgb_depth_crop_fill.png
312
+ TOFDC_split/Test/RGB/20200819_153344_rgb_crop.png,TOFDC_split/Test/GT/20200819_153344_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_153344_depth_crop_fill.png
313
+ TOFDC_split/Test/RGB/20200903_175058_rgb_crop.png,TOFDC_split/Test/GT/20200903_175058_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200903_175058_depth_crop_fill.png
314
+ TOFDC_split/Test/RGB/20200720_195905_rgb_crop.png,TOFDC_split/Test/GT/20200720_195905_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200720_195905_depth_crop_fill.png
315
+ TOFDC_split/Test/RGB/20200819_130736_rgb_crop.png,TOFDC_split/Test/GT/20200819_130736_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_130736_depth_crop_fill.png
316
+ TOFDC_split/Test/RGB/20200919_102349_rgb_crop.png,TOFDC_split/Test/GT/20200919_102349_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_102349_depth_crop_fill.png
317
+ TOFDC_split/Test/RGB/20200718_153607_rgb_crop.png,TOFDC_split/Test/GT/20200718_153607_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200718_153607_depth_crop_fill.png
318
+ TOFDC_split/Test/RGB/20200927_161039_rgb_crop.png,TOFDC_split/Test/GT/20200927_161039_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_161039_depth_crop_fill.png
319
+ TOFDC_split/Test/RGB/20200725_164346_rgb_crop.png,TOFDC_split/Test/GT/20200725_164346_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200725_164346_depth_crop_fill.png
320
+ TOFDC_split/Test/RGB/20200919_172225_rgb_crop.png,TOFDC_split/Test/GT/20200919_172225_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_172225_depth_crop_fill.png
321
+ TOFDC_split/Test/RGB/20200603_094538_rgb_crop.png,TOFDC_split/Test/GT/20200603_094538_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200603_094538_depth_crop_fill.png
322
+ TOFDC_split/Test/RGB/20200531_164918_rgb_crop.png,TOFDC_split/Test/GT/20200531_164918_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200531_164918_depth_crop_fill.png
323
+ TOFDC_split/Test/RGB/20200818_154515_rgb_crop.png,TOFDC_split/Test/GT/20200818_154515_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_154515_depth_crop_fill.png
324
+ TOFDC_split/Test/RGB/20200726_101759_rgb_crop.png,TOFDC_split/Test/GT/20200726_101759_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200726_101759_depth_crop_fill.png
325
+ TOFDC_split/Test/RGB/2020_09_22_15_50_03_213_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_22_15_50_03_213_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_22_15_50_03_213_rgb_depth_crop_fill.png
326
+ TOFDC_split/Test/RGB/20200819_200819_rgb_crop.png,TOFDC_split/Test/GT/20200819_200819_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_200819_depth_crop_fill.png
327
+ TOFDC_split/Test/RGB/20200927_194033_rgb_crop.png,TOFDC_split/Test/GT/20200927_194033_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_194033_depth_crop_fill.png
328
+ TOFDC_split/Test/RGB/20200920_164646_rgb_crop.png,TOFDC_split/Test/GT/20200920_164646_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_164646_depth_crop_fill.png
329
+ TOFDC_split/Test/RGB/20200927_141806_rgb_crop.png,TOFDC_split/Test/GT/20200927_141806_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_141806_depth_crop_fill.png
330
+ TOFDC_split/Test/RGB/20200726_095225_rgb_crop.png,TOFDC_split/Test/GT/20200726_095225_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200726_095225_depth_crop_fill.png
331
+ TOFDC_split/Test/RGB/20200824_173922_rgb_crop.png,TOFDC_split/Test/GT/20200824_173922_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_173922_depth_crop_fill.png
332
+ TOFDC_split/Test/RGB/20200824_170457_rgb_crop.png,TOFDC_split/Test/GT/20200824_170457_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_170457_depth_crop_fill.png
333
+ TOFDC_split/Test/RGB/2020_09_23_22_14_40_589_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_23_22_14_40_589_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_23_22_14_40_589_rgb_depth_crop_fill.png
334
+ TOFDC_split/Test/RGB/2020_09_23_22_09_59_260_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_23_22_09_59_260_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_23_22_09_59_260_rgb_depth_crop_fill.png
335
+ TOFDC_split/Test/RGB/2020_09_12_14_51_15_488_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_14_51_15_488_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_14_51_15_488_rgb_depth_crop_fill.png
336
+ TOFDC_split/Test/RGB/20200723_202303_rgb_crop.png,TOFDC_split/Test/GT/20200723_202303_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200723_202303_depth_crop_fill.png
337
+ TOFDC_split/Test/RGB/2020_09_26_15_10_41_668_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_26_15_10_41_668_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_26_15_10_41_668_rgb_depth_crop_fill.png
338
+ TOFDC_split/Test/RGB/2020_09_09_17_16_13_642_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_17_16_13_642_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_17_16_13_642_rgb_depth_crop_fill.png
339
+ TOFDC_split/Test/RGB/20200721_165000_rgb_crop.png,TOFDC_split/Test/GT/20200721_165000_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200721_165000_depth_crop_fill.png
340
+ TOFDC_split/Test/RGB/20200601_161016_rgb_crop.png,TOFDC_split/Test/GT/20200601_161016_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200601_161016_depth_crop_fill.png
341
+ TOFDC_split/Test/RGB/20200921_143002_rgb_crop.png,TOFDC_split/Test/GT/20200921_143002_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200921_143002_depth_crop_fill.png
342
+ TOFDC_split/Test/RGB/20200824_111302_rgb_crop.png,TOFDC_split/Test/GT/20200824_111302_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_111302_depth_crop_fill.png
343
+ TOFDC_split/Test/RGB/2020_09_10_21_43_45_695_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_10_21_43_45_695_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_10_21_43_45_695_rgb_depth_crop_fill.png
344
+ TOFDC_split/Test/RGB/2020_09_12_11_15_44_885_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_11_15_44_885_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_11_15_44_885_rgb_depth_crop_fill.png
345
+ TOFDC_split/Test/RGB/20200920_112430_rgb_crop.png,TOFDC_split/Test/GT/20200920_112430_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_112430_depth_crop_fill.png
346
+ TOFDC_split/Test/RGB/20200720_201235_rgb_crop.png,TOFDC_split/Test/GT/20200720_201235_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200720_201235_depth_crop_fill.png
347
+ TOFDC_split/Test/RGB/2020_09_09_17_11_38_064_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_17_11_38_064_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_17_11_38_064_rgb_depth_crop_fill.png
348
+ TOFDC_split/Test/RGB/2020_09_22_14_46_09_471_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_22_14_46_09_471_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_22_14_46_09_471_rgb_depth_crop_fill.png
349
+ TOFDC_split/Test/RGB/2020_09_12_15_09_33_394_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_15_09_33_394_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_15_09_33_394_rgb_depth_crop_fill.png
350
+ TOFDC_split/Test/RGB/20200820_095802_rgb_crop.png,TOFDC_split/Test/GT/20200820_095802_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200820_095802_depth_crop_fill.png
351
+ TOFDC_split/Test/RGB/20200719_203741_rgb_crop.png,TOFDC_split/Test/GT/20200719_203741_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200719_203741_depth_crop_fill.png
352
+ TOFDC_split/Test/RGB/20200927_202732_rgb_crop.png,TOFDC_split/Test/GT/20200927_202732_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_202732_depth_crop_fill.png
353
+ TOFDC_split/Test/RGB/20200722_194018_rgb_crop.png,TOFDC_split/Test/GT/20200722_194018_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200722_194018_depth_crop_fill.png
354
+ TOFDC_split/Test/RGB/20200604_145419_rgb_crop.png,TOFDC_split/Test/GT/20200604_145419_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200604_145419_depth_crop_fill.png
355
+ TOFDC_split/Test/RGB/20200607_102459_rgb_crop.png,TOFDC_split/Test/GT/20200607_102459_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200607_102459_depth_crop_fill.png
356
+ TOFDC_split/Test/RGB/20200919_100905_rgb_crop.png,TOFDC_split/Test/GT/20200919_100905_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_100905_depth_crop_fill.png
357
+ TOFDC_split/Test/RGB/20200717_193033_rgb_crop.png,TOFDC_split/Test/GT/20200717_193033_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200717_193033_depth_crop_fill.png
358
+ TOFDC_split/Test/RGB/20200818_122903_rgb_crop.png,TOFDC_split/Test/GT/20200818_122903_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_122903_depth_crop_fill.png
359
+ TOFDC_split/Test/RGB/20200824_165134_rgb_crop.png,TOFDC_split/Test/GT/20200824_165134_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_165134_depth_crop_fill.png
360
+ TOFDC_split/Test/RGB/20200818_103715_rgb_crop.png,TOFDC_split/Test/GT/20200818_103715_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_103715_depth_crop_fill.png
361
+ TOFDC_split/Test/RGB/20200605_094604_rgb_crop.png,TOFDC_split/Test/GT/20200605_094604_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200605_094604_depth_crop_fill.png
362
+ TOFDC_split/Test/RGB/2020_09_08_13_59_59_435_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_13_59_59_435_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_13_59_59_435_rgb_depth_crop_fill.png
363
+ TOFDC_split/Test/RGB/20200920_153941_rgb_crop.png,TOFDC_split/Test/GT/20200920_153941_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_153941_depth_crop_fill.png
364
+ TOFDC_split/Test/RGB/20200824_162137_rgb_crop.png,TOFDC_split/Test/GT/20200824_162137_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_162137_depth_crop_fill.png
365
+ TOFDC_split/Test/RGB/20200927_160450_rgb_crop.png,TOFDC_split/Test/GT/20200927_160450_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_160450_depth_crop_fill.png
366
+ TOFDC_split/Test/RGB/20200927_201454_rgb_crop.png,TOFDC_split/Test/GT/20200927_201454_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_201454_depth_crop_fill.png
367
+ TOFDC_split/Test/RGB/20200927_140533_rgb_crop.png,TOFDC_split/Test/GT/20200927_140533_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_140533_depth_crop_fill.png
368
+ TOFDC_split/Test/RGB/20200608_092324_rgb_crop.png,TOFDC_split/Test/GT/20200608_092324_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200608_092324_depth_crop_fill.png
369
+ TOFDC_split/Test/RGB/2020_09_10_00_00_06_823_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_10_00_00_06_823_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_10_00_00_06_823_rgb_depth_crop_fill.png
370
+ TOFDC_split/Test/RGB/20200727_170712_rgb_crop.png,TOFDC_split/Test/GT/20200727_170712_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200727_170712_depth_crop_fill.png
371
+ TOFDC_split/Test/RGB/20200921_141709_rgb_crop.png,TOFDC_split/Test/GT/20200921_141709_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200921_141709_depth_crop_fill.png
372
+ TOFDC_split/Test/RGB/20200819_131119_rgb_crop.png,TOFDC_split/Test/GT/20200819_131119_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_131119_depth_crop_fill.png
373
+ TOFDC_split/Test/RGB/20200824_154613_rgb_crop.png,TOFDC_split/Test/GT/20200824_154613_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_154613_depth_crop_fill.png
374
+ TOFDC_split/Test/RGB/20200818_120653_rgb_crop.png,TOFDC_split/Test/GT/20200818_120653_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_120653_depth_crop_fill.png
375
+ TOFDC_split/Test/RGB/2020_09_12_21_11_12_504_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_21_11_12_504_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_21_11_12_504_rgb_depth_crop_fill.png
376
+ TOFDC_split/Test/RGB/20200721_193506_rgb_crop.png,TOFDC_split/Test/GT/20200721_193506_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200721_193506_depth_crop_fill.png
377
+ TOFDC_split/Test/RGB/20200927_141406_rgb_crop.png,TOFDC_split/Test/GT/20200927_141406_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_141406_depth_crop_fill.png
378
+ TOFDC_split/Test/RGB/20200819_152521_rgb_crop.png,TOFDC_split/Test/GT/20200819_152521_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_152521_depth_crop_fill.png
379
+ TOFDC_split/Test/RGB/2020_09_12_21_15_12_678_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_21_15_12_678_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_21_15_12_678_rgb_depth_crop_fill.png
380
+ TOFDC_split/Test/RGB/2020_09_09_16_31_44_521_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_16_31_44_521_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_16_31_44_521_rgb_depth_crop_fill.png
381
+ TOFDC_split/Test/RGB/2020_09_09_16_54_11_702_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_16_54_11_702_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_16_54_11_702_rgb_depth_crop_fill.png
382
+ TOFDC_split/Test/RGB/2020_09_26_10_14_23_962_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_26_10_14_23_962_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_26_10_14_23_962_rgb_depth_crop_fill.png
383
+ TOFDC_split/Test/RGB/20200927_195755_rgb_crop.png,TOFDC_split/Test/GT/20200927_195755_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_195755_depth_crop_fill.png
384
+ TOFDC_split/Test/RGB/2020_09_11_18_41_09_560_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_11_18_41_09_560_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_11_18_41_09_560_rgb_depth_crop_fill.png
385
+ TOFDC_split/Test/RGB/2020_09_12_16_03_39_173_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_16_03_39_173_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_16_03_39_173_rgb_depth_crop_fill.png
386
+ TOFDC_split/Test/RGB/2020_09_08_11_28_32_209_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_11_28_32_209_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_11_28_32_209_rgb_depth_crop_fill.png
387
+ TOFDC_split/Test/RGB/20200819_195656_rgb_crop.png,TOFDC_split/Test/GT/20200819_195656_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_195656_depth_crop_fill.png
388
+ TOFDC_split/Test/RGB/20200602_170101_rgb_crop.png,TOFDC_split/Test/GT/20200602_170101_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200602_170101_depth_crop_fill.png
389
+ TOFDC_split/Test/RGB/20200831_110112_rgb_crop.png,TOFDC_split/Test/GT/20200831_110112_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200831_110112_depth_crop_fill.png
390
+ TOFDC_split/Test/RGB/20200919_192838_rgb_crop.png,TOFDC_split/Test/GT/20200919_192838_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_192838_depth_crop_fill.png
391
+ TOFDC_split/Test/RGB/20200601_102116_rgb_crop.png,TOFDC_split/Test/GT/20200601_102116_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200601_102116_depth_crop_fill.png
392
+ TOFDC_split/Test/RGB/2020_09_08_11_14_07_770_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_11_14_07_770_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_11_14_07_770_rgb_depth_crop_fill.png
393
+ TOFDC_split/Test/RGB/20200927_153159_rgb_crop.png,TOFDC_split/Test/GT/20200927_153159_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_153159_depth_crop_fill.png
394
+ TOFDC_split/Test/RGB/20200927_191616_rgb_crop.png,TOFDC_split/Test/GT/20200927_191616_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_191616_depth_crop_fill.png
395
+ TOFDC_split/Test/RGB/20200716_152956_rgb_crop.png,TOFDC_split/Test/GT/20200716_152956_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200716_152956_depth_crop_fill.png
396
+ TOFDC_split/Test/RGB/20200824_175438_rgb_crop.png,TOFDC_split/Test/GT/20200824_175438_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_175438_depth_crop_fill.png
397
+ TOFDC_split/Test/RGB/20200603_085938_rgb_crop.png,TOFDC_split/Test/GT/20200603_085938_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200603_085938_depth_crop_fill.png
398
+ TOFDC_split/Test/RGB/20200927_195432_rgb_crop.png,TOFDC_split/Test/GT/20200927_195432_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_195432_depth_crop_fill.png
399
+ TOFDC_split/Test/RGB/20200919_194324_rgb_crop.png,TOFDC_split/Test/GT/20200919_194324_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_194324_depth_crop_fill.png
400
+ TOFDC_split/Test/RGB/2020_09_12_20_13_55_135_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_20_13_55_135_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_20_13_55_135_rgb_depth_crop_fill.png
401
+ TOFDC_split/Test/RGB/20200919_154328_rgb_crop.png,TOFDC_split/Test/GT/20200919_154328_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_154328_depth_crop_fill.png
402
+ TOFDC_split/Test/RGB/2020_09_26_11_04_10_535_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_26_11_04_10_535_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_26_11_04_10_535_rgb_depth_crop_fill.png
403
+ TOFDC_split/Test/RGB/20200928_152852_rgb_crop.png,TOFDC_split/Test/GT/20200928_152852_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200928_152852_depth_crop_fill.png
404
+ TOFDC_split/Test/RGB/20200824_170929_rgb_crop.png,TOFDC_split/Test/GT/20200824_170929_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_170929_depth_crop_fill.png
405
+ TOFDC_split/Test/RGB/20200927_154347_rgb_crop.png,TOFDC_split/Test/GT/20200927_154347_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_154347_depth_crop_fill.png
406
