Hugging Face's logo

STMicroelectronics
/
st_yolo_lc_v1

Object Detection
Model card Files Community
FBAGSTM commited on
Commit
a63a269
·
verified ·
1 Parent(s): 0caf7c8

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +164 -6
README.md CHANGED
@@ -1,6 +1,164 @@
1
- ---
2
- license: other
3
- license_name: sla0044
4
- license_link: >-
5
- https://github.com/STMicroelectronics/stm32aimodelzoo/object_detection/st_yolo_lc_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_yolo_lc_v1/ST_pretrainedmodel_public_dataset/LICENSE.md
6
+ pipeline_tag: object-detection
7
+ ---
8
+ # ST YOLO LC V1 quantized
9
+
10
+ ## **Use case** : `Object detection`
11
+
12
+ # Model description
13
+
14
+
15
+ ST Yolo LC v1 is a real-time object detection model targeted for real-time processing implemented in Tensorflow.
16
+
17
+ The model is quantized in int8 format using tensorflow lite converter.
18
+
19
+ ## Network information
20
+
21
+
22
+ | Network information | Value |
23
+ |-------------------------|-----------------|
24
+ | Framework | TensorFlow Lite |
25
+ | Quantization | int8 |
26
+ | Paper | https://pjreddie.com/media/files/papers/YOLO9000.pdf |
27
+
28
+ The models are quantized using tensorflow lite converter.
29
+
30
+
31
+ ## Network inputs / outputs
32
+
33
+
34
+ For an image resolution of NxM and NC classes
35
+
36
+ | Input Shape | Description |
37
+ | ----- | ----------- |
38
+ | (1, W, H, 3) | Single NxM RGB image with UINT8 values between 0 and 255 |
39
+
40
+ | Output Shape | Description |
41
+ | ----- | ----------- |
42
+ | (1, WxH, NAx(5+NC)) | FLOAT values Where WXH is the resolution of the output grid cell, NA is the number of anchors and NC is the number of classes|
43
+
44
+
45
+ ## Recommended Platforms
46
+
47
+
48
+ | Platform | Supported | Recommended |
49
+ |----------|-----------|-------------|
50
+ | STM32L0 | [] | [] |
51
+ | STM32L4 | [] | [] |
52
+ | STM32U5 | [] | [] |
53
+ | STM32H7 | [x] | [x] |
54
+ | STM32MP1 | [x] | [x] |
55
+ | STM32MP2 | [x] | [] |
56
+ | STM32N6 | [x] | [] |
57
+
58
+ # Performances
59
+
60
+ ## Metrics
61
+
62
+ Measures are done with default STM32Cube.AI configuration with enabled input / output allocated option.
63
+
64
+
65
+
66
+ ### Reference **NPU** memory footprint based on COCO Person dataset (see Accuracy for details on dataset)
67
+ |Model | Dataset | Format | Resolution | Series | Internal RAM (KiB) | External RAM (KiB)| Weights Flash (KiB)| STM32Cube.AI version | STEdgeAI Core version |
68
+ |----------|------------------|--------|-------------|------------------|------------------|---------------------|-------|----------------------|-------------------------|
69
+ | [st_yolo_lc_v1](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_lc_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_lc_v1_192/st_yolo_lc_v1_192_int8.tflite)| COCO-Person | Int8 | 192x192x3 | STM32N6 | 252 | 0.0 | 328.19 | 10.0.0 | 2.0.0 |
70
+ | [st_yolo_lc_v1](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_lc_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_lc_v1_224/st_yolo_lc_v1_224_int8.tflite)| COCO-Person | Int8 | 256x256x3 | STM32N6 | 343 | 0.0 | 328.19 | 10.0.0 | 2.0.0 |
71
+ | [st_yolo_lc_v1](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_lc_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_lc_v1_256/st_yolo_lc_v1_256_int8.tflite)| COCO-Person | Int8 | 256x256x3 | STM32N6 | 576 | 0.0 | 328.19 | 10.0.0 | 2.0.0 |
72
+
73
+
74
+ ### Reference **NPU** inference time based on COCO Person dataset (see Accuracy for details on dataset)
