Hugging Face's logo

qualcomm
/
Segformer-Base

Image Segmentation
PyTorch
LiteRT
ONNX
real_time
android
Model card Files Community
qaihm-bot commited on
Commit
183cd95
·
verified ·
1 Parent(s): 39306d1

v0.30.5

Browse files

See https://github.com/quic/ai-hub-models/releases/v0.30.5 for changelog.

Files changed (6) hide show
  1. README.md +303 -0
  2. Segformer-Base.onnx +3 -0
  3. Segformer-Base.tflite +3 -0
  4. Segformer-Base_w8a16.onnx +3 -0
  5. Segformer-Base_w8a8.onnx +3 -0
  6. Segformer-Base_w8a8.tflite +3 -0
README.md ADDED
@@ -0,0 +1,303 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: pytorch
3
+ license: other
4
+ tags:
5
+ - real_time
6
+ - android
7
+ pipeline_tag: image-segmentation
8
+
9
+ ---
10
+
11
+ ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/segformer_base/web-assets/model_demo.png)
12
+
13
+ # Segformer-Base: Optimized for Mobile Deployment
14
+ ## Real-time object segmentation
15
+
16
+
17
+ Segformer Base is a machine learning model that predicts masks and classes of objects in an image.
18
+
19
+ This model is an implementation of Segformer-Base found [here](https://github.com/huggingface/transformers/tree/main/src/transformers/models/segformer).
20
+
21
+
22
+ This repository provides scripts to run Segformer-Base on Qualcomm® devices.
23
+ More details on model performance across various devices, can be found
24
+ [here](https://aihub.qualcomm.com/models/segformer_base).
25
+
26
+
27
+ ### Model Details
28
+
29
+ - **Model Type:** Model_use_case.semantic_segmentation
30
+ - **Model Stats:**
31
+ - Model checkpoint: nvidia/segformer-b0-finetuned-ade-512-512
32
+ - Input resolution: 512x512
33
+ - Number of parameters: 3.7M
34
+ - Model size: 14.4 MB
35
+ - Number of output classes: 150
36
+
37
+ | Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model
38
+ |---|---|---|---|---|---|---|---|---|
39
+ | Segformer-Base | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 315.099 ms | 9 - 65 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base.tflite) |
40
+ | Segformer-Base | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN | 321.201 ms | 2 - 11 MB | NPU | Use Export Script |
41
+ | Segformer-Base | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 114.238 ms | 9 - 66 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base.tflite) |
42
+ | Segformer-Base | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN | 121.029 ms | 3 - 55 MB | NPU | Use Export Script |
43
+ | Segformer-Base | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 115.018 ms | 0 - 36 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base.tflite) |
44
+ | Segformer-Base | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN | 116.65 ms | 3 - 6 MB | NPU | Use Export Script |
45
+ | Segformer-Base | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 117.193 ms | 9 - 66 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base.tflite) |
46
+ | Segformer-Base | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN | 119.038 ms | 1 - 10 MB | NPU | Use Export Script |
47
+ | Segformer-Base | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 315.099 ms | 9 - 65 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base.tflite) |
48
+ | Segformer-Base | float | SA7255P ADP | Qualcomm® SA7255P | QNN | 321.201 ms | 2 - 11 MB | NPU | Use Export Script |
49
+ | Segformer-Base | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 114.665 ms | 0 - 38 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base.tflite) |
50
+ | Segformer-Base | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN | 116.561 ms | 3 - 6 MB | NPU | Use Export Script |
51
+ | Segformer-Base | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 125.933 ms | 9 - 61 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base.tflite) |
52
+ | Segformer-Base | float | SA8295P ADP | Qualcomm® SA8295P | QNN | 130.6 ms | 0 - 17 MB | NPU | Use Export Script |
53
+ | Segformer-Base | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 114.784 ms | 0 - 33 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base.tflite) |
54
+ | Segformer-Base | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN | 116.849 ms | 3 - 5 MB | NPU | Use Export Script |
55
+ | Segformer-Base | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 117.193 ms | 9 - 66 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base.tflite) |
56
+ | Segformer-Base | float | SA8775P ADP | Qualcomm® SA8775P | QNN | 119.038 ms | 1 - 10 MB | NPU | Use Export Script |
57
+ | Segformer-Base | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 114.86 ms | 0 - 36 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base.tflite) |
58
+ | Segformer-Base | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN | 116.832 ms | 3 - 18 MB | NPU | Use Export Script |
59
+ | Segformer-Base | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 111.55 ms | 19 - 52 MB | NPU | [Segformer-Base.onnx](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base.onnx) |