+ TOFDC_split/Test/RGB/20200818_150838_rgb_crop.png,TOFDC_split/Test/GT/20200818_150838_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_150838_depth_crop_fill.png
407
+ TOFDC_split/Test/RGB/20200819_204558_rgb_crop.png,TOFDC_split/Test/GT/20200819_204558_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_204558_depth_crop_fill.png
408
+ TOFDC_split/Test/RGB/20200818_152458_rgb_crop.png,TOFDC_split/Test/GT/20200818_152458_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_152458_depth_crop_fill.png
409
+ TOFDC_split/Test/RGB/20200927_155427_rgb_crop.png,TOFDC_split/Test/GT/20200927_155427_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_155427_depth_crop_fill.png
410
+ TOFDC_split/Test/RGB/2020_09_23_21_11_41_067_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_23_21_11_41_067_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_23_21_11_41_067_rgb_depth_crop_fill.png
411
+ TOFDC_split/Test/RGB/20200719_193804_rgb_crop.png,TOFDC_split/Test/GT/20200719_193804_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200719_193804_depth_crop_fill.png
412
+ TOFDC_split/Test/RGB/20200927_194253_rgb_crop.png,TOFDC_split/Test/GT/20200927_194253_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_194253_depth_crop_fill.png
413
+ TOFDC_split/Test/RGB/20200722_193954_rgb_crop.png,TOFDC_split/Test/GT/20200722_193954_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200722_193954_depth_crop_fill.png
414
+ TOFDC_split/Test/RGB/20200927_134952_rgb_crop.png,TOFDC_split/Test/GT/20200927_134952_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_134952_depth_crop_fill.png
415
+ TOFDC_split/Test/RGB/20200927_160719_rgb_crop.png,TOFDC_split/Test/GT/20200927_160719_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_160719_depth_crop_fill.png
416
+ TOFDC_split/Test/RGB/20200607_091021_rgb_crop.png,TOFDC_split/Test/GT/20200607_091021_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200607_091021_depth_crop_fill.png
417
+ TOFDC_split/Test/RGB/20200603_103139_rgb_crop.png,TOFDC_split/Test/GT/20200603_103139_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200603_103139_depth_crop_fill.png
418
+ TOFDC_split/Test/RGB/2020_09_13_19_58_14_264_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_13_19_58_14_264_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_13_19_58_14_264_rgb_depth_crop_fill.png
419
+ TOFDC_split/Test/RGB/20200605_145212_rgb_crop.png,TOFDC_split/Test/GT/20200605_145212_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200605_145212_depth_crop_fill.png
420
+ TOFDC_split/Test/RGB/20200601_203824_rgb_crop.png,TOFDC_split/Test/GT/20200601_203824_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200601_203824_depth_crop_fill.png
421
+ TOFDC_split/Test/RGB/2020_09_22_10_47_09_469_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_22_10_47_09_469_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_22_10_47_09_469_rgb_depth_crop_fill.png
422
+ TOFDC_split/Test/RGB/2020_09_11_18_37_38_345_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_11_18_37_38_345_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_11_18_37_38_345_rgb_depth_crop_fill.png
423
+ TOFDC_split/Test/RGB/2020_09_08_16_17_32_475_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_16_17_32_475_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_16_17_32_475_rgb_depth_crop_fill.png
424
+ TOFDC_split/Test/RGB/20200927_200510_rgb_crop.png,TOFDC_split/Test/GT/20200927_200510_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_200510_depth_crop_fill.png
425
+ TOFDC_split/Test/RGB/2020_09_08_18_12_17_961_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_18_12_17_961_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_18_12_17_961_rgb_depth_crop_fill.png
426
+ TOFDC_split/Test/RGB/20200602_174024_rgb_crop.png,TOFDC_split/Test/GT/20200602_174024_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200602_174024_depth_crop_fill.png
427
+ TOFDC_split/Test/RGB/20200927_202013_rgb_crop.png,TOFDC_split/Test/GT/20200927_202013_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_202013_depth_crop_fill.png
428
+ TOFDC_split/Test/RGB/20200819_160331_rgb_crop.png,TOFDC_split/Test/GT/20200819_160331_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_160331_depth_crop_fill.png
429
+ TOFDC_split/Test/RGB/20200919_102820_rgb_crop.png,TOFDC_split/Test/GT/20200919_102820_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_102820_depth_crop_fill.png
430
+ TOFDC_split/Test/RGB/20200819_202145_rgb_crop.png,TOFDC_split/Test/GT/20200819_202145_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_202145_depth_crop_fill.png
431
+ TOFDC_split/Test/RGB/20200723_150644_rgb_crop.png,TOFDC_split/Test/GT/20200723_150644_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200723_150644_depth_crop_fill.png
432
+ TOFDC_split/Test/RGB/20200818_153114_rgb_crop.png,TOFDC_split/Test/GT/20200818_153114_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_153114_depth_crop_fill.png
433
+ TOFDC_split/Test/RGB/20200920_171921_rgb_crop.png,TOFDC_split/Test/GT/20200920_171921_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_171921_depth_crop_fill.png
434
+ TOFDC_split/Test/RGB/20200824_113913_rgb_crop.png,TOFDC_split/Test/GT/20200824_113913_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_113913_depth_crop_fill.png
435
+ TOFDC_split/Test/RGB/20200818_104721_rgb_crop.png,TOFDC_split/Test/GT/20200818_104721_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_104721_depth_crop_fill.png
436
+ TOFDC_split/Test/RGB/20200819_103046_rgb_crop.png,TOFDC_split/Test/GT/20200819_103046_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_103046_depth_crop_fill.png
437
+ TOFDC_split/Test/RGB/2020_09_08_16_03_28_589_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_16_03_28_589_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_16_03_28_589_rgb_depth_crop_fill.png
438
+ TOFDC_split/Test/RGB/20200726_105128_rgb_crop.png,TOFDC_split/Test/GT/20200726_105128_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200726_105128_depth_crop_fill.png
439
+ TOFDC_split/Test/RGB/20200928_151105_rgb_crop.png,TOFDC_split/Test/GT/20200928_151105_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200928_151105_depth_crop_fill.png
440
+ TOFDC_split/Test/RGB/20200719_204036_rgb_crop.png,TOFDC_split/Test/GT/20200719_204036_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200719_204036_depth_crop_fill.png
441
+ TOFDC_split/Test/RGB/20200824_143415_rgb_crop.png,TOFDC_split/Test/GT/20200824_143415_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_143415_depth_crop_fill.png
442
+ TOFDC_split/Test/RGB/20200824_174420_rgb_crop.png,TOFDC_split/Test/GT/20200824_174420_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_174420_depth_crop_fill.png
443
+ TOFDC_split/Test/RGB/20200820_145156_rgb_crop.png,TOFDC_split/Test/GT/20200820_145156_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200820_145156_depth_crop_fill.png
444
+ TOFDC_split/Test/RGB/2020_09_11_18_54_26_868_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_11_18_54_26_868_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_11_18_54_26_868_rgb_depth_crop_fill.png
445
+ TOFDC_split/Test/RGB/20200818_153411_rgb_crop.png,TOFDC_split/Test/GT/20200818_153411_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_153411_depth_crop_fill.png
446
+ TOFDC_split/Test/RGB/20200919_104611_rgb_crop.png,TOFDC_split/Test/GT/20200919_104611_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_104611_depth_crop_fill.png
447
+ TOFDC_split/Test/RGB/20200602_091635_rgb_crop.png,TOFDC_split/Test/GT/20200602_091635_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200602_091635_depth_crop_fill.png
448
+ TOFDC_split/Test/RGB/20200927_201911_rgb_crop.png,TOFDC_split/Test/GT/20200927_201911_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_201911_depth_crop_fill.png
449
+ TOFDC_split/Test/RGB/20200921_141824_rgb_crop.png,TOFDC_split/Test/GT/20200921_141824_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200921_141824_depth_crop_fill.png
450
+ TOFDC_split/Test/RGB/20200927_143240_rgb_crop.png,TOFDC_split/Test/GT/20200927_143240_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_143240_depth_crop_fill.png
451
+ TOFDC_split/Test/RGB/20200819_161411_rgb_crop.png,TOFDC_split/Test/GT/20200819_161411_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_161411_depth_crop_fill.png
452
+ TOFDC_split/Test/RGB/20200725_185103_rgb_crop.png,TOFDC_split/Test/GT/20200725_185103_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200725_185103_depth_crop_fill.png
453
+ TOFDC_split/Test/RGB/20200718_160045_rgb_crop.png,TOFDC_split/Test/GT/20200718_160045_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200718_160045_depth_crop_fill.png
454
+ TOFDC_split/Test/RGB/20200920_155235_rgb_crop.png,TOFDC_split/Test/GT/20200920_155235_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_155235_depth_crop_fill.png
455
+ TOFDC_split/Test/RGB/20200927_201812_rgb_crop.png,TOFDC_split/Test/GT/20200927_201812_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_201812_depth_crop_fill.png
456
+ TOFDC_split/Test/RGB/20200920_165234_rgb_crop.png,TOFDC_split/Test/GT/20200920_165234_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_165234_depth_crop_fill.png
457
+ TOFDC_split/Test/RGB/20200820_110758_rgb_crop.png,TOFDC_split/Test/GT/20200820_110758_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200820_110758_depth_crop_fill.png
458
+ TOFDC_split/Test/RGB/20200927_141503_rgb_crop.png,TOFDC_split/Test/GT/20200927_141503_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_141503_depth_crop_fill.png
459
+ TOFDC_split/Test/RGB/2020_09_11_19_32_37_601_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_11_19_32_37_601_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_11_19_32_37_601_rgb_depth_crop_fill.png
460
+ TOFDC_split/Test/RGB/2020_09_14_23_27_00_069_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_14_23_27_00_069_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_14_23_27_00_069_rgb_depth_crop_fill.png
461
+ TOFDC_split/Test/RGB/2020_09_23_22_00_56_293_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_23_22_00_56_293_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_23_22_00_56_293_rgb_depth_crop_fill.png
462
+ TOFDC_split/Test/RGB/2020_09_13_19_39_13_040_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_13_19_39_13_040_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_13_19_39_13_040_rgb_depth_crop_fill.png
463
+ TOFDC_split/Test/RGB/20200927_194952_rgb_crop.png,TOFDC_split/Test/GT/20200927_194952_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_194952_depth_crop_fill.png
464
+ TOFDC_split/Test/RGB/2020_09_12_20_05_16_368_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_20_05_16_368_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_20_05_16_368_rgb_depth_crop_fill.png
465
+ TOFDC_split/Test/RGB/2020_09_10_22_56_54_090_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_10_22_56_54_090_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_10_22_56_54_090_rgb_depth_crop_fill.png
466
+ TOFDC_split/Test/RGB/20200718_193838_rgb_crop.png,TOFDC_split/Test/GT/20200718_193838_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200718_193838_depth_crop_fill.png
467
+ TOFDC_split/Test/RGB/20200927_193414_rgb_crop.png,TOFDC_split/Test/GT/20200927_193414_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_193414_depth_crop_fill.png
468
+ TOFDC_split/Test/RGB/20200819_203755_rgb_crop.png,TOFDC_split/Test/GT/20200819_203755_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_203755_depth_crop_fill.png
469
+ TOFDC_split/Test/RGB/20200927_191236_rgb_crop.png,TOFDC_split/Test/GT/20200927_191236_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_191236_depth_crop_fill.png
470
+ TOFDC_split/Test/RGB/20200819_103927_rgb_crop.png,TOFDC_split/Test/GT/20200819_103927_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_103927_depth_crop_fill.png
471
+ TOFDC_split/Test/RGB/20200718_201858_rgb_crop.png,TOFDC_split/Test/GT/20200718_201858_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200718_201858_depth_crop_fill.png
472
+ TOFDC_split/Test/RGB/20200919_104745_rgb_crop.png,TOFDC_split/Test/GT/20200919_104745_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_104745_depth_crop_fill.png
473
+ TOFDC_split/Test/RGB/20200818_113846_rgb_crop.png,TOFDC_split/Test/GT/20200818_113846_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_113846_depth_crop_fill.png
474
+ TOFDC_split/Test/RGB/20200928_150528_rgb_crop.png,TOFDC_split/Test/GT/20200928_150528_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200928_150528_depth_crop_fill.png
475
+ TOFDC_split/Test/RGB/20200603_195754_rgb_crop.png,TOFDC_split/Test/GT/20200603_195754_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200603_195754_depth_crop_fill.png
476
+ TOFDC_split/Test/RGB/20200607_092259_rgb_crop.png,TOFDC_split/Test/GT/20200607_092259_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200607_092259_depth_crop_fill.png
477
+ TOFDC_split/Test/RGB/20200725_154655_rgb_crop.png,TOFDC_split/Test/GT/20200725_154655_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200725_154655_depth_crop_fill.png
478
+ TOFDC_split/Test/RGB/20200919_155325_rgb_crop.png,TOFDC_split/Test/GT/20200919_155325_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_155325_depth_crop_fill.png
479
+ TOFDC_split/Test/RGB/20200919_195123_rgb_crop.png,TOFDC_split/Test/GT/20200919_195123_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_195123_depth_crop_fill.png
480
+ TOFDC_split/Test/RGB/2020_09_21_20_03_59_031_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_21_20_03_59_031_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_21_20_03_59_031_rgb_depth_crop_fill.png
481
+ TOFDC_split/Test/RGB/2020_09_10_22_34_50_072_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_10_22_34_50_072_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_10_22_34_50_072_rgb_depth_crop_fill.png
482
+ TOFDC_split/Test/RGB/2020_09_12_16_23_05_697_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_16_23_05_697_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_16_23_05_697_rgb_depth_crop_fill.png
483
+ TOFDC_split/Test/RGB/2020_09_26_10_44_05_488_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_26_10_44_05_488_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_26_10_44_05_488_rgb_depth_crop_fill.png
484
+ TOFDC_split/Test/RGB/2020_09_14_19_42_05_367_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_14_19_42_05_367_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_14_19_42_05_367_rgb_depth_crop_fill.png
485
+ TOFDC_split/Test/RGB/2020_09_14_19_34_21_962_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_14_19_34_21_962_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_14_19_34_21_962_rgb_depth_crop_fill.png
486
+ TOFDC_split/Test/RGB/2020_09_09_14_49_11_337_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_14_49_11_337_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_14_49_11_337_rgb_depth_crop_fill.png
487
+ TOFDC_split/Test/RGB/2020_09_09_21_00_29_568_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_21_00_29_568_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_21_00_29_568_rgb_depth_crop_fill.png
488
+ TOFDC_split/Test/RGB/2020_09_26_14_26_08_445_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_26_14_26_08_445_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_26_14_26_08_445_rgb_depth_crop_fill.png
489
+ TOFDC_split/Test/RGB/2020_09_09_20_58_26_019_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_20_58_26_019_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_20_58_26_019_rgb_depth_crop_fill.png
490
+ TOFDC_split/Test/RGB/2020_09_12_11_54_38_467_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_11_54_38_467_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_11_54_38_467_rgb_depth_crop_fill.png
491
+ TOFDC_split/Test/RGB/20200603_110751_rgb_crop.png,TOFDC_split/Test/GT/20200603_110751_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200603_110751_depth_crop_fill.png
492
+ TOFDC_split/Test/RGB/20200601_190016_rgb_crop.png,TOFDC_split/Test/GT/20200601_190016_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200601_190016_depth_crop_fill.png
493
+ TOFDC_split/Test/RGB/20200919_112001_rgb_crop.png,TOFDC_split/Test/GT/20200919_112001_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_112001_depth_crop_fill.png
494
+ TOFDC_split/Test/RGB/20200602_100348_rgb_crop.png,TOFDC_split/Test/GT/20200602_100348_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200602_100348_depth_crop_fill.png
495
+ TOFDC_split/Test/RGB/20200604_164300_rgb_crop.png,TOFDC_split/Test/GT/20200604_164300_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200604_164300_depth_crop_fill.png
496
+ TOFDC_split/Test/RGB/20200927_200211_rgb_crop.png,TOFDC_split/Test/GT/20200927_200211_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_200211_depth_crop_fill.png
497
+ TOFDC_split/Test/RGB/2020_09_09_21_25_08_609_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_21_25_08_609_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_21_25_08_609_rgb_depth_crop_fill.png
498
+ TOFDC_split/Test/RGB/2020_09_22_10_58_03_795_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_22_10_58_03_795_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_22_10_58_03_795_rgb_depth_crop_fill.png
499
