Hugging Face's logo

STMicroelectronics
/
st_ssd_mobilenet_v1

Object Detection
Model card Files Community
FBAGSTM commited on
Commit
8161e20
·
verified ·
1 Parent(s): 25a8ab8

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +144 -6
README.md CHANGED
@@ -1,6 +1,144 @@
1
- ---
2
- license: other
3
- license_name: sla0044
4
- license_link: >-
5
- https://github.com/STMicroelectronics/stm32aimodelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/LICENSE.md
6
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ license_name: sla0044
4
+ license_link: >-
5
+ https://github.com/STMicroelectronics/stm32aimodelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/LICENSE.md
6
+ pipeline_tag: object-detection
7
+ ---
8
+ # SSD MobileNet v1 quantized
9
+
10
+ ## **Use case** : `Object detection`
11
+
12
+ # Model description
13
+
14
+
15
+ The mobilenet-ssd model is a Single-Shot multibox Detection (SSD) network intended to perform object detection.
16
+ Mobilenet-ssd is using MobileNet as a backbone which is a general architecture that can be used for multiple use cases.
17
+ Depending on the use case, it can use different input layer size and
18
+ different width factors. This allows different width models to reduce
19
+ the number of multiply-adds and thereby reduce inference cost on mobile devices.
20
+
21
+ The model is quantized in int8 using tensorflow lite converter.
22
+
23
+ ## Network information
24
+
25
+
26
+ | Network information | Value |
27
+ |-------------------------|-----------------|
28
+ | Framework | TensorFlow Lite |
29
+ | Quantization | int8 |
30
+ | Provenance | https://www.tensorflow.org/api_docs/python/tf/keras/applications/mobilenet |
31
+ | Paper | https://arxiv.org/abs/1704.04861, https://arxiv.org/abs/1512.02325 |
32
+
33
+ The models are quantized using tensorflow lite converter.
34
+
35
+
36
+ ## Network inputs / outputs
37
+
38
+
39
+ For an image resolution of NxM and NC classes
40
+
41
+ | Input Shape | Description |
42
+ | ----- | ----------- |
43
+ | (1, N, M, 3) | Single NxM RGB image with UINT8 values between 0 and 255 |
44
+
45
+ | Output Shape | Description |
46
+ | ----- | ----------- |
47
+ | (1, NA, 8 + NC) | FLOAT values Where NA is thge number of anchors and NC is the number of classes|
48
+
49
+
50
+ ## Recommended Platforms
51
+
52
+
53
+ | Platform | Supported | Recommended |
54
+ |----------|-----------|-------------|
55
+ | STM32L0 | [] | [] |
56
+ | STM32L4 | [] | [] |
57
+ | STM32U5 | [] | [] |
58
+ | STM32H7 | [x] | [x] |
59
+ | STM32MP1 | [x] | [x] |
60
+ | STM32MP2 | [x] | [x] |
61
+ | STM32N6 | [x] | [x] |
62
+
63
+ # Performances
64
+
65
+ ## Metrics
66
+
67
+ Measures are done with default STM32Cube.AI configuration with enabled input / output allocated option.
68
+ ### Reference **NPU** memory footprint based on COCO Person dataset (see Accuracy for details on dataset)
69
+ |Model | Dataset | Format | Resolution | Series | Internal RAM | External RAM | Weights Flash | STM32Cube.AI version | STEdgeAI Core version |
70
+ |----------|------------------|--------|-------------|------------------|------------------|---------------------|-------|----------------------|-------------------------|
71
+ |[ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_224/st_ssd_mobilenet_v1_025_224_int8.tflite) | COCO-Person | Int8 | 224x224x3 | STM32N6 | 694.64 | 0.0 | 827.16 | 10.0.0 | 2.0.0 |
72
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_256/st_ssd_mobilenet_v1_025_256_int8.tflite) | COCO-Person | Int8 | 256x256x3 | STM32N6 | 1002.64 | 0.0 | 826.91 | 10.0.0 | 2.0.0 |
73
+
74
+
75
+ ### Reference **NPU** inference time based on COCO Person dataset (see Accuracy for details on dataset)
76
+ | Model | Dataset | Format | Resolution | Board | Execution Engine | Inference time (ms) | Inf / sec | STM32Cube.AI version | STEdgeAI Core version |
77
+ |--------|------------------|--------|-------------|------------------|------------------|---------------------|-------|----------------------|-------------------------|
78
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_224/st_ssd_mobilenet_v1_025_224_int8.tflite) | COCO-Person | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 9.63 | 103.84 | 10.0.0 | 2.0.0 |
79
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_256/st_ssd_mobilenet_v1_025_256_int8.tflite) | COCO-Person | Int8 | 256x256x3 | STM32N6570-DK | NPU/MCU | 10.37 | 96.43 | 10.0.0 | 2.0.0 |
80
+
81
+
82
+ ### Reference MCU memory footprint based on COCO Person dataset (see Accuracy for details on dataset)
83
+ | Model | Format | Resolution | Series | Activation RAM | Runtime RAM | Weights Flash | Code Flash | Total RAM | Total Flash | STM32Cube.AI version |
84
+ |-------------------|--------|------------|---------|----------------|-------------|---------------|------------|-------------|-------------|-----------------------|
85
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_192/st_ssd_mobilenet_v1_025_192_int8.tflite) | Int8 | 192x192x3 | STM32H7 | 266.3 | 29.93 | 483.16 | 95.39 | 296.23 | 578.55 | 10.0.0 | |
86
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_224/st_ssd_mobilenet_v1_025_224_int8.tflite) | Int8 | 224x224x3 | STM32H7 | 379.6 | 34.34 | 675.64 | 106.01 | 413.94 | 781.65 | 10.0.0 | |
87
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_256/st_ssd_mobilenet_v1_025_256_int8.tflite) | Int8 | 256x256x3 | STM32H7 | 456.1 KiB | 33.75 | 675.64 | 105.26| 489.85 | 780.9 | 10.0.0 |
88
+
89
+
90
+ ### Reference **MCU** inference time based on COCO Person dataset (see Accuracy for details on dataset)
91
+
92
+
93