60
+ | Segformer-Base | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 86.521 ms | 9 - 75 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base.tflite) |
61
+ | Segformer-Base | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN | 87.898 ms | 3 - 62 MB | NPU | Use Export Script |
62
+ | Segformer-Base | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 83.731 ms | 23 - 81 MB | NPU | [Segformer-Base.onnx](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base.onnx) |
63
+ | Segformer-Base | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 93.336 ms | 8 - 59 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base.tflite) |
64
+ | Segformer-Base | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN | 97.56 ms | 3 - 53 MB | NPU | Use Export Script |
65
+ | Segformer-Base | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 94.304 ms | 23 - 70 MB | NPU | [Segformer-Base.onnx](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base.onnx) |
66
+ | Segformer-Base | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 121.73 ms | 3 - 3 MB | NPU | Use Export Script |
67
+ | Segformer-Base | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 118.315 ms | 35 - 35 MB | NPU | [Segformer-Base.onnx](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base.onnx) |
68
+ | Segformer-Base | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 40.96 ms | 31 - 99 MB | NPU | [Segformer-Base.onnx](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a16.onnx) |
69
+ | Segformer-Base | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 28.528 ms | 28 - 288 MB | NPU | [Segformer-Base.onnx](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a16.onnx) |
70
+ | Segformer-Base | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 31.159 ms | 27 - 239 MB | NPU | [Segformer-Base.onnx](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a16.onnx) |
71
+ | Segformer-Base | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 44.24 ms | 58 - 58 MB | NPU | [Segformer-Base.onnx](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a16.onnx) |
72
+ | Segformer-Base | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 35.944 ms | 2 - 41 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a8.tflite) |
73
+ | Segformer-Base | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN | 25.442 ms | 1 - 10 MB | NPU | Use Export Script |
74
+ | Segformer-Base | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 13.76 ms | 2 - 49 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a8.tflite) |
75
+ | Segformer-Base | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN | 9.382 ms | 1 - 44 MB | NPU | Use Export Script |
76
+ | Segformer-Base | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 13.209 ms | 2 - 14 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a8.tflite) |
77
+ | Segformer-Base | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN | 7.852 ms | 1 - 4 MB | NPU | Use Export Script |
78
+ | Segformer-Base | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 13.755 ms | 2 - 40 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a8.tflite) |
79
+ | Segformer-Base | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN | 8.462 ms | 1 - 13 MB | NPU | Use Export Script |
80
+ | Segformer-Base | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | TFLITE | 122.719 ms | 14 - 55 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a8.tflite) |
81
+ | Segformer-Base | w8a8 | RB3 Gen 2 (Proxy) | Qualcomm® QCS6490 (Proxy) | QNN | 33.447 ms | 1 - 13 MB | NPU | Use Export Script |
82
+ | Segformer-Base | w8a8 | RB5 (Proxy) | Qualcomm® QCS8250 (Proxy) | TFLITE | 415.858 ms | 50 - 68 MB | GPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a8.tflite) |
83
+ | Segformer-Base | w8a8 | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 35.944 ms | 2 - 41 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a8.tflite) |
84
+ | Segformer-Base | w8a8 | SA7255P ADP | Qualcomm® SA7255P | QNN | 25.442 ms | 1 - 10 MB | NPU | Use Export Script |
85
+ | Segformer-Base | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 13.24 ms | 2 - 15 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a8.tflite) |
86
+ | Segformer-Base | w8a8 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN | 7.806 ms | 1 - 4 MB | NPU | Use Export Script |
87
+ | Segformer-Base | w8a8 | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 15.501 ms | 2 - 42 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a8.tflite) |
88
+ | Segformer-Base | w8a8 | SA8295P ADP | Qualcomm® SA8295P | QNN | 9.948 ms | 1 - 18 MB | NPU | Use Export Script |
89
+ | Segformer-Base | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 13.25 ms | 2 - 15 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a8.tflite) |
90
+ | Segformer-Base | w8a8 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN | 7.81 ms | 1 - 3 MB | NPU | Use Export Script |