+ TOFDC_split/Test/RGB/2020_09_11_17_27_16_826_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_11_17_27_16_826_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_11_17_27_16_826_rgb_depth_crop_fill.png
500
+ TOFDC_split/Test/RGB/20200927_155151_rgb_crop.png,TOFDC_split/Test/GT/20200927_155151_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_155151_depth_crop_fill.png
501
+ TOFDC_split/Test/RGB/20200818_213742_rgb_crop.png,TOFDC_split/Test/GT/20200818_213742_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_213742_depth_crop_fill.png
502
+ TOFDC_split/Test/RGB/20200716_095332_rgb_crop.png,TOFDC_split/Test/GT/20200716_095332_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200716_095332_depth_crop_fill.png
503
+ TOFDC_split/Test/RGB/20200820_165419_rgb_crop.png,TOFDC_split/Test/GT/20200820_165419_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200820_165419_depth_crop_fill.png
504
+ TOFDC_split/Test/RGB/20200818_152524_rgb_crop.png,TOFDC_split/Test/GT/20200818_152524_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_152524_depth_crop_fill.png
505
+ TOFDC_split/Test/RGB/20200721_144926_rgb_crop.png,TOFDC_split/Test/GT/20200721_144926_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200721_144926_depth_crop_fill.png
506
+ TOFDC_split/Test/RGB/2020_09_12_20_42_13_700_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_20_42_13_700_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_20_42_13_700_rgb_depth_crop_fill.png
507
+ TOFDC_split/Test/RGB/20200927_204750_rgb_crop.png,TOFDC_split/Test/GT/20200927_204750_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_204750_depth_crop_fill.png
508
+ TOFDC_split/Test/RGB/20200928_153521_rgb_crop.png,TOFDC_split/Test/GT/20200928_153521_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200928_153521_depth_crop_fill.png
509
+ TOFDC_split/Test/RGB/20200727_170228_rgb_crop.png,TOFDC_split/Test/GT/20200727_170228_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200727_170228_depth_crop_fill.png
510
+ TOFDC_split/Test/RGB/20200927_192222_rgb_crop.png,TOFDC_split/Test/GT/20200927_192222_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_192222_depth_crop_fill.png
511
+ TOFDC_split/Test/RGB/2020_09_22_10_15_20_018_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_22_10_15_20_018_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_22_10_15_20_018_rgb_depth_crop_fill.png
512
+ TOFDC_split/Test/RGB/20200727_173216_rgb_crop.png,TOFDC_split/Test/GT/20200727_173216_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200727_173216_depth_crop_fill.png
513
+ TOFDC_split/Test/RGB/20200725_154516_rgb_crop.png,TOFDC_split/Test/GT/20200725_154516_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200725_154516_depth_crop_fill.png
514
+ TOFDC_split/Test/RGB/2020_09_12_20_31_01_219_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_20_31_01_219_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_20_31_01_219_rgb_depth_crop_fill.png
515
+ TOFDC_split/Test/RGB/20200819_104229_rgb_crop.png,TOFDC_split/Test/GT/20200819_104229_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_104229_depth_crop_fill.png
516
+ TOFDC_split/Test/RGB/20200920_170926_rgb_crop.png,TOFDC_split/Test/GT/20200920_170926_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_170926_depth_crop_fill.png
517
+ TOFDC_split/Test/RGB/20200903_160917_rgb_crop.png,TOFDC_split/Test/GT/20200903_160917_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200903_160917_depth_crop_fill.png
518
+ TOFDC_split/Test/RGB/20200819_201906_rgb_crop.png,TOFDC_split/Test/GT/20200819_201906_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200819_201906_depth_crop_fill.png
519
+ TOFDC_split/Test/RGB/20200717_193107_rgb_crop.png,TOFDC_split/Test/GT/20200717_193107_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200717_193107_depth_crop_fill.png
520
+ TOFDC_split/Test/RGB/20200719_094105_rgb_crop.png,TOFDC_split/Test/GT/20200719_094105_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200719_094105_depth_crop_fill.png
521
+ TOFDC_split/Test/RGB/20200927_202626_rgb_crop.png,TOFDC_split/Test/GT/20200927_202626_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_202626_depth_crop_fill.png
522
+ TOFDC_split/Test/RGB/2020_09_08_11_02_59_336_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_11_02_59_336_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_11_02_59_336_rgb_depth_crop_fill.png
523
+ TOFDC_split/Test/RGB/2020_09_08_17_44_15_116_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_17_44_15_116_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_17_44_15_116_rgb_depth_crop_fill.png
524
+ TOFDC_split/Test/RGB/2020_09_12_20_47_27_409_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_20_47_27_409_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_20_47_27_409_rgb_depth_crop_fill.png
525
+ TOFDC_split/Test/RGB/2020_09_11_21_56_01_588_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_11_21_56_01_588_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_11_21_56_01_588_rgb_depth_crop_fill.png
526
+ TOFDC_split/Test/RGB/2020_09_10_15_34_37_293_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_10_15_34_37_293_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_10_15_34_37_293_rgb_depth_crop_fill.png
527
+ TOFDC_split/Test/RGB/2020_09_26_15_03_55_533_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_26_15_03_55_533_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_26_15_03_55_533_rgb_depth_crop_fill.png
528
+ TOFDC_split/Test/RGB/2020_09_09_21_18_08_414_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_21_18_08_414_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_21_18_08_414_rgb_depth_crop_fill.png
529
+ TOFDC_split/Test/RGB/2020_09_12_15_38_42_237_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_15_38_42_237_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_15_38_42_237_rgb_depth_crop_fill.png
530
+ TOFDC_split/Test/RGB/2020_09_12_15_17_16_617_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_15_17_16_617_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_15_17_16_617_rgb_depth_crop_fill.png
531
+ TOFDC_split/Test/RGB/20200818_215847_rgb_crop.png,TOFDC_split/Test/GT/20200818_215847_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_215847_depth_crop_fill.png
532
+ TOFDC_split/Test/RGB/20200818_145458_rgb_crop.png,TOFDC_split/Test/GT/20200818_145458_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_145458_depth_crop_fill.png
533
+ TOFDC_split/Test/RGB/2020_09_12_11_06_14_851_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_12_11_06_14_851_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_12_11_06_14_851_rgb_depth_crop_fill.png
534
+ TOFDC_split/Test/RGB/2020_09_13_19_13_22_503_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_13_19_13_22_503_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_13_19_13_22_503_rgb_depth_crop_fill.png
535
+ TOFDC_split/Test/RGB/20200927_154637_rgb_crop.png,TOFDC_split/Test/GT/20200927_154637_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_154637_depth_crop_fill.png
536
+ TOFDC_split/Test/RGB/20200919_191743_rgb_crop.png,TOFDC_split/Test/GT/20200919_191743_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200919_191743_depth_crop_fill.png
537
+ TOFDC_split/Test/RGB/20200824_165648_rgb_crop.png,TOFDC_split/Test/GT/20200824_165648_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200824_165648_depth_crop_fill.png
538
+ TOFDC_split/Test/RGB/20200927_195308_rgb_crop.png,TOFDC_split/Test/GT/20200927_195308_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200927_195308_depth_crop_fill.png
539
+ TOFDC_split/Test/RGB/2020_09_09_11_00_54_989_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_11_00_54_989_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_11_00_54_989_rgb_depth_crop_fill.png
540
+ TOFDC_split/Test/RGB/2020_09_21_20_29_27_975_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_21_20_29_27_975_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_21_20_29_27_975_rgb_depth_crop_fill.png
541
+ TOFDC_split/Test/RGB/20200723_105307_rgb_crop.png,TOFDC_split/Test/GT/20200723_105307_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200723_105307_depth_crop_fill.png
542
+ TOFDC_split/Test/RGB/20200720_154423_rgb_crop.png,TOFDC_split/Test/GT/20200720_154423_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200720_154423_depth_crop_fill.png
543
+ TOFDC_split/Test/RGB/2020_09_08_17_22_53_558_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_17_22_53_558_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_17_22_53_558_rgb_depth_crop_fill.png
544
+ TOFDC_split/Test/RGB/2020_09_22_16_18_39_356_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_22_16_18_39_356_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_22_16_18_39_356_rgb_depth_crop_fill.png
545
+ TOFDC_split/Test/RGB/20200818_203330_rgb_crop.png,TOFDC_split/Test/GT/20200818_203330_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_203330_depth_crop_fill.png
546
+ TOFDC_split/Test/RGB/20200717_105043_rgb_crop.png,TOFDC_split/Test/GT/20200717_105043_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200717_105043_depth_crop_fill.png
547
+ TOFDC_split/Test/RGB/20200920_112805_rgb_crop.png,TOFDC_split/Test/GT/20200920_112805_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200920_112805_depth_crop_fill.png
548
+ TOFDC_split/Test/RGB/20200723_110356_rgb_crop.png,TOFDC_split/Test/GT/20200723_110356_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200723_110356_depth_crop_fill.png
549
+ TOFDC_split/Test/RGB/20200601_171237_rgb_crop.png,TOFDC_split/Test/GT/20200601_171237_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200601_171237_depth_crop_fill.png
550
+ TOFDC_split/Test/RGB/2020_09_09_15_15_11_689_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_09_15_15_11_689_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_09_15_15_11_689_rgb_depth_crop_fill.png
551
+ TOFDC_split/Test/RGB/20200717_161032_rgb_crop.png,TOFDC_split/Test/GT/20200717_161032_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200717_161032_depth_crop_fill.png
552
+ TOFDC_split/Test/RGB/20200725_163031_rgb_crop.png,TOFDC_split/Test/GT/20200725_163031_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200725_163031_depth_crop_fill.png
553
+ TOFDC_split/Test/RGB/20200718_100632_rgb_crop.png,TOFDC_split/Test/GT/20200718_100632_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200718_100632_depth_crop_fill.png
554
+ TOFDC_split/Test/RGB/20200721_152622_rgb_crop.png,TOFDC_split/Test/GT/20200721_152622_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200721_152622_depth_crop_fill.png
555
+ TOFDC_split/Test/RGB/2020_09_08_16_24_37_249_rgb_rgb_crop.png,TOFDC_split/Test/GT/2020_09_08_16_24_37_249_rgb_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/2020_09_08_16_24_37_249_rgb_depth_crop_fill.png
556
+ TOFDC_split/Test/RGB/20200818_203600_rgb_crop.png,TOFDC_split/Test/GT/20200818_203600_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200818_203600_depth_crop_fill.png
557
+ TOFDC_split/Test/RGB/20200721_164403_rgb_crop.png,TOFDC_split/Test/GT/20200721_164403_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200721_164403_depth_crop_fill.png
558
+ TOFDC_split/Test/RGB/20200716_201254_rgb_crop.png,TOFDC_split/Test/GT/20200716_201254_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200716_201254_depth_crop_fill.png
559
+ TOFDC_split/Test/RGB/20200725_164021_rgb_crop.png,TOFDC_split/Test/GT/20200725_164021_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200725_164021_depth_crop_fill.png
560
+ TOFDC_split/Test/RGB/20200603_202348_rgb_crop.png,TOFDC_split/Test/GT/20200603_202348_tgv_gt_crop.png,TOFDC_split/Test/LR_Filled/20200603_202348_depth_crop_fill.png
data/TOFDSR_Train.txt ADDED
The diff for this file is too large to render. See raw diff
 
data/__pycache__/nyu_dataloader.cpython-311.pyc ADDED
Binary file (3.04 kB). View file
 
data/__pycache__/rgbdd_dataloader.cpython-311.pyc ADDED
Binary file (7.64 kB). View file
 
data/nyu_dataloader.py ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from torch.utils.data import Dataset
2
+ from PIL import Image
3
+ import numpy as np
4
+
5
+
6
+ class NYU_v2_datset(Dataset):
7
+ """NYUDataset."""
8
+
9
+ def __init__(self, root_dir, scale=8, train=True, transform=None):
10
+ """
11
+ Args:
12
+ root_dir (string): Directory with all the images.
13
+ scale (float): dataset scale
14
+ train (bool): train or test
15
+ transform (callable, optional): Optional transform to be applied on a sample.
16
+
17
+ """
18
+ self.root_dir = root_dir
19
+ self.transform = transform
20
+ self.scale = scale
21
+ self.train = train
22
+
23
+ if train:
24
+ self.depths = np.load('%s/train_depth_split.npy' % root_dir)
25
+ self.images = np.load('%s/train_images_split.npy' % root_dir)
26
+ else:
27
+ self.depths = np.load('%s/test_depth.npy' % root_dir)
28
+ self.images = np.load('%s/test_images_v2.npy' % root_dir)
29
+
30
+ def __len__(self):
31
+ return self.depths.shape[0]
32
+
33
+ def __getitem__(self, idx):
34
+ depth = self.depths[idx]
35
+ image = self.images[idx]
36
+ h, w = depth.shape[:2]
37
+ s = self.scale
38
+ lr = np.array(Image.fromarray(depth.squeeze()).resize((w // s, h // s), Image.BICUBIC).resize((w, h), Image.BICUBIC))
39
+
40
+ if self.transform:
41
+ image = self.transform(image).float()
42
+ depth = self.transform(depth).float()
43
+ lr = self.transform(np.expand_dims(lr, 2)).float()
44
+
45
+ sample = {'guidance': image, 'lr': lr, 'gt': depth}
46
+
47
+ return sample
data/rgbdd_dataloader.py ADDED
@@ -0,0 +1,118 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import numpy as np
2
+ import os
3
+
4
+ from torch.utils.data import Dataset
5
+ from PIL import Image
6
+ from scipy.ndimage import gaussian_filter
7
+
8
+
9
+ class RGBDD_Dataset(Dataset):
10
+ """RGB-D-D Dataset."""
11
+
12
+ def __init__(self, root_dir="./dataset/RGB-D-D/", scale=4, downsample='real', train=True,
13
+ transform=None, isNoisy=False, blur_sigma=1.2):
14
+
15
+ self.root_dir = root_dir
16
+ self.transform = transform
17
+ self.scale = scale
18
+ self.downsample = downsample
19
+ self.train = train
20
+ self.isNoisy = isNoisy
21
+ self.blur_sigma = blur_sigma
22
+
23
+ types = ['models', 'plants', 'portraits']
24
+
25
+ if train:
26
+ if self.downsample == 'real':
27
+ self.GTs = []
28
+ self.LRs = []
29
+ self.RGBs = []
30
+ for type in types:
31
+ list_dir = os.listdir('%s/%s/%s_train'% (root_dir, type, type))
32
+ for n in list_dir:
33
+ self.RGBs.append('%s/%s/%s_train/%s/%s_RGB.jpg' % (root_dir, type, type, n, n))
34
+ self.GTs.append('%s/%s/%s_train/%s/%s_HR_gt.png' % (root_dir, type, type, n, n))
35
+ self.LRs.append('%s/%s/%s_train/%s/%s_LR_fill_depth.png' % (root_dir, type, type, n, n))
36
+ else:
37
+ self.GTs = []
38
+ self.RGBs = []
39
+ for type in types:
40
+ list_dir = os.listdir('%s/%s/%s_train'% (root_dir, type, type))
41
+ for n in list_dir:
42
+ self.RGBs.append('%s/%s/%s_train/%s/%s_RGB.jpg' % (root_dir, type, type, n, n))
43
+ self.GTs.append('%s/%s/%s_train/%s/%s_HR_gt.png' % (root_dir, type, type, n, n))
44
+
45
+ else:
46
+ if self.downsample == 'real':
47
+ self.GTs = []
48
+ self.LRs = []
49
+ self.RGBs = []
50
+ for type in types:
51
+ list_dir = os.listdir('%s/%s/%s_test'% (root_dir, type, type))
52
+ for n in list_dir:
53
+ self.RGBs.append('%s/%s/%s_test/%s/%s_RGB.jpg' % (root_dir, type, type, n, n))
54
+ self.GTs.append('%s/%s/%s_test/%s/%s_HR_gt.png' % (root_dir, type, type, n, n))
55
+ self.LRs.append('%s/%s/%s_test/%s/%s_LR_fill_depth.png' % (root_dir, type, type, n, n))
56
+ else:
57
+ self.GTs = []
58
+ self.RGBs = []
59
+ for type in types:
60
+ list_dir = os.listdir('%s/%s/%s_test'% (root_dir, type, type))
61
+ for n in list_dir:
62
+ self.RGBs.append('%s/%s/%s_test/%s/%s_RGB.jpg' % (root_dir, type, type, n, n))
63
+ self.GTs.append('%s/%s/%s_test/%s/%s_HR_gt.png' % (root_dir, type, type, n, n))
64
+
65
+ def __len__(self):
66
+ return len(self.GTs)
67
+
68
+ def __getitem__(self, idx):
69
+ if self.downsample == 'real':
70
+ image = np.array(Image.open(self.RGBs[idx]).convert("RGB")).astype(np.float32)
71
+ name = self.RGBs[idx][-22:-8]
72
+ gt = np.array(Image.open(self.GTs[idx])).astype(np.float32)
73
+ h, w = gt.shape
74
+ lr = np.array(Image.open(self.LRs[idx]).resize((w, h), Image.BICUBIC)).astype(np.float32)
75
+ else:
76
+ image = Image.open(self.RGBs[idx]).convert("RGB")
77
+ name = self.RGBs[idx][-22:-8]
78
+ image = np.array(image).astype(np.float32)
79
+ gt = Image.open(self.GTs[idx])
80
+ w, h = gt.size
81
+ s = self.scale
82
+ lr = np.array(gt.resize((w // s, h // s), Image.BICUBIC).resize((w, h), Image.BICUBIC)).astype(np.float32)
83
+ gt = np.array(gt).astype(np.float32)
84
+
85
+ # normalization
86
+ if self.train:
87
+ max_out = 5000.0
88
+ min_out = 0.0
89
+ lr = (lr - min_out) / (max_out - min_out)
90
+ gt = (gt-min_out)/(max_out-min_out)
91
+ else:
92
+ max_out = 5000.0
93
+ min_out = 0.0
94
+ lr = (lr - min_out) / (max_out - min_out)
95
+
96
+ maxx = np.max(image)
97
+ minn = np.min(image)
98
+ image = (image - minn) / (maxx - minn)
99
+
100
+ lr_minn = np.min(lr)
101
+ lr_maxx = np.max(lr)
102
+
103
+ if not self.train:
104
+ np.random.seed(42)
105
+
106
+ if self.isNoisy:
107
+ lr = gaussian_filter(lr, sigma=self.blur_sigma)
108
+
109
+ gaussian_noise = np.random.normal(0, 0.07, lr.shape)
110
+ lr = lr + gaussian_noise
111
+ lr = np.clip(lr, lr_minn, lr_maxx)
112
+
113
+ image = self.transform(image).float()
114