+ | Model | Format | Resolution | Board | Execution Engine | Frequency | Inference time (ms) | STM32Cube.AI version |
94
+ |-------------------|--------|------------|------------------|------------------|-------------|---------------------|-----------------------|
95
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_192/st_ssd_mobilenet_v1_025_192_int8.tflite) | Int8 | 192x192x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 149.18 ms | 10.0.0 |
96
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_224/st_ssd_mobilenet_v1_025_224_int8.tflite) | Int8 | 224x224x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 218.99 ms | 10.0.0 |
97
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_256/st_ssd_mobilenet_v1_025_256_int8.tflite) | Int8 | 256x256x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 267.18 ms | 10.0.0 |
98
+
99
+
100
+ ### Reference **MPU** inference time based on COCO Person dataset (see Accuracy for details on dataset)
101
+
102
+ | Model | Format | Resolution | Quantization | Board | Execution Engine | Frequency | Inference time (ms) | %NPU | %GPU | %CPU | X-LINUX-AI version | Framework |
103
+ |----------------------------------------------------------------------------------------------------------------------------------------------------|--------|------------|---------------|-------------------|------------------|-----------|---------------------|-------|-------|------|--------------------|-----------------------|
104
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_192/st_ssd_mobilenet_v1_025_192_int8.tflite) | Int8 | 192x192x3 | per-channel** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 12.34 ms | 15.35 | 84.65 |0 | v5.1.0 | OpenVX |
105
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_224/st_ssd_mobilenet_v1_025_224_int8.tflite) | Int8 | 224x224x3 | per-channel** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 18.65 ms | 14.02 | 85.98 |0 | v5.1.0 | OpenVX |
106
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_256/st_ssd_mobilenet_v1_025_256_int8.tflite) | Int8 | 256x256x3 | per-channel** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 14.33 ms | 14.12 | 85.88 |0 | v5.1.0 | OpenVX |
107
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_192/st_ssd_mobilenet_v1_025_192_int8.tflite) | Int8 | 192x192x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 67.80 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
108
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_224/st_ssd_mobilenet_v1_025_224_int8.tflite) | Int8 | 224x224x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 100.20 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
109
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_256/st_ssd_mobilenet_v1_025_256_int8.tflite) | Int8 | 256x256x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 119.00 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
110
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_192/st_ssd_mobilenet_v1_025_192_int8.tflite) | Int8 | 192x192x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 95.36 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
111
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_224/st_ssd_mobilenet_v1_025_224_int8.tflite) | Int8 | 224x224x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 139.00 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
112
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_256/st_ssd_mobilenet_v1_025_256_int8.tflite) | Int8 | 256x256x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 168.80 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
113
+
114
+ ** **To get the most out of MP25 NPU hardware acceleration, please use per-tensor quantization**
115
+
116
+ ### AP on COCO Person dataset
117
+
118
+
119
+ Dataset details: [link](https://cocodataset.org/#download) , License [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/legalcode) , Quotation[[1]](#1) , Number of classes: 80, Number of images: 118,287
120
+
121
+ | Model | Format | Resolution | AP* |
122
+ |-------|--------|------------|----------------|
123
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_192/st_ssd_mobilenet_v1_025_192_int8.tflite) | Int8 | 192x192x3 | 35.80 % |
124
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_192/st_ssd_mobilenet_v1_025_192.h5) | Float | 192x192x3 | 35.80 % |
125
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_224/st_ssd_mobilenet_v1_025_224_int8.tflite) | Int8 | 224x224x3 | 46.10 % |
126
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_224/st_ssd_mobilenet_v1_025_224.h5) | Float | 224x224x3 | 46.90 % |
127
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_256/st_ssd_mobilenet_v1_025_256_int8.tflite) | Int8 | 256x256x3 | 50.50 % |
128
+ | [ST SSD Mobilenet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_ssd_mobilenet_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_ssd_mobilenet_v1_025_256/st_ssd_mobilenet_v1_025_256.h5) | Float | 256x256x3 | 51 % |
129
+
130
+ \* EVAL_IOU = 0.4, NMS_THRESH = 0.5, SCORE_THRESH =0.001
131
+
132
+
133
+ ## Retraining and Integration in a simple example:
134
+
135
+ Please refer to the stm32ai-modelzoo-services GitHub [here](https://github.com/STMicroelectronics/stm32ai-modelzoo-services)
136
+
137
+
138
+
139
+ # References
140
+
141
+
142
+ <a id="1">[1]</a>
143
+ Lin, T.Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P. and Zitnick, C.L., 2014. "Microsoft coco: Common objects in context". In Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part V 13 (pp. 740-755). Springer International Publishing. [Online]. Available: https://cocodataset.org/#download.
144
+