91
+ | Segformer-Base | w8a8 | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 13.755 ms | 2 - 40 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a8.tflite) |
92
+ | Segformer-Base | w8a8 | SA8775P ADP | Qualcomm® SA8775P | QNN | 8.462 ms | 1 - 13 MB | NPU | Use Export Script |
93
+ | Segformer-Base | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 13.244 ms | 2 - 15 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a8.tflite) |
94
+ | Segformer-Base | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN | 7.828 ms | 1 - 12 MB | NPU | Use Export Script |
95
+ | Segformer-Base | w8a8 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | ONNX | 17.979 ms | 3 - 25 MB | NPU | [Segformer-Base.onnx](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a8.onnx) |
96
+ | Segformer-Base | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 9.065 ms | 2 - 52 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a8.tflite) |
97
+ | Segformer-Base | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN | 5.255 ms | 1 - 61 MB | NPU | Use Export Script |
98
+ | Segformer-Base | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 12.47 ms | 5 - 66 MB | NPU | [Segformer-Base.onnx](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a8.onnx) |
99
+ | Segformer-Base | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 9.279 ms | 2 - 43 MB | NPU | [Segformer-Base.tflite](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a8.tflite) |
100
+ | Segformer-Base | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN | 4.788 ms | 1 - 53 MB | NPU | Use Export Script |
101
+ | Segformer-Base | w8a8 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | ONNX | 11.956 ms | 5 - 56 MB | NPU | [Segformer-Base.onnx](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a8.onnx) |
102
+ | Segformer-Base | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN | 8.692 ms | 1 - 1 MB | NPU | Use Export Script |
103
+ | Segformer-Base | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 19.372 ms | 9 - 9 MB | NPU | [Segformer-Base.onnx](https://huggingface.co/qualcomm/Segformer-Base/blob/main/Segformer-Base_w8a8.onnx) |
104
+
105
+
106
+
107
+
108
+ ## Installation
109
+
110
+
111
+ Install the package via pip:
112
+ ```bash
113
+ pip install qai-hub-models
114
+ ```
115
+
116
+
117
+ ## Configure Qualcomm® AI Hub to run this model on a cloud-hosted device
118
+
119
+ Sign-in to [Qualcomm® AI Hub](https://app.aihub.qualcomm.com/) with your
120
+ Qualcomm® ID. Once signed in navigate to `Account -> Settings -> API Token`.
121
+
122
+ With this API token, you can configure your client to run models on the cloud
123
+ hosted devices.
124
+ ```bash
125
+ qai-hub configure --api_token API_TOKEN
126
+ ```
127
+ Navigate to [docs](https://app.aihub.qualcomm.com/docs/) for more information.
128
+
129
+
130
+
131
+ ## Demo off target
132
+
133
+ The package contains a simple end-to-end demo that downloads pre-trained
134
+ weights and runs this model on a sample input.
135
+
136
+ ```bash
137
+ python -m qai_hub_models.models.segformer_base.demo
138
+ ```
139
+
140
+ The above demo runs a reference implementation of pre-processing, model
141
+ inference, and post processing.
142
+
143
+ **NOTE**: If you want running in a Jupyter Notebook or Google Colab like
144
+ environment, please add the following to your cell (instead of the above).
145
+ ```
146
+ %run -m qai_hub_models.models.segformer_base.demo
147
+ ```
148
+
149
+
150
+ ### Run model on a cloud-hosted device
151
+
152
+ In addition to the demo, you can also run the model on a cloud-hosted Qualcomm®
153
+ device. This script does the following:
154
+ * Performance check on-device on a cloud-hosted device
155
+ * Downloads compiled assets that can be deployed on-device for Android.
156
+ * Accuracy check between PyTorch and on-device outputs.
157
+
158
+ ```bash
159
+ python -m qai_hub_models.models.segformer_base.export
160
+ ```
161
+ ```
162
+ Profiling Results
163
+ ------------------------------------------------------------
164
+ Segformer-Base
165
+ Device : cs_8275 (ANDROID 14)
166
+ Runtime : TFLITE
167
+ Estimated inference time (ms) : 315.1
168
+ Estimated peak memory usage (MB): [9, 65]
169
+ Total # Ops : 545
170
+ Compute Unit(s) : npu (545 ops) gpu (0 ops) cpu (0 ops)
171
+ ```
172
+
173
+
174
+ ## How does this work?
175
+
176
+ This [export script](https://aihub.qualcomm.com/models/segformer_base/qai_hub_models/models/Segformer-Base/export.py)
177
+ leverages [Qualcomm® AI Hub](https://aihub.qualcomm.com/) to optimize, validate, and deploy this model
178
+ on-device. Lets go through each step below in detail:
179
+
180
+ Step 1: **Compile model for on-device deployment**
181
+
182
+ To compile a PyTorch model for on-device deployment, we first trace the model
183
+ in memory using the `jit.trace` and then call the `submit_compile_job` API.