+ gt = self.transform(np.expand_dims(gt, 2)).float()
115
+ lr = self.transform(np.expand_dims(lr, 2)).float()
116
+ sample = {'guidance': image, 'lr': lr, 'gt': gt, 'max': max_out, 'min': min_out, 'name':name}
117
+
118
+ return sample
data/tofdc_dataloader.py ADDED
@@ -0,0 +1,90 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import numpy as np
2
+ import os
3
+
4
+ from torch.utils.data import Dataset, DataLoader
5
+ from PIL import Image
6
+ from scipy.ndimage import gaussian_filter
7
+
8
+
9
+ class TOFDSR_Dataset(Dataset):
10
+
11
+ def __init__(self, root_dir="./dataset/", scale=4, downsample='real', train=True, txt_file='./TOFDSR_Train.txt' ,
12
+ transform=None, isNoisy=False, blur_sigma=1.2):
13
+
14
+ self.root_dir = root_dir
15
+ self.transform = transform
16
+ self.scale = scale
17
+ self.downsample = downsample
18
+ self.train = train
19
+ self.isNoisy = isNoisy
20
+ self.blur_sigma = blur_sigma
21
+ self.image_list = txt_file
22
+
23
+ with open(self.image_list, 'r') as f:
24
+ self.filename = f.readlines()
25
+
26
+ def __len__(self):
27
+ return len(self.filename)
28
+
29
+ def __getitem__(self, idx):
30
+
31
+ sample_path = self.filename[idx].strip('\n')
32
+ sample_path_ = sample_path.split(',')
33
+ rgb_path = sample_path_[0]
34
+ gt_path = sample_path_[1]
35
+ lr_path = sample_path_[2]
36
+ name = gt_path[20:-4]
37
+
38
+ rgb_path = os.path.join(self.root_dir, rgb_path)
39
+ gt_path = os.path.join(self.root_dir, gt_path)
40
+ lr_path = os.path.join(self.root_dir, lr_path)
41
+
42
+ if self.downsample == 'real':
43
+ image = np.array(Image.open(rgb_path).convert("RGB")).astype(np.float32)
44
+ gt = np.array(Image.open(gt_path)).astype(np.float32)
45
+ h, w = gt.shape
46
+ lr = np.array(Image.open(lr_path).resize((w, h), Image.BICUBIC)).astype(np.float32)
47
+
48
+ else:
49
+ image = np.array(Image.open(rgb_path).convert("RGB")).astype(np.float32)
50
+ gt = Image.open(gt_path)
51
+ w, h = gt.size
52
+ lr = np.array(gt.resize((w, h), Image.BICUBIC)).astype(np.float32)
53
+ gt = np.array(gt).astype(np.float32)
54
+
55
+ image_max = np.max(image)
56
+ image_min = np.min(image)
57
+ image = (image - image_min) / (image_max - image_min)
58
+
59
+ # normalization
60
+ if self.train:
61
+ max_out = 5000.0
62
+ min_out = 0.0
63
+ lr = (lr - min_out) / (max_out - min_out)
64
+ gt = (gt-min_out)/(max_out-min_out)
65
+ else:
66
+ max_out = 5000.0
67
+ min_out = 0.0
68
+ lr = (lr - min_out) / (max_out - min_out)
69
+
70
+ lr_minn = np.min(lr)
71
+ lr_maxx = np.max(lr)
72
+
73
+ if not self.train:
74
+ np.random.seed(42)
75
+
76
+ if self.isNoisy:
77
+ lr = gaussian_filter(lr, sigma=self.blur_sigma)
78
+
79
+ gaussian_noise = np.random.normal(0, 0.07, lr.shape)
80
+ lr = lr + gaussian_noise
81
+ lr = np.clip(lr, lr_minn, lr_maxx)
82
+
83
+ if self.transform:
84
+ image = self.transform(image).float()
85
+ gt = self.transform(np.expand_dims(gt, 2)).float()
86
+ lr = self.transform(np.expand_dims(lr, 2)).float()
87
+
88
+ sample = {'guidance': image, 'lr': lr, 'gt': gt, 'max': max_out, 'min': min_out,'name': name}
89
+
90
+ return sample
examples/RGB-D-D/20200518160957_LR_fill_depth.png ADDED
examples/RGB-D-D/20200518160957_RGB.jpg ADDED
examples/TOFDSR/2020_09_08_13_59_59_435_rgb_depth_crop_fill.png ADDED
examples/TOFDSR/2020_09_08_13_59_59_435_rgb_rgb_crop.png ADDED

Git LFS Details

  • SHA256: caac13848f252ad22044e7a4789595f150f939a4c94e7595b2dce7ed1d8330a0
  • Pointer size: 131 Bytes
  • Size of remote file: 236 kB
net/CR.py ADDED
@@ -0,0 +1,63 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import torch.nn as nn
2
+ import torch
3
+ from torchvision import models
4
+
5
+ class Vgg19(torch.nn.Module):
6
+ def __init__(self, requires_grad=False):
7
+ super(Vgg19, self).__init__()
8
+ vgg_pretrained_features = models.vgg19(pretrained=True).features
9
+ self.slice1 = torch.nn.Sequential()
10
+ self.slice2 = torch.nn.Sequential()
11
+ self.slice3 = torch.nn.Sequential()
12
+ self.slice4 = torch.nn.Sequential()
13
+ self.slice5 = torch.nn.Sequential()
14
+ for x in range(2):
15
+ self.slice1.add_module(str(x), vgg_pretrained_features[x])
16
+ for x in range(2, 7):
17
+ self.slice2.add_module(str(x), vgg_pretrained_features[x])
18
+ for x in range(7, 12):
19
+ self.slice3.add_module(str(x), vgg_pretrained_features[x])
20
+ for x in range(12, 21):
21
+ self.slice4.add_module(str(x), vgg_pretrained_features[x])
22
+ for x in range(21, 30):
23
+ self.slice5.add_module(str(x), vgg_pretrained_features[x])
24
+ if not requires_grad:
25
+ for param in self.parameters():
26
+ param.requires_grad = False
27
+
28
+ def forward(self, X):
29
+ h_relu1 = self.slice1(X)
30
+ h_relu2 = self.slice2(h_relu1)
31
+ h_relu3 = self.slice3(h_relu2)
32
+ h_relu4 = self.slice4(h_relu3)
33
+ h_relu5 = self.slice5(h_relu4)
34
+ return [h_relu1, h_relu2, h_relu3, h_relu4, h_relu5]
35
+
36
+ class ContrastLoss(nn.Module):
37
+ def __init__(self, ablation=False):
38
+
39
+ super(ContrastLoss, self).__init__()
40
+ self.vgg = Vgg19().cuda()
41
+ self.l1 = nn.L1Loss()
42
+ self.weights = [1.0/32, 1.0/16, 1.0/8, 1.0/4, 1.0]
43
+ self.ab = ablation
44
+
45
+ def forward(self, a, p, n):
46
+
47
+ a_re = a.repeat(1, 3, 1, 1)
48
+ p_re = p.repeat(1, 3, 1, 1)
49
+ n_re = n.repeat(1, 3, 1, 1)
50
+ a_vgg, p_vgg, n_vgg = self.vgg(a_re), self.vgg(p_re), self.vgg(n_re)
51
+ loss = 0
52
+
53
+ d_ap, d_an = 0, 0
54
+ for i in range(len(a_vgg)):
55
+ d_ap = self.l1(a_vgg[i], p_vgg[i].detach())
56
+ if not self.ab:
57
+ d_an = self.l1(a_vgg[i], n_vgg[i].detach())
58
+ contrastive = d_ap / (d_an + 1e-7)
59
+ else:
60
+ contrastive = d_ap
61
+
62
+ loss += self.weights[i] * contrastive
63
+ return loss
net/__pycache__/CR.cpython-311.pyc ADDED
Binary file (5.26 kB). View file
 
net/__pycache__/deform_conv.cpython-311.pyc ADDED
Binary file (5.2 kB). View file
 
net/__pycache__/dornet.cpython-311.pyc ADDED
Binary file (38.3 kB). View file
 
net/__pycache__/dornet_ddp.cpython-311.pyc ADDED
Binary file (39.2 kB). View file
 
net/deform_conv.py ADDED
@@ -0,0 +1,75 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import math
2
+
3
+ import torch
4
+ import torch.nn as nn
5
+ from torch.nn.modules.utils import _pair
6
+ from mmcv.ops import modulated_deform_conv2d
7
+
8
+ class DCN_layer_rgb(nn.Module):
9
+ def __init__(self, in_channels, out_channels, kernel_size, stride=1, padding=0, dilation=1,
10
+ groups=1, deformable_groups=1, bias=True, extra_offset_mask=True):
11
+ super(DCN_layer_rgb, self).__init__()
12
+ self.in_channels = in_channels
13
+ self.out_channels = out_channels
14
+ self.kernel_size = _pair(kernel_size)
15
+ self.stride = stride
16
+ self.padding = padding
17
+ self.dilation = dilation
18
+ self.groups = groups
19
+ self.deformable_groups = deformable_groups
20
+ self.with_bias = bias
21
+
22
+ self.weight = nn.Parameter(
23
+ torch.Tensor(out_channels, in_channels, *self.kernel_size))
24
+
25
+ self.extra_offset_mask = extra_offset_mask
26
+ self.conv_offset_mask = nn.Conv2d(
27
+ self.in_channels,
28
+ self.deformable_groups * 3 * self.kernel_size[0] * self.kernel_size[1],
29
+ kernel_size=self.kernel_size, stride=_pair(self.stride), padding=_pair(self.padding),
30
+ bias=True
31
+ )
32
+
33
+ self.c1 = nn.Conv2d(in_channels*4, out_channels, 1, 1, 0, bias=False)
34
+ self.c2 = nn.Conv2d(out_channels, out_channels, 1, 1, 0, bias=False)
35
+
36
+ if bias:
37
+ self.bias = nn.Parameter(torch.Tensor(out_channels))
38
+ else:
39
+ self.register_parameter('bias', None)
40
+
41
+ self.init_offset()
42
+ self.reset_parameters()
43
+
44
+ def reset_parameters(self):
45
+ n = self.in_channels
46
+ for k in self.kernel_size:
47
+ n *= k
48
+ stdv = 1. / math.sqrt(n)
49
+ self.weight.data.uniform_(-stdv, stdv)
50
+ if self.bias is not None:
51
+ self.bias.data.zero_()
52
+
53
+ def init_offset(self):
54
+ self.conv_offset_mask.weight.data.zero_()
55
+ self.conv_offset_mask.bias.data.zero_()
56
+
57
+ def forward(self, input_feat, inter, fea):
58
+ b, c, h, w = input_feat.shape
59
+ fea = self.c1(fea).unsqueeze(1)
60
+ weight = self.weight.unsqueeze(0) * fea
61
+ weight = weight.view(b * self.out_channels, self.in_channels, self.kernel_size[0],
62
+ self.kernel_size[1]).contiguous()
63
+ input_feat = input_feat.view(1, b * self.in_channels, h, w)
64
+
65
+ out = self.conv_offset_mask(inter)
66
+ o1, o2, mask = torch.chunk(out, 3, dim=1)
67
+ offset = torch.cat((o1, o2), dim=1)
68
+ mask = torch.sigmoid(mask)
69
+
70
+ out = modulated_deform_conv2d(input_feat.contiguous(), offset, mask, weight, self.bias, self.stride,
71
+ self.padding, self.dilation, b, b)
72
+ _, _, height, width = out.shape
73
+ out = out.view(b, self.out_channels, height, width).contiguous()
74
+ out2 = self.c2(out)
75
+ return out2
net/dornet.py ADDED
@@ -0,0 +1,586 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import torch
2
+ import torch.nn as nn
3
+ from .deform_conv import DCN_layer_rgb
4
+ import torch.nn.functional as F
5
+ import math
6
+
7
+ from torch.distributions.normal import Normal
8
+ import numpy as np
9
+
10
+
11
+ class SparseDispatcher(object):
12
+ """Helper for implementing a mixture of experts.
13
+ The purpose of this class is to create input minibatches for the
14
+ experts and to combine the results of the experts to form a unified
15
+ output tensor.
16
+ There are two functions:
17
+ dispatch - take an input Tensor and create input Tensors for each expert.
18
+ combine - take output Tensors from each expert and form a combined output
19
+ Tensor. Outputs from different experts for the same batch element are
20
+ summed together, weighted by the provided "gates".
21
+ The class is initialized with a "gates" Tensor, which specifies which
22
+ batch elements go to which experts, and the weights to use when combining
23
+ the outputs. Batch element b is sent to expert e iff gates[b, e] != 0.
24
+ The inputs and outputs are all two-dimensional [batch, depth].
25
+ Caller is responsible for collapsing additional dimensions prior to
26
+ calling this class and reshaping the output to the original shape.
27
+ See common_layers.reshape_like().
28
+ Example use:
29
+ gates: a float32 `Tensor` with shape `[batch_size, num_experts]`
30
+ inputs: a float32 `Tensor` with shape `[batch_size, input_size]`
31
+ experts: a list of length `num_experts` containing sub-networks.
32
+ dispatcher = SparseDispatcher(num_experts, gates)
33
+ expert_inputs = dispatcher.dispatch(inputs)
34
+ expert_outputs = [experts[i](expert_inputs[i]) for i in range(num_experts)]
35
+ outputs = dispatcher.combine(expert_outputs)
36
+ The preceding code sets the output for a particular example b to:
37
+ output[b] = Sum_i(gates[b, i] * experts[i](inputs[b]))
38
+ This class takes advantage of sparsity in the gate matrix by including in the
39
+ `Tensor`s for expert i only the batch elements for which `gates[b, i] > 0`.
40
+ """
41
+
42
+ def __init__(self, num_experts, gates):
43
+ """Create a SparseDispatcher."""
44
+
45
+ self._gates = gates
46
+ self._num_experts = num_experts
47
+ # sort experts
48
+ sorted_experts, index_sorted_experts = torch.nonzero(gates).sort(0)
49
+ # drop indices
50
+ _, self._expert_index = sorted_experts.split(1, dim=1)
51
+ # get according batch index for each expert
52
+ self._batch_index = torch.nonzero(gates)[index_sorted_experts[:, 1], 0]
53
+ # calculate num samples that each expert gets
54
+ self._part_sizes = (gates > 0).sum(0).tolist()
55
+ # expand gates to match with self._batch_index
56
+ gates_exp = gates[self._batch_index.flatten()]
57
+ self._nonzero_gates = torch.gather(gates_exp, 1, self._expert_index)
58
+
59
+ def dispatch(self, D_Kernel, index_1):
60
+ b, c = D_Kernel.shape
61
+
62
+ D_Kernel_exp = D_Kernel[self._batch_index]
63
+
64
+ list1 = torch.zeros((1, self._num_experts))
65
+ list1[0, index_1] = b
66
+
67
+ return torch.split(D_Kernel_exp, list1[0].int().tolist(), dim=0)
68
+
69
+ def combine(self, expert_out, multiply_by_gates=True):
70
+ stitched = torch.cat(expert_out, 0).exp()
71
+ if multiply_by_gates:
72
+ stitched = stitched.mul(self._nonzero_gates.unsqueeze(1).unsqueeze(1))
73
+
74
+ zeros = torch.zeros(
75
+ (self._gates.size(0), expert_out[-1].size(1), expert_out[-1].size(2), expert_out[-1].size(3)),
76
+ requires_grad=True, device=stitched.device)
77
+
78
+ combined = zeros.index_add(0, self._batch_index, stitched.float())
79
+
80
+ # add eps to all zero values in order to avoid nans when going back to log space
81
+ combined[combined == 0] = np.finfo(float).eps
82
+ # back to log space
83
+ return combined.log()
84
+
85
+ def expert_to_gates(self):
86
+ """Gate values corresponding to the examples in the per-expert `Tensor`s.
87
+ Returns:
88
+ a list of `num_experts` one-dimensional `Tensor`s with type `tf.float32`
89
+ and shapes `[expert_batch_size_i]`
90
+ """
91
+ # split nonzero gates for each expert
92
+ return torch.split(self._nonzero_gates, self._part_sizes, dim=0)
93
+
94
+
95
+ class DecMoE(nn.Module):
96
+ """Call a Sparsely gated mixture of experts layer with 1-layer Feed-Forward networks as experts.
97
+ Args:
98
+ input_size: integer - size of the input
99
+ output_size: integer - size of the input
100
+ num_experts: an integer - number of experts
101
+ hidden_size: an integer - hidden size of the experts
102
+ noisy_gating: a boolean
103
+ k: an integer - how many experts to use for each batch element
104
+ """
105
+
106
+ def __init__(self, ds_inputsize, input_size, output_size, num_experts, hidden_size, noisy_gating=True, k=2,
107
+ trainingmode=True):
108
+ super(DecMoE, self).__init__()
109
+ self.noisy_gating = noisy_gating
110
+ self.num_experts = num_experts
111
+ self.output_size = output_size
112
+ self.input_size = input_size
113
+ self.hidden_size = hidden_size
114
+ self.training = trainingmode
115
+ self.k = k
116
+ # instantiate experts
117
+ self.experts = nn.ModuleList(
118
+ [generateKernel(hidden_size, 3), generateKernel(hidden_size, 5), generateKernel(hidden_size, 7),
119
+ generateKernel(hidden_size, 9)])
120
+ self.w_gate = nn.Parameter(torch.zeros(ds_inputsize, num_experts), requires_grad=True)
121
+ self.w_noise = nn.Parameter(torch.zeros(ds_inputsize, num_experts), requires_grad=True)
122
+
123
+ self.softplus = nn.Softplus()
124
+ self.softmax = nn.Softmax(1)
125
+ self.register_buffer("mean", torch.tensor([0.0]))
126
+ self.register_buffer("std", torch.tensor([1.0]))
127
+ assert (self.k <= self.num_experts)
128
+
129
+ def cv_squared(self, x):
130
+ """The squared coefficient of variation of a sample.
131
+ Useful as a loss to encourage a positive distribution to be more uniform.
132
+ Epsilons added for numerical stability.
133
+ Returns 0 for an empty Tensor.
134
+ Args:
135
+ x: a `Tensor`.
136
+ Returns:
137
+ a `Scalar`.
138
+ """
139
+ eps = 1e-10
140
+ # if only num_experts = 1
141
+
142
+ if x.shape[0] == 1:
143
+ return torch.tensor([0], device=x.device, dtype=x.dtype)
144
+ return x.float().var() / (x.float().mean() ** 2 + eps)
145
+
146
+ def _gates_to_load(self, gates):
147
+ """Compute the true load per expert, given the gates.
148
+ The load is the number of examples for which the corresponding gate is >0.
149
+ Args:
150
+ gates: a `Tensor` of shape [batch_size, n]
151
+ Returns:
152
+ a float32 `Tensor` of shape [n]
153
+ """
154
+ return (gates > 0).sum(0)
155
+
156
+ def _prob_in_top_k(self, clean_values, noisy_values, noise_stddev, noisy_top_values):
157
+ """Helper function to NoisyTopKGating.
158
+ Computes the probability that value is in top k, given different random noise.
159
+ This gives us a way of backpropagating from a loss that balances the number
160
+ of times each expert is in the top k experts per example.
161
+ In the case of no noise, pass in None for noise_stddev, and the result will
162
+ not be differentiable.
163
+ Args:
164
+ clean_values: a `Tensor` of shape [batch, n].
165
+ noisy_values: a `Tensor` of shape [batch, n]. Equal to clean values plus
166
+ normally distributed noise with standard deviation noise_stddev.
167
+ noise_stddev: a `Tensor` of shape [batch, n], or None
168
+ noisy_top_values: a `Tensor` of shape [batch, m].
169
+ "values" Output of tf.top_k(noisy_top_values, m). m >= k+1
170
+ Returns:
171
+ a `Tensor` of shape [batch, n].
172
+ """
173
+ batch = clean_values.size(0)
174
+ m = noisy_top_values.size(1)
175
+ top_values_flat = noisy_top_values.flatten()
176
+
177
+ threshold_positions_if_in = torch.arange(batch, device=clean_values.device) * m + self.k
178
+ threshold_if_in = torch.unsqueeze(torch.gather(top_values_flat, 0, threshold_positions_if_in), 1)
179
+ is_in = torch.gt(noisy_values, threshold_if_in)
180
+ threshold_positions_if_out = threshold_positions_if_in - 1
181
+ threshold_if_out = torch.unsqueeze(torch.gather(top_values_flat, 0, threshold_positions_if_out), 1)
182
+ # is each value currently in the top k.
183
+ normal = Normal(self.mean, self.std)
184
+ prob_if_in = normal.cdf((clean_values - threshold_if_in) / noise_stddev)
185
+ prob_if_out = normal.cdf((clean_values - threshold_if_out) / noise_stddev)
186
+ prob = torch.where(is_in, prob_if_in, prob_if_out)
187
+ return prob
188
+
189
+ def noisy_top_k_gating(self, x, train, noise_epsilon=1e-2):
190
+ """Noisy top-k gating.
191
+ See paper: https://arxiv.org/abs/1701.06538.
192
+ Args:
193
+ x: input Tensor with shape [batch_size, input_size]
194
+ train: a boolean - we only add noise at training time.
195
+ noise_epsilon: a float
196
+ Returns:
197
+ gates: a Tensor with shape [batch_size, num_experts]
198
+ load: a Tensor with shape [num_experts]
199
+ """
200
+ clean_logits = x @ self.w_gate
201
+ if self.noisy_gating and train:
202
+ raw_noise_stddev = x @ self.w_noise
203
+ noise_stddev = ((self.softplus(raw_noise_stddev) + noise_epsilon))
204
+ noisy_logits = clean_logits + (torch.randn_like(clean_logits) * noise_stddev)
205
+ logits = noisy_logits
206
+ else:
207
+ logits = clean_logits
208
+
209
+ # calculate topk + 1 that will be needed for the noisy gates
210
+ top_logits, top_indices = logits.topk(min(self.k + 1, self.num_experts), dim=1)
211
+ top_k_logits = top_logits[:, :self.k]
212
+ top_k_indices = top_indices[:, :self.k]
213
+ top_k_gates = self.softmax(top_k_logits)
214
+
215
+ zeros = torch.zeros_like(logits, requires_grad=True)
216
+ gates = zeros.scatter(1, top_k_indices, top_k_gates)
217
+
218
+ if self.noisy_gating and self.k < self.num_experts and train:
219
+ load = (self._prob_in_top_k(clean_logits, noisy_logits, noise_stddev, top_logits)).sum(0)
220
+ else:
221
+ load = self._gates_to_load(gates)
222
+ return gates, load, top_k_indices[0]
223
+
224
+ def forward(self, x_ds, D_Kernel, loss_coef=1e-2):
225
+ gates, load, index_1 = self.noisy_top_k_gating(x_ds, self.training)
226
+ # calculate importance loss
227
+ importance = gates.sum(0)
228
+
229
+ loss = self.cv_squared(importance) + self.cv_squared(load)
230
+ loss *= loss_coef
231
+
232
+ dispatcher = SparseDispatcher(self.num_experts, gates)
233
+ expert_kernel = dispatcher.dispatch(D_Kernel, index_1)
234
+ expert_outputs = [self.experts[i](expert_kernel[i]) for i in range(self.num_experts)]
235
+ return expert_outputs, loss
236
+
237
+
238
+ def default_conv(in_channels, out_channels, kernel_size, bias=True):
239
+ return nn.Conv2d(in_channels, out_channels, kernel_size, padding=(kernel_size // 2), bias=bias)
240
+
241
+
242
+ class CALayer(nn.Module):
243
+ def __init__(self, channel, reduction=16):
244
+ super(CALayer, self).__init__()
245
+ self.avg_pool = nn.AdaptiveAvgPool2d(1)
246
+ self.conv_du = nn.Sequential(
247
+ nn.Conv2d(channel, channel // reduction, 1, padding=0, bias=True),
248
+ nn.ReLU(inplace=True),
249
+ nn.Conv2d(channel // reduction, channel, 1, padding=0, bias=True),
250
+ nn.Sigmoid()
251
+ )
252
+
253
+ def forward(self, x):
254
+ y = self.avg_pool(x)
255
+ y = self.conv_du(y)
256
+ return x * y
257
+
258
+
259
+ class RCAB(nn.Module):
260
+ def __init__(
261
+ self, conv, n_feat, kernel_size, reduction,
262
+ bias=True, bn=False, act=nn.ReLU(True), res_scale=1):
263
+
264
+ super(RCAB, self).__init__()
265
+ modules_body = []
266
+ for i in range(2):
267
+ modules_body.append(conv(n_feat, n_feat, kernel_size, bias=bias))
268
+ if bn: modules_body.append(nn.BatchNorm2d(n_feat))
269
+ if i == 0: modules_body.append(act)
270
+ modules_body.append(CALayer(n_feat, reduction))
271
+ self.body = nn.Sequential(*modules_body)
272
+ self.res_scale = res_scale
273
+
274
+ def forward(self, x):
275
+ res = self.body(x)
276
+ res += x
277
+ return res
278
+
279
+
280
+ class ResidualGroup(nn.Module):
281
+ def __init__(self, conv, n_feat, kernel_size, reduction, n_resblocks):
282
+ super(ResidualGroup, self).__init__()
283
+ modules_body = []
284
+ modules_body = [
285
+ RCAB(
286
+ conv, n_feat, kernel_size, reduction, bias=True, bn=False,
287
+ act=nn.LeakyReLU(negative_slope=0.2, inplace=True), res_scale=1) \
288
+ for _ in range(n_resblocks)]
289
+ modules_body.append(conv(n_feat, n_feat, kernel_size))
290
+ self.body = nn.Sequential(*modules_body)
291
+
292
+ def forward(self, x):
293
+ res = self.body(x)
294
+ res += x
295
+ return res
296
+
297
+
298
+ class ResBlock(nn.Module):
299
+ def __init__(self, in_feat, out_feat, stride=1):
300
+ super(ResBlock, self).__init__()
301
+ self.backbone = nn.Sequential(
302
+ nn.Conv2d(in_feat, out_feat, kernel_size=3, stride=stride, padding=1, bias=False),
303
+ nn.BatchNorm2d(out_feat),
304
+ nn.LeakyReLU(0.1, True),
305
+ nn.Conv2d(out_feat, out_feat, kernel_size=3, padding=1, bias=False),
306
+ nn.BatchNorm2d(out_feat),
307
+ )
308
+ self.shortcut = nn.Sequential(
309
+ nn.Conv2d(in_feat, out_feat, kernel_size=1, stride=stride, bias=False),
310
+ nn.BatchNorm2d(out_feat)
311
+ )
312
+
313
+ def forward(self, x):
314
+ return nn.LeakyReLU(0.1, True)(self.backbone(x) + self.shortcut(x))
315
+
316
+
317
+ class DaEncoder(nn.Module):
318
+ def __init__(self, nfeats):
319
+ super(DaEncoder, self).__init__()
320
+
321
+ self.E_pre = nn.Sequential(
322
+ ResBlock(in_feat=1, out_feat=nfeats // 2, stride=1),
323
+ ResBlock(in_feat=nfeats // 2, out_feat=nfeats, stride=1),
324
+ ResBlock(in_feat=nfeats, out_feat=nfeats, stride=1)
325
+ )
326
+ self.E = nn.Sequential(
327
+ nn.Conv2d(nfeats, nfeats * 2, kernel_size=3, stride=2, padding=1),
328
+ nn.BatchNorm2d(nfeats * 2),
329
+ nn.LeakyReLU(0.1, True),
330
+ nn.Conv2d(nfeats * 2, nfeats * 4, kernel_size=3, stride=2, padding=1),
331
+ nn.BatchNorm2d(nfeats * 4),
332
+ nn.AdaptiveAvgPool2d(1)
333
+ )
334
+
335
+ def forward(self, x):
336
+ inter = self.E_pre(x)
337
+ fea = self.E(inter)
338
+
339
+ out = fea.squeeze(-1).squeeze(-1)
340
+
341
+ return fea, out, inter
342
+
343
+
344
+ class generateKernel(nn.Module):
345
+ def __init__(self, nfeats, kernel_size=5):
346
+ super(generateKernel, self).__init__()
347
+
348
+ self.mlp = nn.Sequential(
349
+ nn.Linear(nfeats * 4, nfeats),
350
+ nn.LeakyReLU(0.1, True),
351
+ nn.Linear(nfeats, kernel_size * kernel_size)
352
+ )
353
+
354
+ def forward(self, D_Kernel):
355
+ D_Kernel = self.mlp(D_Kernel)
356
+ return D_Kernel
357
+
358
+
359
+ class DAB(nn.Module):
360
+ def __init__(self):
361
+ super(DAB, self).__init__()
362
+ self.relu = nn.LeakyReLU(0.1, True)
363
+ self.conv = default_conv(1, 1, 1)
364
+
365
+ def forward(self, x, D_Kernel):
366
+ b, c, h, w = x.size()
367
+ b1, l = D_Kernel.shape
368
+ kernel_size = int(math.sqrt(l))
369
+ with torch.no_grad():
370
+ kernel = D_Kernel.view(-1, 1, kernel_size, kernel_size)
371
+ out = F.conv2d(x.view(1, -1, h, w), kernel, groups=b * c, padding=(kernel_size - 1) // 2)
372
+ out = out.view(b, -1, h, w)
373
+ out = self.conv(self.relu(out).view(b, -1, h, w))
374
+ return out
375
+
376
+
377
+ class DR(nn.Module):
378
+ def __init__(self, nfeats, num_experts=4, k=3):
379
+ super(DR, self).__init__()
380
+
381
+ self.topK = k
382
+ self.num_experts = num_experts
383
+ self.start_idx = num_experts - k
384
+
385
+ self.c1 = ResBlock(in_feat=1, out_feat=nfeats, stride=1)
386
+ self.gap = nn.AdaptiveMaxPool2d(1)
387
+ self.gap2 = nn.AdaptiveAvgPool2d(1)
388
+ self.fc1 = nn.Linear(nfeats, nfeats * 4)
389
+
390
+ self.dab = [DAB(), DAB(), DAB()]
391
+ self.dab_list = nn.ModuleList(self.dab)
392
+
393
+ self.DecoderMoE = DecMoE(ds_inputsize=nfeats * 4, input_size=1, output_size=1, num_experts=num_experts,
394
+ hidden_size=nfeats,
395
+ noisy_gating=True, k=k, trainingmode=True)
396
+
397
+ self.conv = default_conv(1, 1, 1)
398
+
399
+ def forward(self, lr, sr, D_Kernel):
400
+
401
+ y1 = F.interpolate(lr, scale_factor=0.125, mode='bicubic', align_corners=True,
402
+ recompute_scale_factor=True)
403
+ y2 = self.c1(y1)
404
+ y3 = self.gap(y2) + self.gap2(y2)
405
+ y4 = y3.view(y3.shape[0], -1)
406
+ y5 = self.fc1(y4)
407
+
408
+ D_Kernel_list, aux_loss = self.DecoderMoE(y5, D_Kernel, loss_coef=0.02)
409
+
410
+ sorted_D_Kernel_list = sorted(D_Kernel_list, key=lambda x: (x.size(0), x.size(1)))
411
+
412
+ sum_result = None
413
+ for iidx in range(self.start_idx, self.num_experts):
414
+ res_d = self.dab_list[iidx - self.start_idx](sr, sorted_D_Kernel_list[iidx])
415
+ if sum_result is None:
416
+ sum_result = res_d
417
+ else:
418
+ sum_result += res_d
419
+
420
+ out = self.conv(sum_result)
421
+ return out, aux_loss
422
+
423
+
424
+ class DA_rgb(nn.Module):
425
+ def __init__(self, channels_in, channels_out, kernel_size, reduction):
426
+ super(DA_rgb, self).__init__()
427
+
428
+ self.kernel_size = kernel_size
429
+ self.channels_out = channels_out
430
+ self.channels_in = channels_in
431
+
432
+ self.dcnrgb = DCN_layer_rgb(self.channels_in, self.channels_out, kernel_size,
433
+ padding=(kernel_size - 1) // 2, bias=False)
434
+
435
+ self.rcab1 = RCAB(default_conv, channels_out, 3, reduction)
436
+ self.relu = nn.LeakyReLU(0.1, True)
437
+ self.conv = default_conv(channels_in, channels_out, 3)
438
+
439
+ def forward(self, x, inter, fea):
440
+ out1 = self.rcab1(x)
441
+ out2 = self.dcnrgb(out1, inter, fea)
442
+ out = self.conv(out2 + out1)
443
+ return out
444
+
445
+
446
+ class FusionBlock(nn.Module):
447
+ def __init__(self, channels_in, channels_out):
448
+ super(FusionBlock, self).__init__()
449
+ self.conv1 = default_conv(channels_in, channels_in // 4, 1)
450
+ self.conv2 = default_conv(channels_in, channels_in // 4, 1)
451
+ self.conv3 = default_conv(channels_in // 4, channels_in, 1)
452
+ self.sigmoid = nn.Sigmoid()
453
+
454
+ self.conv = default_conv(2 * channels_in, channels_out, 3)
455
+
456
+ def forward(self, rgb, dep, inter):
457
+ inter1 = self.conv1(inter)
458
+ rgb1 = self.conv2(rgb)
459
+
460
+ w = torch.sigmoid(inter1)
461
+ rgb2 = rgb1 * w
462
+ rgb3 = self.conv3(rgb2) + rgb
463
+ cat1 = torch.cat([rgb3, dep], dim=1)
464
+ out = self.conv(cat1)
465
+
466
+ return out
467
+
468
+
469
+ class DOFT(nn.Module):
470
+ def __init__(self, channels_in, channels_out, kernel_size, reduction):
471
+ super(DOFT, self).__init__()
472
+ self.channels_out = channels_out
473
+ self.channels_in = channels_in
474
+ self.kernel_size = kernel_size
475
+
476
+ self.DA_rgb = DA_rgb(channels_in, channels_out, kernel_size, reduction)
477
+ self.fb = FusionBlock(channels_in, channels_out)
478
+
479
+ self.relu = nn.LeakyReLU(0.1, True)
480
+
481
+ def forward(self, x, inter, rgb, fea):
482
+ rgb = self.DA_rgb(rgb, inter, fea)
483
+
484
+ out1 = self.fb(rgb, x, inter)
485
+ out = x + out1
486
+ return out
487
+
488
+
489
+ class DSRN(nn.Module):
490
+ def __init__(self, nfeats=64, reduction=16, conv=default_conv):
491
+ super(DSRN, self).__init__()
492
+
493
+ kernel_size = 3
494
+
495
+ n_feats = nfeats
496
+
497
+ # head module
498
+ modules_head = [conv(1, n_feats, kernel_size)]
499
+ self.head = nn.Sequential(*modules_head)
500
+
501
+ modules_head_rgb = [conv(3, n_feats, kernel_size)]
502
+ self.head_rgb = nn.Sequential(*modules_head_rgb)
503
+
504
+ self.dgm1 = DOFT(n_feats, n_feats, 3, reduction)
505
+ self.dgm2 = DOFT(n_feats, n_feats, 3, reduction)
506
+ self.dgm3 = DOFT(n_feats, n_feats, 3, reduction)
507
+ self.dgm4 = DOFT(n_feats, n_feats, 3, reduction)
508
+ self.dgm5 = DOFT(n_feats, n_feats, 3, reduction)
509
+
510
+ self.c_d1 = ResidualGroup(conv, n_feats, 3, reduction=reduction, n_resblocks=2)
511
+ self.c_d2 = ResidualGroup(conv, n_feats, 3, reduction=reduction, n_resblocks=2)
512
+ self.c_d3 = ResidualGroup(conv, n_feats, 3, reduction=reduction, n_resblocks=2)
513
+ self.c_d4 = ResidualGroup(conv, n_feats, 3, reduction=reduction, n_resblocks=2)
514
+
515
+ modules_d5 = [conv(5 * n_feats, n_feats, 1),
516
+ ResidualGroup(conv, n_feats, 3, reduction=reduction, n_resblocks=2)]
517
+ self.c_d5 = nn.Sequential(*modules_d5)
518
+
519
+ self.c_r1 = conv(n_feats, n_feats, kernel_size)
520
+ self.c_r2 = conv(n_feats, n_feats, kernel_size)
521
+ self.c_r3 = conv(n_feats, n_feats, kernel_size)
522
+ self.c_r4 = conv(n_feats, n_feats, kernel_size)
523
+
524
+ self.act = nn.LeakyReLU(0.1, True)
525
+
526
+ # tail
527
+ modules_tail = [conv(n_feats, 1, kernel_size)]
528
+ self.tail = nn.Sequential(*modules_tail)
529
+
530
+ def forward(self, x, inter, rgb, fea):
531
+ # head
532
+ x = self.head(x)
533
+ rgb = self.head_rgb(rgb)
534
+ rgb1 = self.c_r1(rgb)
535
+ rgb2 = self.c_r2(self.act(rgb1))
536
+ rgb3 = self.c_r3(self.act(rgb2))
537
+ rgb4 = self.c_r4(self.act(rgb3))
538
+
539
+ dep10 = self.dgm1(x, inter, rgb, fea)
540
+ dep1 = self.c_d1(dep10)
541
+ dep20 = self.dgm2(dep1, inter, rgb1, fea)
542
+ dep2 = self.c_d2(self.act(dep20))
543
+ dep30 = self.dgm3(dep2, inter, rgb2, fea)
544
+ dep3 = self.c_d3(self.act(dep30))
545
+ dep40 = self.dgm4(dep3, inter, rgb3, fea)
546
+ dep4 = self.c_d4(self.act(dep40))
547
+ dep50 = self.dgm5(dep4, inter, rgb4, fea)
548
+
549
+ cat1 = torch.cat([dep1, dep2, dep3, dep4, dep50], dim=1)
550
+ dep6 = self.c_d5(cat1)
551
+
552
+ res = dep6 + x
553
+
554
+ out = self.tail(res)
555
+
556
+ return out
557
+
558
+
559
+ class Net(nn.Module):
560
+ def __init__(self, tiny_model=False):
561
+ super(Net, self).__init__()
562
+
563
+ if tiny_model:
564
+ n_feats = 24
565
+ reduction = 4
566
+ else:
567
+ n_feats = 64
568
+ reduction = 16
569
+
570
+ # Restorer
571
+ self.R = DSRN(nfeats=n_feats, reduction=reduction)
572
+ self.training = False
573
+ # Encoder
574
+ self.Enc = DaEncoder(nfeats=n_feats)
575
+ self.Dab = DR(nfeats=n_feats)
576
+
577
+ def forward(self, x_query, rgb):
578
+
579
+ fea, d_kernel, inter = self.Enc(x_query)
580
+ restored = self.R(x_query, inter, rgb, fea)
581
+
582
+ if self.training:
583
+ d_lr_, aux_loss = self.Dab(x_query, restored, d_kernel)
584
+ return restored, d_lr_, aux_loss
585
+ else:
586
+ return restored
net/dornet_ddp.py ADDED
@@ -0,0 +1,600 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import torch
2
+ import torch.nn as nn
3
+ from .deform_conv import DCN_layer_rgb
4
+ import torch.nn.functional as F
5
+ import math
6
+ from net.CR import *
7
+ from torch.distributions.normal import Normal
8
+ import numpy as np
9
+
10
+
11
+ class SparseDispatcher(object):
12
+ """Helper for implementing a mixture of experts.
13
+ The purpose of this class is to create input minibatches for the
14
+ experts and to combine the results of the experts to form a unified
15
+ output tensor.
16
+ There are two functions:
17
+ dispatch - take an input Tensor and create input Tensors for each expert.
18
+ combine - take output Tensors from each expert and form a combined output
19
+ Tensor. Outputs from different experts for the same batch element are
20
+ summed together, weighted by the provided "gates".
21
+ The class is initialized with a "gates" Tensor, which specifies which
22
+ batch elements go to which experts, and the weights to use when combining
23
+ the outputs. Batch element b is sent to expert e iff gates[b, e] != 0.