184
+
185
+ ```python
186
+ import torch
187
+
188
+ import qai_hub as hub
189
+ from qai_hub_models.models.segformer_base import Model
190
+
191
+ # Load the model
192
+ torch_model = Model.from_pretrained()
193
+
194
+ # Device
195
+ device = hub.Device("Samsung Galaxy S24")
196
+
197
+ # Trace model
198
+ input_shape = torch_model.get_input_spec()
199
+ sample_inputs = torch_model.sample_inputs()
200
+
201
+ pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()])
202
+
203
+ # Compile model on a specific device
204
+ compile_job = hub.submit_compile_job(
205
+ model=pt_model,
206
+ device=device,
207
+ input_specs=torch_model.get_input_spec(),
208
+ )
209
+
210
+ # Get target model to run on-device
211
+ target_model = compile_job.get_target_model()
212
+
213
+ ```
214
+
215
+
216
+ Step 2: **Performance profiling on cloud-hosted device**
217
+
218
+ After compiling models from step 1. Models can be profiled model on-device using the
219
+ `target_model`. Note that this scripts runs the model on a device automatically
220
+ provisioned in the cloud. Once the job is submitted, you can navigate to a
221
+ provided job URL to view a variety of on-device performance metrics.
222
+ ```python
223
+ profile_job = hub.submit_profile_job(
224
+ model=target_model,
225
+ device=device,
226
+ )
227
+
228
+ ```
229
+
230
+ Step 3: **Verify on-device accuracy**
231
+
232
+ To verify the accuracy of the model on-device, you can run on-device inference
233
+ on sample input data on the same cloud hosted device.
234
+ ```python
235
+ input_data = torch_model.sample_inputs()
236
+ inference_job = hub.submit_inference_job(
237
+ model=target_model,
238
+ device=device,
239
+ inputs=input_data,
240
+ )
241
+ on_device_output = inference_job.download_output_data()
242
+
243
+ ```
244
+ With the output of the model, you can compute like PSNR, relative errors or
245
+ spot check the output with expected output.
246
+
247
+ **Note**: This on-device profiling and inference requires access to Qualcomm®
248
+ AI Hub. [Sign up for access](https://myaccount.qualcomm.com/signup).
249
+
250
+
251
+
252
+ ## Run demo on a cloud-hosted device
253
+
254
+ You can also run the demo on-device.
255
+
256
+ ```bash
257
+ python -m qai_hub_models.models.segformer_base.demo --eval-mode on-device
258
+ ```
259
+
260
+ **NOTE**: If you want running in a Jupyter Notebook or Google Colab like
261
+ environment, please add the following to your cell (instead of the above).
262
+ ```
263
+ %run -m qai_hub_models.models.segformer_base.demo -- --eval-mode on-device
264
+ ```
265
+
266
+
267
+ ## Deploying compiled model to Android
268
+
269
+
270
+ The models can be deployed using multiple runtimes:
271
+ - TensorFlow Lite (`.tflite` export): [This
272
+ tutorial](https://www.tensorflow.org/lite/android/quickstart) provides a
273
+ guide to deploy the .tflite model in an Android application.
274
+
275
+
276
+ - QNN (`.so` export ): This [sample
277
+ app](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/sample_app.html)
278
+ provides instructions on how to use the `.so` shared library in an Android application.
279
+
280
+
281
+ ## View on Qualcomm® AI Hub
282
+ Get more details on Segformer-Base's performance across various devices [here](https://aihub.qualcomm.com/models/segformer_base).
283
+ Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/)
284
+
285
+
286
+ ## License
287
+ * The license for the original implementation of Segformer-Base can be found
288
+ [here](https://github.com/huggingface/transformers/blob/main/LICENSE).
289
+ * The license for the compiled assets for on-device deployment can be found [here](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/Qualcomm+AI+Hub+Proprietary+License.pdf)
290
+
291
+
292
+
293
+ ## References
294
+ * [SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers](https://arxiv.org/abs/2105.15203)
295
+ * [Source Model Implementation](https://github.com/huggingface/transformers/tree/main/src/transformers/models/segformer)
296
+
297
+
298
+
299
+ ## Community
300
+ * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
301
+ * For questions or feedback please [reach out to us](mailto:[email protected]).
302
+
303
+
Segformer-Base.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e7a91ccebd74cb978a659e02b127bb5ae4d1bf9970be76f9688f974721337336
3
+ size 15145277
Segformer-Base.tflite ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:add87dabfd682f27acc59d810a5e6055e09491a4769c8b96190dd77726a30ef4
3
+ size 15113856
Segformer-Base_w8a16.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:825a6b5fc08372f08925e151ec7dc9a87df66ab42277ec0e7b51056f026b87c7
3
+ size 15454331
Segformer-Base_w8a8.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7d98860398daeea7f59e1b0b518a17bbbdefb1cf227c989ac9871e0178f75bcf
3
+ size 15445355
Segformer-Base_w8a8.tflite ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ef433cde49eb9aeb6007d0826a4330b74109ea1acb0566e41480a9dc668c510c
3
+ size 4087088