24
+ The inputs and outputs are all two-dimensional [batch, depth].
25
+ Caller is responsible for collapsing additional dimensions prior to
26
+ calling this class and reshaping the output to the original shape.
27
+ See common_layers.reshape_like().
28
+ Example use:
29
+ gates: a float32 `Tensor` with shape `[batch_size, num_experts]`
30
+ inputs: a float32 `Tensor` with shape `[batch_size, input_size]`
31
+ experts: a list of length `num_experts` containing sub-networks.
32
+ dispatcher = SparseDispatcher(num_experts, gates)
33
+ expert_inputs = dispatcher.dispatch(inputs)
34
+ expert_outputs = [experts[i](expert_inputs[i]) for i in range(num_experts)]
35
+ outputs = dispatcher.combine(expert_outputs)
36
+ The preceding code sets the output for a particular example b to:
37
+ output[b] = Sum_i(gates[b, i] * experts[i](inputs[b]))
38
+ This class takes advantage of sparsity in the gate matrix by including in the
39
+ `Tensor`s for expert i only the batch elements for which `gates[b, i] > 0`.
40
+ """
41
+
42
+ def __init__(self, num_experts, gates):
43
+ """Create a SparseDispatcher."""
44
+
45
+ self._gates = gates
46
+ self._num_experts = num_experts
47
+ # sort experts
48
+ sorted_experts, index_sorted_experts = torch.nonzero(gates).sort(0)
49
+ # drop indices
50
+ _, self._expert_index = sorted_experts.split(1, dim=1)
51
+ # get according batch index for each expert
52
+ self._batch_index = torch.nonzero(gates)[index_sorted_experts[:, 1], 0]
53
+ # calculate num samples that each expert gets
54
+ self._part_sizes = (gates > 0).sum(0).tolist()
55
+ # expand gates to match with self._batch_index
56
+ gates_exp = gates[self._batch_index.flatten()]
57
+ self._nonzero_gates = torch.gather(gates_exp, 1, self._expert_index)
58
+
59
+ def dispatch(self, D_Kernel, index_1):
60
+ b, c = D_Kernel.shape
61
+
62
+ D_Kernel_exp = D_Kernel[self._batch_index]
63
+
64
+ list1 = torch.zeros((1, self._num_experts))
65
+ list1[0, index_1] = b
66
+
67
+ return torch.split(D_Kernel_exp, list1[0].int().tolist(), dim=0)
68
+
69
+ def combine(self, expert_out, multiply_by_gates=True):
70
+ stitched = torch.cat(expert_out, 0).exp()
71
+ if multiply_by_gates:
72
+ stitched = stitched.mul(self._nonzero_gates.unsqueeze(1).unsqueeze(1))
73
+
74
+ zeros = torch.zeros(
75
+ (self._gates.size(0), expert_out[-1].size(1), expert_out[-1].size(2), expert_out[-1].size(3)),
76
+ requires_grad=True, device=stitched.device)
77
+
78
+ combined = zeros.index_add(0, self._batch_index, stitched.float())
79
+
80
+ # add eps to all zero values in order to avoid nans when going back to log space
81
+ combined[combined == 0] = np.finfo(float).eps
82
+ # back to log space
83
+ return combined.log()
84
+
85
+ def expert_to_gates(self):
86
+ """Gate values corresponding to the examples in the per-expert `Tensor`s.
87
+ Returns:
88
+ a list of `num_experts` one-dimensional `Tensor`s with type `tf.float32`
89
+ and shapes `[expert_batch_size_i]`
90
+ """
91
+ # split nonzero gates for each expert
92
+ return torch.split(self._nonzero_gates, self._part_sizes, dim=0)
93
+
94
+
95
+ class DecMoE(nn.Module):
96
+ """Call a Sparsely gated mixture of experts layer with 1-layer Feed-Forward networks as experts.
97
+ Args:
98
+ input_size: integer - size of the input
99
+ output_size: integer - size of the input
100
+ num_experts: an integer - number of experts
101
+ hidden_size: an integer - hidden size of the experts
102
+ noisy_gating: a boolean
103
+ k: an integer - how many experts to use for each batch element
104
+ """
105
+
106
+ def __init__(self, ds_inputsize, input_size, output_size, num_experts, hidden_size, noisy_gating=True, k=2,
107
+ trainingmode=True):
108
+ super(DecMoE, self).__init__()
109
+ self.noisy_gating = noisy_gating
110
+ self.num_experts = num_experts
111
+ self.output_size = output_size
112
+ self.input_size = input_size
113
+ self.hidden_size = hidden_size
114
+ self.training = trainingmode
115
+ self.k = k
116
+ # instantiate experts
117
+ self.experts = nn.ModuleList(
118
+ [generateKernel(hidden_size, 3), generateKernel(hidden_size, 5), generateKernel(hidden_size, 7),
119
+ generateKernel(hidden_size, 9)])
120
+ self.w_gate = nn.Parameter(torch.zeros(ds_inputsize, num_experts), requires_grad=True)
121
+ self.w_noise = nn.Parameter(torch.zeros(ds_inputsize, num_experts), requires_grad=True)
122
+
123
+ self.softplus = nn.Softplus()
124
+ self.softmax = nn.Softmax(1)
125
+ self.register_buffer("mean", torch.tensor([0.0]))
126
+ self.register_buffer("std", torch.tensor([1.0]))
127
+ assert (self.k <= self.num_experts)
128
+
129
+ def cv_squared(self, x):
130
+ """The squared coefficient of variation of a sample.
131
+ Useful as a loss to encourage a positive distribution to be more uniform.
132
+ Epsilons added for numerical stability.
133
+ Returns 0 for an empty Tensor.
134
+ Args:
135
+ x: a `Tensor`.
136
+ Returns:
137
+ a `Scalar`.
138
+ """
139
+ eps = 1e-10
140
+ # if only num_experts = 1
141
+
142
+ if x.shape[0] == 1:
143
+ return torch.tensor([0], device=x.device, dtype=x.dtype)
144
+ return x.float().var() / (x.float().mean() ** 2 + eps)
145
+
146
+ def _gates_to_load(self, gates):
147
+ """Compute the true load per expert, given the gates.
148
+ The load is the number of examples for which the corresponding gate is >0.
149
+ Args:
150
+ gates: a `Tensor` of shape [batch_size, n]
151
+ Returns:
152
+ a float32 `Tensor` of shape [n]
153
+ """
154
+ return (gates > 0).sum(0)
155
+
156
+ def _prob_in_top_k(self, clean_values, noisy_values, noise_stddev, noisy_top_values):
157
+ """Helper function to NoisyTopKGating.
158
+ Computes the probability that value is in top k, given different random noise.
159
+ This gives us a way of backpropagating from a loss that balances the number
160
+ of times each expert is in the top k experts per example.
161
+ In the case of no noise, pass in None for noise_stddev, and the result will
162
+ not be differentiable.
163
+ Args:
164
+ clean_values: a `Tensor` of shape [batch, n].
165
+ noisy_values: a `Tensor` of shape [batch, n]. Equal to clean values plus
166
+ normally distributed noise with standard deviation noise_stddev.
167
+ noise_stddev: a `Tensor` of shape [batch, n], or None
168
+ noisy_top_values: a `Tensor` of shape [batch, m].
169
+ "values" Output of tf.top_k(noisy_top_values, m). m >= k+1
170
+ Returns:
171
+ a `Tensor` of shape [batch, n].
172
+ """
173
+ batch = clean_values.size(0)
174
+ m = noisy_top_values.size(1)
175
+ top_values_flat = noisy_top_values.flatten()
176
+
177
+ threshold_positions_if_in = torch.arange(batch, device=clean_values.device) * m + self.k
178
+ threshold_if_in = torch.unsqueeze(torch.gather(top_values_flat, 0, threshold_positions_if_in), 1)
179
+ is_in = torch.gt(noisy_values, threshold_if_in)
180
+ threshold_positions_if_out = threshold_positions_if_in - 1
181
+ threshold_if_out = torch.unsqueeze(torch.gather(top_values_flat, 0, threshold_positions_if_out), 1)
182
+ # is each value currently in the top k.
183
+ normal = Normal(self.mean, self.std)
184
+ prob_if_in = normal.cdf((clean_values - threshold_if_in) / noise_stddev)
185
+ prob_if_out = normal.cdf((clean_values - threshold_if_out) / noise_stddev)
186
+ prob = torch.where(is_in, prob_if_in, prob_if_out)
187
+ return prob
188
+
189
+ def noisy_top_k_gating(self, x, train, noise_epsilon=1e-2):
190
+ """Noisy top-k gating.
191
+ See paper: https://arxiv.org/abs/1701.06538.
192
+ Args:
193
+ x: input Tensor with shape [batch_size, input_size]
194
+ train: a boolean - we only add noise at training time.
195
+ noise_epsilon: a float
196
+ Returns:
197
+ gates: a Tensor with shape [batch_size, num_experts]
198
+ load: a Tensor with shape [num_experts]
199
+ """
200
+ clean_logits = x @ self.w_gate
201
+ if self.noisy_gating and train:
202
+ raw_noise_stddev = x @ self.w_noise
203
+ noise_stddev = ((self.softplus(raw_noise_stddev) + noise_epsilon))
204
+ noisy_logits = clean_logits + (torch.randn_like(clean_logits) * noise_stddev)
205
+ logits = noisy_logits
206
+ else:
207
+ logits = clean_logits
208
+
209
+ # calculate topk + 1 that will be needed for the noisy gates
210
+ top_logits, top_indices = logits.topk(min(self.k + 1, self.num_experts), dim=1)
211
+ top_k_logits = top_logits[:, :self.k]
212
+ top_k_indices = top_indices[:, :self.k]
213
+ top_k_gates = self.softmax(top_k_logits)
214
+
215
+ zeros = torch.zeros_like(logits, requires_grad=True)
216
+ gates = zeros.scatter(1, top_k_indices, top_k_gates)
217
+
218
+ if self.noisy_gating and self.k < self.num_experts and train:
219
+ load = (self._prob_in_top_k(clean_logits, noisy_logits, noise_stddev, top_logits)).sum(0)
220
+ else:
221
+ load = self._gates_to_load(gates)
222
+ return gates, load, top_k_indices[0]
223
+
224
+ def forward(self, x_ds, D_Kernel, loss_coef=1e-2):
225
+ gates, load, index_1 = self.noisy_top_k_gating(x_ds, self.training)
226
+ # calculate importance loss
227
+ importance = gates.sum(0)
228
+
229
+ loss = self.cv_squared(importance) + self.cv_squared(load)
230
+ loss *= loss_coef
231
+
232
+ dispatcher = SparseDispatcher(self.num_experts, gates)
233
+ expert_kernel = dispatcher.dispatch(D_Kernel, index_1)
234
+ expert_outputs = [self.experts[i](expert_kernel[i]) for i in range(self.num_experts)]
235
+
236
+ return expert_outputs, loss
237
+
238
+
239
+ def default_conv(in_channels, out_channels, kernel_size, bias=True):
240
+ return nn.Conv2d(in_channels, out_channels, kernel_size, padding=(kernel_size // 2), bias=bias)
241
+
242
+
243
+ class CALayer(nn.Module):
244
+ def __init__(self, channel, reduction=16):
245
+ super(CALayer, self).__init__()
246
+ self.avg_pool = nn.AdaptiveAvgPool2d(1)
247
+ self.conv_du = nn.Sequential(
248
+ nn.Conv2d(channel, channel // reduction, 1, padding=0, bias=True),
249
+ nn.ReLU(inplace=True),
250
+ nn.Conv2d(channel // reduction, channel, 1, padding=0, bias=True),
251
+ nn.Sigmoid()
252
+ )
253
+
254
+ def forward(self, x):
255
+ y = self.avg_pool(x)
256
+ y = self.conv_du(y)
257
+ return x * y
258
+
259
+
260
+ class RCAB(nn.Module):
261
+ def __init__(
262
+ self, conv, n_feat, kernel_size, reduction,
263
+ bias=True, bn=False, act=nn.ReLU(True), res_scale=1):
264
+
265
+ super(RCAB, self).__init__()
266
+ modules_body = []
267
+ for i in range(2):
268
+ modules_body.append(conv(n_feat, n_feat, kernel_size, bias=bias))
269
+ if bn: modules_body.append(nn.BatchNorm2d(n_feat))
270
+ if i == 0: modules_body.append(act)
271
+ modules_body.append(CALayer(n_feat, reduction))
272
+ self.body = nn.Sequential(*modules_body)
273
+ self.res_scale = res_scale
274
+
275
+ def forward(self, x):
276
+ res = self.body(x)
277
+ res += x
278
+ return res
279
+
280
+
281
+ class ResidualGroup(nn.Module):
282
+ def __init__(self, conv, n_feat, kernel_size, reduction, n_resblocks):
283
+ super(ResidualGroup, self).__init__()
284
+ modules_body = []
285
+ modules_body = [
286
+ RCAB(
287
+ conv, n_feat, kernel_size, reduction, bias=True, bn=False,
288
+ act=nn.LeakyReLU(negative_slope=0.2, inplace=True), res_scale=1) \
289
+ for _ in range(n_resblocks)]
290
+ modules_body.append(conv(n_feat, n_feat, kernel_size))
291
+ self.body = nn.Sequential(*modules_body)
292
+
293
+ def forward(self, x):
294
+ res = self.body(x)
295
+ res += x
296
+ return res
297
+
298
+
299
+ class ResBlock(nn.Module):
300
+ def __init__(self, in_feat, out_feat, stride=1):
301
+ super(ResBlock, self).__init__()
302
+ self.backbone = nn.Sequential(
303
+ nn.Conv2d(in_feat, out_feat, kernel_size=3, stride=stride, padding=1, bias=False),
304
+ nn.BatchNorm2d(out_feat),
305
+ nn.LeakyReLU(0.1, True),
306
+ nn.Conv2d(out_feat, out_feat, kernel_size=3, padding=1, bias=False),
307
+ nn.BatchNorm2d(out_feat),
308
+ )
309
+ self.shortcut = nn.Sequential(
310
+ nn.Conv2d(in_feat, out_feat, kernel_size=1, stride=stride, bias=False),
311
+ nn.BatchNorm2d(out_feat)
312
+ )
313
+
314
+ def forward(self, x):
315
+ return nn.LeakyReLU(0.1, True)(self.backbone(x) + self.shortcut(x))
316
+
317
+
318
+ class DaEncoder(nn.Module):
319
+ def __init__(self, nfeats):
320
+ super(DaEncoder, self).__init__()
321
+
322
+ self.E_pre = nn.Sequential(
323
+ ResBlock(in_feat=1, out_feat=nfeats // 2, stride=1),
324
+ ResBlock(in_feat=nfeats // 2, out_feat=nfeats, stride=1),
325
+ ResBlock(in_feat=nfeats, out_feat=nfeats, stride=1)
326
+ )
327
+ self.E = nn.Sequential(
328
+ nn.Conv2d(nfeats, nfeats * 2, kernel_size=3, stride=2, padding=1),
329
+ nn.BatchNorm2d(nfeats * 2),
330
+ nn.LeakyReLU(0.1, True),
331
+ nn.Conv2d(nfeats * 2, nfeats * 4, kernel_size=3, stride=2, padding=1),
332
+ nn.BatchNorm2d(nfeats * 4),
333
+ nn.AdaptiveAvgPool2d(1)
334
+ )
335
+
336
+ def forward(self, x):
337
+ inter = self.E_pre(x)
338
+ fea = self.E(inter)
339
+
340
+ out = fea.squeeze(-1).squeeze(-1)
341
+
342
+ return fea, out, inter
343
+
344
+
345
+ class generateKernel(nn.Module):
346
+ def __init__(self, nfeats, kernel_size=5):
347
+ super(generateKernel, self).__init__()
348
+
349
+ self.mlp = nn.Sequential(
350
+ nn.Linear(nfeats * 4, nfeats),
351
+ nn.LeakyReLU(0.1, True),
352
+ nn.Linear(nfeats, kernel_size * kernel_size)
353
+ )
354
+
355
+ def forward(self, D_Kernel):
356
+ D_Kernel = self.mlp(D_Kernel)
357
+ return D_Kernel
358
+
359
+
360
+ class DAB(nn.Module):
361
+ def __init__(self):
362
+ super(DAB, self).__init__()
363
+ self.relu = nn.LeakyReLU(0.1, True)
364
+ self.conv = default_conv(1, 1, 1)
365
+
366
+ def forward(self, x, D_Kernel):
367
+ b, c, h, w = x.size()
368
+ b1, l = D_Kernel.shape
369
+ kernel_size = int(math.sqrt(l))
370
+ with torch.no_grad():
371
+ kernel = D_Kernel.view(-1, 1, kernel_size, kernel_size)
372
+ out = F.conv2d(x.view(1, -1, h, w), kernel, groups=b * c, padding=(kernel_size - 1) // 2)
373
+ out = out.view(b, -1, h, w)
374
+ out = self.conv(self.relu(out).view(b, -1, h, w))
375
+ return out
376
+
377
+
378
+ class DR(nn.Module):
379
+ def __init__(self, nfeats, num_experts=4, k=3):
380
+ super(DR, self).__init__()
381
+
382
+ self.topK = k
383
+ self.num_experts = num_experts
384
+ self.start_idx = num_experts - k
385
+
386
+ self.c1 = ResBlock(in_feat=1, out_feat=nfeats, stride=1)
387
+ self.gap = nn.AdaptiveMaxPool2d(1)
388
+ self.gap2 = nn.AdaptiveAvgPool2d(1)
389
+ self.fc1 = nn.Linear(nfeats, nfeats * 4)
390
+
391
+ self.dab = [DAB(), DAB(), DAB()]
392
+ self.dab_list = nn.ModuleList(self.dab)
393
+
394
+ self.DecoderMoE = DecMoE(ds_inputsize=nfeats * 4, input_size=1, output_size=1, num_experts=num_experts,
395
+ hidden_size=nfeats,
396
+ noisy_gating=True, k=k, trainingmode=True)
397
+
398
+ self.conv = default_conv(1, 1, 1)
399
+
400
+ def forward(self, lr, sr, D_Kernel):
401
+
402
+ y1 = F.interpolate(lr, scale_factor=0.125, mode='bicubic', align_corners=True,
403
+ recompute_scale_factor=True)
404
+ y2 = self.c1(y1)
405
+ y3 = self.gap(y2) + self.gap2(y2)
406
+ y4 = y3.view(y3.shape[0], -1)
407
+ y5 = self.fc1(y4)
408
+
409
+ D_Kernel_list, aux_loss = self.DecoderMoE(y5, D_Kernel, loss_coef=0.02)
410
+
411
+ sorted_D_Kernel_list = sorted(D_Kernel_list, key=lambda x: (x.size(0), x.size(1)))
412
+
413
+ sum_result = None
414
+ for iidx in range(self.start_idx, self.num_experts):
415
+ res_d = self.dab_list[iidx - self.start_idx](sr, sorted_D_Kernel_list[iidx])
416
+ if sum_result is None:
417
+ sum_result = res_d
418
+ else:
419
+ sum_result += res_d
420
+
421
+ out = self.conv(sum_result)
422
+ return out, aux_loss
423
+
424
+
425
+ class DA_rgb(nn.Module):
426
+ def __init__(self, channels_in, channels_out, kernel_size, reduction):
427
+ super(DA_rgb, self).__init__()
428
+
429
+ self.kernel_size = kernel_size
430
+ self.channels_out = channels_out
431
+ self.channels_in = channels_in
432
+
433
+ self.dcnrgb = DCN_layer_rgb(self.channels_in, self.channels_out, kernel_size,
434
+ padding=(kernel_size - 1) // 2, bias=False)
435
+
436
+ self.rcab1 = RCAB(default_conv, channels_out, 3, reduction)
437
+ self.relu = nn.LeakyReLU(0.1, True)
438
+ self.conv = default_conv(channels_in, channels_out, 3)
439
+
440
+ def forward(self, x, inter, fea):
441
+ out1 = self.rcab1(x)
442
+ out2 = self.dcnrgb(out1, inter, fea)
443
+ out = self.conv(out2 + out1)
444
+ return out
445
+
446
+
447
+ class FusionBlock(nn.Module):
448
+ def __init__(self, channels_in, channels_out):
449
+ super(FusionBlock, self).__init__()
450
+ self.conv1 = default_conv(channels_in, channels_in // 4, 1)
451
+ self.conv2 = default_conv(channels_in, channels_in // 4, 1)
452
+ self.conv3 = default_conv(channels_in // 4, channels_in, 1)
453
+ self.sigmoid = nn.Sigmoid()
454
+
455
+ self.conv = default_conv(2 * channels_in, channels_out, 3)
456
+
457
+ def forward(self, rgb, dep, inter):
458
+ inter1 = self.conv1(inter)
459
+ rgb1 = self.conv2(rgb)
460
+
461
+ w = torch.sigmoid(inter1)
462
+ rgb2 = rgb1 * w
463
+ rgb3 = self.conv3(rgb2) + rgb
464
+ cat1 = torch.cat([rgb3, dep], dim=1)
465
+ out = self.conv(cat1)
466
+
467
+ return out
468
+
469
+
470
+ class DOFT(nn.Module):
471
+ def __init__(self, channels_in, channels_out, kernel_size, reduction):
472
+ super(DOFT, self).__init__()
473
+ self.channels_out = channels_out
474
+ self.channels_in = channels_in
475
+ self.kernel_size = kernel_size
476
+
477
+ self.DA_rgb = DA_rgb(channels_in, channels_out, kernel_size, reduction)
478
+ self.fb = FusionBlock(channels_in, channels_out)
479
+
480
+ self.relu = nn.LeakyReLU(0.1, True)
481
+
482
+ def forward(self, x, inter, rgb, fea):
483
+ rgb = self.DA_rgb(rgb, inter, fea)
484
+
485
+ out1 = self.fb(rgb, x, inter)
486
+ out = x + out1
487
+ return out
488
+
489
+
490
+ class DSRN(nn.Module):
491
+ def __init__(self, nfeats=64, reduction=16, conv=default_conv):
492
+ super(DSRN, self).__init__()
493
+
494
+ kernel_size = 3
495
+
496
+ n_feats = nfeats
497
+
498
+ # head module
499
+ modules_head = [conv(1, n_feats, kernel_size)]
500
+ self.head = nn.Sequential(*modules_head)
501
+
502
+ modules_head_rgb = [conv(3, n_feats, kernel_size)]
503
+ self.head_rgb = nn.Sequential(*modules_head_rgb)
504
+
505
+ self.dgm1 = DOFT(n_feats, n_feats, 3, reduction)
506
+ self.dgm2 = DOFT(n_feats, n_feats, 3, reduction)
507
+ self.dgm3 = DOFT(n_feats, n_feats, 3, reduction)
508
+ self.dgm4 = DOFT(n_feats, n_feats, 3, reduction)
509
+ self.dgm5 = DOFT(n_feats, n_feats, 3, reduction)
510
+
511
+ self.c_d1 = ResidualGroup(conv, n_feats, 3, reduction=reduction, n_resblocks=2)
512
+ self.c_d2 = ResidualGroup(conv, n_feats, 3, reduction=reduction, n_resblocks=2)
513
+ self.c_d3 = ResidualGroup(conv, n_feats, 3, reduction=reduction, n_resblocks=2)
514
+ self.c_d4 = ResidualGroup(conv, n_feats, 3, reduction=reduction, n_resblocks=2)
515
+
516
+ modules_d5 = [conv(5 * n_feats, n_feats, 1),
517
+ ResidualGroup(conv, n_feats, 3, reduction=reduction, n_resblocks=2)]
518
+ self.c_d5 = nn.Sequential(*modules_d5)
519
+
520
+ self.c_r1 = conv(n_feats, n_feats, kernel_size)
521
+ self.c_r2 = conv(n_feats, n_feats, kernel_size)
522
+ self.c_r3 = conv(n_feats, n_feats, kernel_size)
523
+ self.c_r4 = conv(n_feats, n_feats, kernel_size)
524
+
525
+ self.act = nn.LeakyReLU(0.1, True)
526
+
527
+ # tail
528
+ modules_tail = [conv(n_feats, 1, kernel_size)]
529
+ self.tail = nn.Sequential(*modules_tail)
530
+
531
+ def forward(self, x, inter, rgb, fea):
532
+ # head
533
+ x = self.head(x)
534
+ rgb = self.head_rgb(rgb)
535
+ rgb1 = self.c_r1(rgb)
536
+ rgb2 = self.c_r2(self.act(rgb1))
537
+ rgb3 = self.c_r3(self.act(rgb2))
538
+ rgb4 = self.c_r4(self.act(rgb3))
539
+
540
+ dep10 = self.dgm1(x, inter, rgb, fea)
541
+ dep1 = self.c_d1(dep10)
542
+ dep20 = self.dgm2(dep1, inter, rgb1, fea)
543
+ dep2 = self.c_d2(self.act(dep20))
544
+ dep30 = self.dgm3(dep2, inter, rgb2, fea)
545
+ dep3 = self.c_d3(self.act(dep30))
546
+ dep40 = self.dgm4(dep3, inter, rgb3, fea)
547
+ dep4 = self.c_d4(self.act(dep40))
548
+ dep50 = self.dgm5(dep4, inter, rgb4, fea)
549
+
550
+ cat1 = torch.cat([dep1, dep2, dep3, dep4, dep50], dim=1)
551
+ dep6 = self.c_d5(cat1)
552
+
553
+ res = dep6 + x
554
+
555
+ out = self.tail(res)
556
+
557
+ return out
558
+
559
+ class SRN(nn.Module):
560
+ def __init__(self, nfeats, reduction):
561
+ super(SRN, self).__init__()
562
+
563
+ # Restorer
564
+ self.R = DSRN(nfeats=nfeats, reduction=reduction)
565
+
566
+ # Encoder
567
+ self.Enc = DaEncoder(nfeats=nfeats)
568
+
569
+ def forward(self, x_query, rgb):
570
+
571
+ fea, d_kernel, inter = self.Enc(x_query)
572
+ restored = self.R(x_query, inter, rgb, fea)
573
+
574
+ return restored, d_kernel
575
+
576
+
577
+ class Net_ddp(nn.Module):
578
+ def __init__(self, tiny_model=False):
579
+ super(Net_ddp, self).__init__()
580
+
581
+ if tiny_model:
582
+ n_feats = 24
583
+ reduction = 4
584
+ else:
585
+ n_feats = 64
586
+ reduction = 16
587
+
588
+ self.srn = SRN(nfeats=n_feats, reduction=reduction)
589
+ self.Dab = DR(nfeats=n_feats)
590
+
591
+ self.CLLoss = ContrastLoss(ablation=False)
592
+
593
+ def forward(self, x_query, rgb):
594
+
595
+ restored, d_kernel = self.srn(x_query, rgb)
596
+
597
+ d_lr_, aux_loss = self.Dab(x_query,restored, d_kernel)
598
+ CLLoss1 = self.CLLoss(d_lr_, x_query, restored)
599
+
600
+ return restored, d_lr_, aux_loss, CLLoss1
test_img.py ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import numpy as np
2
+ import os
3
+ import torch
4
+ import cv2
5
+ from net.dornet import Net
6
+ from data.rgbdd_dataloader import *
7
+ from PIL import Image
8
+ import torchvision.transforms as transforms
9
+
10
+ # RGBDD_Dataset data process
11
+
12
+ # S1: load data
13
+
14
+ lr_path = r"C:\Users\wuyuan\Downloads/model_out.png"
15
+
16
+ lr_pil = Image.open(lr_path)
17
+ import pdb;pdb.set_trace()
test_img/RGB-D-D/20200518160957_LR_fill_depth.png ADDED
test_img/RGB-D-D/20200518160957_RGB.jpg ADDED
test_img/TOFDSR/2020_09_08_13_59_59_435_rgb_depth_crop_fill.png ADDED
test_img/TOFDSR/2020_09_08_13_59_59_435_rgb_rgb_crop.png ADDED

Git LFS Details

  • SHA256: caac13848f252ad22044e7a4789595f150f939a4c94e7595b2dce7ed1d8330a0
  • Pointer size: 131 Bytes
  • Size of remote file: 236 kB
test_nyu_rgbdd.py ADDED
@@ -0,0 +1,103 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import argparse
2
+
3
+ from utils import *
4
+ import torchvision.transforms as transforms
5
+
6
+ from net.dornet import Net
7
+ from torch.utils.data import Dataset, DataLoader
8
+ from data.nyu_dataloader import *
9
+ from data.rgbdd_dataloader import *
10
+ # from data.tofsr_dataloader import *
11
+
12
+ import os
13
+
14
+ import torch
15
+
16
+ parser = argparse.ArgumentParser()
17
+ parser.add_argument('--scale', type=int, default=4, help='scale factor')
18
+ parser.add_argument("--root_dir", type=str, default='./dataset/RGB-D-D', help="root dir of dataset")
19
+ parser.add_argument("--model_dir", type=str, default="./checkpoints/RGBDD.pth", help="path of net")
20
+ parser.add_argument("--results_dir", type=str, default='./results/', help="root dir of results")
21
+ parser.add_argument('--tiny_model', action='store_true', help='tiny model')
22
+ parser.add_argument("--blur_sigma", type=int, default=3.6, help="blur_sigma")
23
+ parser.add_argument('--isNoisy', action='store_true', help='Noisy')
24
+
25
+ opt = parser.parse_args()
26
+
27
+ net = Net(tiny_model=True).cuda()
28
+
29
+ print("*********************************************")
30
+ print(sum(p.numel() for p in net.parameters() if p.requires_grad))
31
+ print("*********************************************")
32
+ net.load_state_dict(torch.load(opt.model_dir, map_location='cuda:0'))
33
+ device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
34
+ net.to(device)
35
+
36
+ data_transform = transforms.Compose([transforms.ToTensor()])
37
+
38
+ dataset_name = opt.root_dir.split('/')[-1]
39
+
40
+ if dataset_name == 'RGB-D-D':
41
+ dataset = RGBDD_Dataset(root_dir=opt.root_dir, scale=opt.scale, downsample='real', train=False,
42
+ transform=data_transform, isNoisy=opt.isNoisy, blur_sigma=opt.blur_sigma)
43
+ rmse = np.zeros(405)
44
+ elif dataset_name == 'NYU-v2':
45
+ dataset = NYU_v2_datset(root_dir=opt.root_dir, scale=opt.scale, transform=data_transform, train=False)
46
+ test_minmax = np.load('%s/test_minmax.npy' % opt.root_dir)
47
+ rmse = np.zeros(449)
48
+
49
+ dataloader = DataLoader(dataset, batch_size=1, shuffle=False, num_workers=8)
50
+ data_num = len(dataloader)
51
+
52
+ with torch.no_grad():
53
+ net.eval()
54
+ if dataset_name == 'RGB-D-D':
55
+ for idx, data in enumerate(dataloader):
56
+ guidance, lr, gt, maxx, minn, name = data['guidance'].cuda(), data['lr'].cuda(), data['gt'].cuda(), data[
57
+ 'max'].cuda(), data['min'].cuda(), data['name']
58
+ out = net(x_query=lr, rgb=guidance)
59
+ rmse[idx] = rgbdd_calc_rmse(gt[0, 0], out[0, 0], [maxx, minn])
60
+
61
+ # Save results (Save the output depth map)
62
+ # path_output = '{}/output'.format(opt.results_dir)
63
+ # os.makedirs(path_output, exist_ok=True)
64
+ # path_save_pred = '{}/{}.png'.format(path_output, name[0])
65
+
66
+ # pred = out[0, 0] * (maxx - minn) + minn
67
+ # pred = pred.cpu().detach().numpy()
68
+ # pred = pred.astype(np.uint16)
69
+ # pred = Image.fromarray(pred)
70
+ # pred.save(path_save_pred)
71
+
72
+ print('idx:%d RMSE:%f' % (idx + 1, rmse[idx]))
73
+ print("==========RGB-D-D=========")
74
+ print(rmse.mean())
75
+ print("==========RGB-D-D=========")
76
+ elif dataset_name == 'NYU-v2':
77
+ # t = np.zeros(449)
78
+ for idx, data in enumerate(dataloader):
79
+ guidance, lr, gt = data['guidance'].cuda(), data['lr'].cuda(), data['gt'].cuda()
80
+ out = net(x_query=lr, rgb=guidance)
81
+
82
+ minmax = test_minmax[:, idx]
83
+ minmax = torch.from_numpy(minmax).cuda()
84
+ rmse[idx] = calc_rmse(gt[0, 0], out[0, 0], minmax)
85
+
86
+ # Save results (Save the output depth map)
87
+ # path_output = '{}/output'.format(opt.results_dir)
88
+ # os.makedirs(path_output, exist_ok=True)
89
+ # path_save_pred = '{}/{:010d}.png'.format(path_output, idx)
90
+
91
+ # pred = out[0,0] * (minmax[0] - minmax[1]) + minmax[1]
92
+ # pred = pred * 1000.0
93
+ # pred = pred.cpu().detach().numpy()
94
+ # pred = pred.astype(np.uint16)
95
+ # pred = Image.fromarray(pred)
96
+ # pred.save(path_save_pred)
97
+
98
+ print('idx:%d RMSE:%f' % (idx + 1, rmse[idx]))
99
+ print("=========NYU-v2==========")
100
+ print(rmse.mean())
101
+ print("=========NYU-v2==========")
102
+
103
+
test_tofdsr.py ADDED
@@ -0,0 +1,66 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import argparse
2
+ import numpy as np
3
+ from utils import *
4
+ import torchvision.transforms as transforms
5
+
6
+ from net.dornet_ddp import Net
7
+
8
+ from data.tofdc_dataloader import *
9
+
10
+ import os
11
+
12
+ import torch
13
+
14
+ parser = argparse.ArgumentParser()
15
+ parser.add_argument('--scale', type=int, default=4, help='scale factor')
16
+ parser.add_argument("--root_dir", type=str, default='/opt/data/private/dataset', help="root dir of dataset")
17
+ parser.add_argument("--model_dir", type=str, default="./checkpoints/TOFDSR.pth", help="path of net")
18
+ parser.add_argument("--results_dir", type=str, default='./results/', help="root dir of results")
19
+ parser.add_argument('--tiny_model', action='store_true', help='tiny model')
20
+ parser.add_argument("--blur_sigma", type=int, default=3.6, help="blur_sigma")
21
+ parser.add_argument('--isNoisy', action='store_true', help='Noisy')
22
+
23
+ opt = parser.parse_args()
24
+
25
+ net = Net(tiny_model=opt.tiny_model).srn.cuda()
26
+
27
+ net.load_state_dict(torch.load(opt.model_dir, map_location='cuda:0'))
28
+ device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
29
+ net.to(device)
30
+
31
+ data_transform = transforms.Compose([transforms.ToTensor()])
32
+
33
+ dataset_name = opt.root_dir.split('/')[-1]
34
+
35
+ dataset = TOFDSR_Dataset(root_dir=opt.root_dir, train=False, txt_file="./data/TOFDSR_Test.txt", transform=data_transform, isNoisy=opt.isNoisy, blur_sigma=opt.blur_sigma)
36
+ dataloader = DataLoader(dataset, batch_size=1, shuffle=False, num_workers=8)
37
+
38
+ data_num = len(dataloader)
39
+ rmse = np.zeros(data_num)
40
+
41
+ with torch.no_grad():
42
+ net.eval()
43
+
44
+ for idx, data in enumerate(dataloader):
45
+ guidance, lr, gt, maxx, minn, name = data['guidance'].cuda(), data['lr'].cuda(), data['gt'].cuda(), data[
46
+ 'max'].cuda(), data['min'].cuda(), data['name']
47
+ out, _ = net(x_query=lr, rgb=guidance)
48
+ rmse[idx] = tofdsr_calc_rmse(gt[0, 0], out[0, 0], [maxx, minn])
49
+
50
+ # Save results (Save the output depth map)
51
+ # path_output = '{}/output'.format(opt.results_dir)
52
+ # os.makedirs(path_output, exist_ok=True)
53
+ # path_save_pred = '{}/{}.png'.format(path_output, name[0])
54
+
55
+ # pred = out[0, 0] * (maxx - minn) + minn
56
+ # pred = pred.cpu().detach().numpy()
57
+ # pred = pred.astype(np.uint16)
58
+ # pred = Image.fromarray(pred)
59
+ # pred.save(path_save_pred)
60
+
61
+ print('idx:%d RMSE:%f' % (idx + 1, rmse[idx]))
62
+ print("=========TOFDSR==========")
63
+ print(rmse.mean())
64
+ print("=========TOFDSR==========")
65
+
66
+
train_nyu_rgbdd.py ADDED
@@ -0,0 +1,158 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import argparse
2
+ from net.dornet import Net
3
+ from net.CR import *
4
+
5
+ from data.rgbdd_dataloader import *
6
+ from data.nyu_dataloader import *
7
+
8
+ from utils import calc_rmse, rgbdd_calc_rmse
9
+
10
+ from torch.utils.data import Dataset
11
+ from torchvision import transforms, utils
12
+ import torch
13
+ import torch.optim as optim
14
+ import torch.nn as nn
15
+
16
+ from tqdm import tqdm
17
+ import logging
18
+ from datetime import datetime
19
+ import os
20
+
21
+ import numpy as np
22
+
23
+ parser = argparse.ArgumentParser()
24
+ parser.add_argument('--scale', type=int, default=4, help='scale factor')
25
+ parser.add_argument('--lr', default='0.0001', type=float, help='learning rate')
26
+ parser.add_argument('--result', default='experiment', help='learning rate')
27
+ parser.add_argument('--tiny_model', action='store_true', help='tiny model')
28
+ parser.add_argument('--epoch', default=300, type=int, help='max epoch')
29
+ parser.add_argument("--decay_iterations", type=list, default=[1.2e5, 2e5, 3.6e5],
30
+ help="steps to start lr decay")
31
+ parser.add_argument("--gamma", type=float, default=0.2, help="decay rate of learning rate")
32
+ parser.add_argument("--root_dir", type=str, default='./dataset/RGB-D-D', help="root dir of dataset")
33
+ parser.add_argument("--batch_size", type=int, default=3, help="batch_size of training dataloader")
34
+ parser.add_argument("--blur_sigma", type=int, default=3.6, help="blur_sigma")
35
+ parser.add_argument('--isNoisy', action='store_true', help='Noisy')
36
+
37
+ opt = parser.parse_args()
38
+ print(opt)
39
+
40
+ s = datetime.now().strftime('%Y%m%d%H%M%S')
41
+ dataset_name = opt.root_dir.split('/')[-1]
42
+ result_root = '%s/%s-lr_%s-s_%s-%s-b_%s' % (opt.result, s, opt.lr, opt.scale, dataset_name, opt.batch_size)
43
+ if not os.path.exists(result_root):
44
+ os.mkdir(result_root)
45
+
46
+ logging.basicConfig(filename='%s/train.log' % result_root, format='%(asctime)s %(message)s', level=logging.INFO)
47
+ logging.info(opt)
48
+
49
+ net = Net(tiny_model=opt.tiny_model).cuda()
50
+
51
+ print("**********************Parameters***********************")
52
+ print(sum(p.numel() for p in net.parameters() if p.requires_grad))
53
+ print("**********************Parameters***********************")
54
+ net.train()
55
+
56
+ optimizer = optim.Adam(net.parameters(), lr=opt.lr)
57
+ scheduler = optim.lr_scheduler.MultiStepLR(optimizer, milestones=opt.decay_iterations, gamma=opt.gamma)
58
+
59
+ CL = ContrastLoss(ablation=False)
60
+ l1 = nn.L1Loss().cuda()
61
+
62
+ data_transform = transforms.Compose([transforms.ToTensor()])
63
+
64
+
65
+ if dataset_name == 'RGB-D-D':
66
+ train_dataset = RGBDD_Dataset(root_dir=opt.root_dir, scale=opt.scale, downsample='real', train=True,
67
+ transform=data_transform, isNoisy=opt.isNoisy, blur_sigma=opt.blur_sigma)
68
+ test_dataset = RGBDD_Dataset(root_dir=opt.root_dir, scale=opt.scale, downsample='real', train=False,
69
+ transform=data_transform, isNoisy=opt.isNoisy, blur_sigma=opt.blur_sigma)
70
+
71
+ elif dataset_name == 'NYU-v2':
72
+ test_minmax = np.load('%s/test_minmax.npy' % opt.root_dir)
73
+ train_dataset = NYU_v2_datset(root_dir=opt.root_dir, scale=opt.scale, transform=data_transform, train=True)
74
+ test_dataset = NYU_v2_datset(root_dir=opt.root_dir, scale=opt.scale, transform=data_transform, train=False)
75
+
76
+
77
+ train_dataloader = torch.utils.data.DataLoader(train_dataset, batch_size=opt.batch_size, shuffle=True, num_workers=8)
78
+ test_dataloader = torch.utils.data.DataLoader(test_dataset, batch_size=1, shuffle=False, num_workers=8)
79
+
80
+ max_epoch = opt.epoch
81
+ num_train = len(train_dataloader)
82
+ best_rmse = 100.0
83
+ best_epoch = 0
84
+ for epoch in range(max_epoch):
85
+ # ---------
86
+ # Training
87
+ # ---------
88
+ net.train()
89
+ running_loss = 0.0
90
+
91
+ t = tqdm(iter(train_dataloader), leave=True, total=len(train_dataloader))
92
+
93
+ for idx, data in enumerate(t):
94
+ batches_done = num_train * epoch + idx
95
+ optimizer.zero_grad()
96
+ guidance, lr, gt = data['guidance'].cuda(), data['lr'].cuda(), data['gt'].cuda()
97
+
98
+ restored, d_lr_, aux_loss = net(x_query=lr, rgb=guidance)
99
+
100
+ rec_loss = l1(restored, gt)
101
+ da_loss = l1(d_lr_, lr)
102
+ cl_loss = CL(d_lr_,lr,restored)
103
+ loss = rec_loss + 0.1 * da_loss + 0.1 * cl_loss + aux_loss
104
+
105
+ loss.backward()
106
+ optimizer.step()
107
+ scheduler.step()
108
+ running_loss += loss.data.item()
109
+
110
+ t.set_description(
111
+ '[train epoch:%d] loss: Rec_loss:%.8f DA_loss:%.8f CL_loss:%.8f' % (epoch + 1, rec_loss.item(), da_loss.item(), cl_loss.item()))
112
+ t.refresh()
113
+
114
+ logging.info('epoch:%d iteration:%d running_loss:%.10f' % (epoch + 1, batches_done + 1, running_loss / num_train))
115
+
116
+
117
+ # -----------
118
+ # Validating
119
+ # -----------
120
+ with torch.no_grad():
121
+
122
+ net.eval()
123
+ if dataset_name == 'RGB-D-D':
124
+ rmse = np.zeros(405)
125
+ elif dataset_name == 'NYU-v2':
126
+ rmse = np.zeros(449)
127
+ t = tqdm(iter(test_dataloader), leave=True, total=len(test_dataloader))
128
+
129
+ for idx, data in enumerate(t):
130
+ if dataset_name == 'RGB-D-D':
131
+ guidance, lr, gt, max, min = data['guidance'].cuda(), data['lr'].cuda(), data['gt'].cuda(), data[
132
+ 'max'].cuda(), data['min'].cuda()
133
+ out = net(x_query=lr, rgb=guidance)
134
+ minmax = [max, min]
135
+ rmse[idx] = rgbdd_calc_rmse(gt[0, 0], out[0, 0], minmax)
136
+ t.set_description('[validate] rmse: %f' % rmse[:idx + 1].mean())
137
+ t.refresh()
138
+ elif dataset_name == 'NYU-v2':
139
+ guidance, lr, gt = data['guidance'].cuda(), data['lr'].cuda(), data['gt'].cuda()
140
+ out = net(x_query=lr, rgb=guidance)
141
+ minmax = test_minmax[:, idx]
142
+ minmax = torch.from_numpy(minmax).cuda()
143
+ rmse[idx] = calc_rmse(gt[0, 0], out[0, 0], minmax)
144
+ t.set_description('[validate] rmse: %f' % rmse[:idx + 1].mean())
145
+ t.refresh()
146
+ r_mean = rmse.mean()
147
+ if r_mean < best_rmse:
148
+ best_rmse = r_mean
149
+ best_epoch = epoch
150
+ torch.save(net.state_dict(),
151
+ os.path.join(result_root, "RMSE%f_8%d.pth" % (best_rmse, best_epoch + 1)))
152
+ logging.info(
153
+ '---------------------------------------------------------------------------------------------------------------------------')
154
+ logging.info('epoch:%d lr:%f-------mean_rmse:%f (BEST: %f @epoch%d)' % (
155
+ epoch + 1, scheduler.get_last_lr()[0], r_mean, best_rmse, best_epoch + 1))
156
+ logging.info(
157
+ '---------------------------------------------------------------------------------------------------------------------------')
158
+
train_tofdsr.py ADDED
@@ -0,0 +1,167 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import argparse
2
+
3
+ from net.dornet_ddp import Net
4
+
5
+ from data.tofsr_dataloader import *
6
+ from utils import tofdsr_calc_rmse
7
+
8
+ from torch.utils.data import Dataset, DataLoader
9
+ import torch.distributed as dist
10
+ from torch.nn.parallel import DistributedDataParallel
11
+ from torch.utils.data.distributed import DistributedSampler
12
+ from torchvision import transforms, utils
13
+ import torch.optim as optim
14
+
15
+ import random
16
+
17
+ from net.CR import *
18
+ from tqdm import tqdm
19
+ import logging
20
+ from datetime import datetime
21
+ import os
22
+
23
+ parser = argparse.ArgumentParser()
24
+
25
+ parser.add_argument("--local-rank", default=-1, type=int)
26
+
27
+ parser.add_argument('--scale', type=int, default=4, help='scale factor')
28
+ parser.add_argument('--lr', default='0.0002', type=float, help='learning rate') # 0.0001
29
+ parser.add_argument('--tiny_model', action='store_true', help='tiny model')
30
+ parser.add_argument('--epoch', default=300, type=int, help='max epoch')
31
+ parser.add_argument('--device', default="0,1", type=str, help='which gpu use')
32
+ parser.add_argument("--decay_iterations", type=list, default=[1.2e5, 2e5, 3.6e5],
33
+ help="steps to start lr decay")
34
+ parser.add_argument("--gamma", type=float, default=0.2, help="decay rate of learning rate")
35
+ parser.add_argument("--root_dir", type=str, default='./dataset/TOFDSR', help="root dir of dataset")
36
+ parser.add_argument("--batchsize", type=int, default=3, help="batchsize of training dataloader")
37
+ parser.add_argument("--num_gpus", type=int, default=2, help="num_gpus")
38
+ parser.add_argument('--seed', type=int, default=7240, help='random seed point')
39
+ parser.add_argument("--result_root", type=str, default='experiment/TOFDSR', help="root dir of dataset")
40
+ parser.add_argument("--blur_sigma", type=int, default=3.6, help="blur_sigma")
41
+ parser.add_argument('--isNoisy', action='store_true', help='Noisy')
42
+
43
+ opt = parser.parse_args()
44
+ print(opt)
45
+
46
+ os.environ["CUDA_VISIBLE_DEVICES"] = "0,1"
47
+
48
+ torch.manual_seed(opt.seed)
49
+ np.random.seed(opt.seed)
50
+ random.seed(opt.seed)
51
+ torch.cuda.manual_seed_all(opt.seed)
52
+
53
+ local_rank = int(os.environ["LOCAL_RANK"])
54
+ torch.cuda.set_device(local_rank)
55
+ dist.init_process_group(backend='nccl')
56
+ device = torch.device("cuda", local_rank)
57
+
58
+ s = datetime.now().strftime('%Y%m%d%H%M%S')
59
+ dataset_name = opt.root_dir.split('/')[-1]
60
+
61
+ rank = dist.get_rank()
62
+
63
+ logging.basicConfig(filename='%s/train.log' % opt.result_root, format='%(asctime)s %(message)s', level=logging.INFO)
64
+ logging.info(opt)
65
+
66
+ net = Net(tiny_model=opt.tiny_model).cuda()
67
+
68
+ data_transform = transforms.Compose([transforms.ToTensor()])
69
+
70
+ train_dataset = TOFDSR_Dataset(root_dir=opt.root_dir, train=True, txt_file="./data/TOFDSR_Train.txt", transform=data_transform,
71
+ isNoisy=opt.isNoisy, blur_sigma=opt.blur_sigma)
72
+ test_dataset = TOFDSR_Dataset(root_dir=opt.root_dir, train=False, txt_file="./data/TOFDSR_Test.txt", transform=data_transform,
73
+ isNoisy=opt.isNoisy, blur_sigma=opt.blur_sigma)
74
+
75
+ if torch.cuda.device_count() > 1:
76
+ train_sampler = DistributedSampler(dataset=train_dataset)
77
+ train_dataloader = DataLoader(train_dataset, batch_size=opt.batchsize, shuffle=False, pin_memory=True, num_workers=8,
78
+ drop_last=True, sampler=train_sampler)
79
+ test_dataloader = DataLoader(test_dataset, batch_size=1, shuffle=False, pin_memory=True, num_workers=8)
80
+
81
+ net = DistributedDataParallel(net, device_ids=[local_rank], output_device=int(local_rank), find_unused_parameters=True)
82
+
83
+ l1 = nn.L1Loss().to(device)
84
+
85
+ optimizer = optim.Adam(net.module.parameters(), lr=opt.lr)
86
+ scheduler = optim.lr_scheduler.MultiStepLR(optimizer, milestones=opt.decay_iterations, gamma=opt.gamma)
87
+ net.train()
88
+
89
+ max_epoch = opt.epoch
90
+ num_train = len(train_dataloader)
91
+ best_rmse = 100.0
92
+ best_epoch = 0
93
+ for epoch in range(max_epoch):
94
+ # ---------
95
+ # Training
96
+ # ---------
97
+ train_sampler.set_epoch(epoch)
98
+ net.train()
99
+ running_loss = 0.0
100
+
101
+ t = tqdm(iter(train_dataloader), leave=True, total=len(train_dataloader))
102
+
103
+ for idx, data in enumerate(t):
104
+ batches_done = num_train * epoch + idx
105
+ optimizer.zero_grad()
106
+ guidance, lr, gt = data['guidance'].to(device), data['lr'].to(device), data['gt'].to(device)
107
+
108
+ restored, d_lr_, aux_loss, cl_loss = net(x_query=lr, rgb=guidance)
109
+
110
+ mask = (gt >= 0.02) & (gt <= 1)
111
+ gt = gt[mask]
112
+ restored = restored[mask]
113
+ lr = lr[mask]
114
+ d_lr_ = d_lr_[mask]
115
+
116
+ rec_loss = l1(restored, gt)
117
+ da_loss = l1(d_lr_, lr)
118
+
119
+ loss = rec_loss + 0.1 * da_loss + 0.1 * cl_loss + aux_loss
120
+ loss.backward()
121
+ optimizer.step()
122
+ scheduler.step()
123
+ running_loss += loss.data.item()
124
+ running_loss_50 = running_loss
125
+
126
+ if idx % 50 == 0:
127
+ running_loss_50 /= 50
128
+ t.set_description(
129
+ '[train epoch:%d] loss: Rec_loss:%.8f DA_loss:%.8f CL_loss:%.8f' % (
130
+ epoch + 1, rec_loss.item(), da_loss.item(), cl_loss.item()))
131
+ t.refresh()
132
+
133
+ logging.info('epoch:%d iteration:%d running_loss:%.10f' % (epoch + 1, batches_done + 1, running_loss / num_train))
134
+
135
+ # -----------
136
+ # Validating
137
+ # -----------
138
+ if rank == 0:
139
+ with torch.no_grad():
140
+
141
+ net.eval()
142
+ rmse = np.zeros(560)
143
+ t = tqdm(iter(test_dataloader), leave=True, total=len(test_dataloader))
144
+
145
+ for idx, data in enumerate(t):
146
+ guidance, lr, gt, maxx, minn = data['guidance'].to(device), data['lr'].to(device), data['gt'].to(
147
+ device), data[
148
+ 'max'].to(device), data['min'].to(device)
149
+ out, _ = net.module.srn(x_query=lr, rgb=guidance)
150
+ minmax = [maxx, minn]
151
+ rmse[idx] = tofdsr_calc_rmse(gt[0, 0], out[0, 0], minmax)
152
+ t.set_description('[validate] rmse: %f' % rmse[:idx + 1].mean())
153
+ t.refresh()
154
+
155
+ r_mean = rmse.mean()
156
+ if r_mean < best_rmse:
157
+ best_rmse = r_mean
158
+ best_epoch = epoch
159
+ torch.save(net.module.srn.state_dict(),
160
+ os.path.join(opt.result_root, "RMSE%f_8%d.pth" % (best_rmse, best_epoch + 1)))
161
+ logging.info(
162
+ '---------------------------------------------------------------------------------------------------------------------------')
163
+ logging.info('epoch:%d lr:%f-------mean_rmse:%f (BEST: %f @epoch%d)' % (
164
+ epoch + 1, scheduler.get_last_lr()[0], r_mean, best_rmse, best_epoch + 1))
165
+ logging.info(
166
+ '---------------------------------------------------------------------------------------------------------------------------')
167
+
utils.py ADDED
@@ -0,0 +1,37 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import torch
2
+
3
+ def calc_rmse(a, b, minmax):
4
+ a = a[6:-6, 6:-6]
5
+ b = b[6:-6, 6:-6]
6
+
7
+ a = a*(minmax[0]-minmax[1]) + minmax[1]
8
+ b = b*(minmax[0]-minmax[1]) + minmax[1]
9
+ a = a * 100
10
+ b = b * 100
11
+
12
+ return torch.sqrt(torch.mean(torch.pow(a-b,2)))
13
+
14
+
15
+ def rgbdd_calc_rmse(gt, out, minmax):
16
+ gt = gt[6:-6, 6:-6]
17
+ out = out[6:-6, 6:-6]
18
+
19
+ out = out*(minmax[0]-minmax[1]) + minmax[1]
20
+ gt = gt / 10.0
21
+ out = out / 10.0
22
+
23
+ return torch.sqrt(torch.mean(torch.pow(gt-out,2)))
24
+
25
+ def tofdsr_calc_rmse(gt, out, minmax):
26
+ gt = gt[6:-6, 6:-6]
27
+ out = out[6:-6, 6:-6]
28
+
29
+ mask = (gt >= 100) & (gt <= 5000)
30
+ gt = gt[mask]
31
+ out = out[mask]
32
+
33
+ out = out*(minmax[0]-minmax[1]) + minmax[1]
34
+ gt = gt / 10.0
35
+ out = out / 10.0
36
+
37
+ return torch.sqrt(torch.mean(torch.pow(gt-out,2)))