75
+ | Model | Dataset | Format | Resolution | Board | Execution Engine | Inference time (ms) | Inf / sec | STM32Cube.AI version | STEdgeAI Core version |
76
+ |--------|------------------|--------|-------------|------------------|------------------|---------------------|-------|----------------------|-------------------------|
77
+ | [st_yolo_lc_v1](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_lc_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_lc_v1_192/st_yolo_lc_v1_192_int8.tflite) | COCO-Person | Int8 | 192x192x3 | STM32N6570-DK | NPU/MCU | 1.96 | 510.20 | 10.0.0 | 2.0.0 |
78
+ | [st_yolo_lc_v1](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_lc_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_lc_v1_224/st_yolo_lc_v1_224_int8.tflite) | COCO-Person | Int8 | 256x256x3 | STM32N6570-DK | NPU/MCU | 2.35 | 425.53 | 10.0.0 | 2.0.0 |
79
+ | [st_yolo_lc_v1](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_lc_v1/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_lc_v1_256/st_yolo_lc_v1_256_int8.tflite) | COCO-Person | Int8 256x256x3 | STM32N6570-DK | NPU/MCU | 3.01 | 332.23 | 10.0.0 | 2.0.0 |
80
+
81
+ ### Reference **MCU** memory footprint based on COCO Person dataset (see Accuracy for details on dataset)
82
+
83
+
84
+ | Model | Format | Resolution | Series | Activation RAM | Runtime RAM | Weights Flash | Code Flash | Total RAM | Total Flash | STM32Cube.AI version |
85
+ |-------------------|--------|------------|---------|----------------|-------------|---------------|-----------------|--------------|-------------|-----------------------|
86
+ | st_yolo_lc_v1 | Int8 | 192x192x3 | STM32H7 | 166.29 KiB | 8.09 KiB | 276.73 KiB | 53.48 KiB | 174.38 KiB | 330.21 KiB | 10.0.0 |
87
+ | st_yolo_lc_v1 | Int8 | 224x224x3 | STM32H7 | 217.29 KiB | 8.09 KiB | 276.73 KiB | 53.48 KiB | 225.38 KiB | 330.21 KiB | 10.0.0 |
88
+ | st_yolo_lc_v1 | Int8 | 256x256x3 | STM32H7 | 278.29 KiB | 8.09 KiB | 276.73 KiB | 53.48 KiB | 286.38 KiB | 330.21 KiB | 10.0.0 |
89
+
90
+
91
+ ### Reference **MCU** inference time based on COCO Person dataset (see Accuracy for details on dataset)
92
+
93
+
94
+ | Model | Format | Resolution | Board | Execution Engine | Frequency | Inference time (ms) | STM32Cube.AI version |
95
+ |-------------------|--------|------------|------------------|------------------|-------------|---------------------|-----------------------|
96
+ | st_yolo_lc_v1 | Int8 | 192x192x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 179.01 | 10.0.0 |
97
+ | st_yolo_lc_v1 | Int8 | 224x224x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 244.7 | 10.0.0 |
98
+ | st_yolo_lc_v1 | Int8 | 256x256x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 321.38 | 10.0.0 |
99
+
100
+
101
+ ### Reference **MPU** inference time based on COCO Person dataset (see Accuracy for details on dataset)
102
+
103
+ | Model | Format | Resolution | Quantization | Board | Execution Engine | Frequency | Inference time (ms) | %NPU | %GPU | %CPU | X-LINUX-AI version | Framework |
104
+ |---------------|--------|------------|---------------|-------------------|------------------|-----------|---------------------|-------|-------|------|--------------------|-----------------------|
105
+ | st_yolo_lc_v1 | Int8 | 192x192x3 | per-channel** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 12.00 ms | 2.62 | 97.38 |0 | v5.1.0 | OpenVX |
106
+ | st_yolo_lc_v1 | Int8 | 224x224x3 | per-channel** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 17.92 ms | 2.43 | 97.57 |0 | v5.1.0 | OpenVX |
107
+ | st_yolo_lc_v1 | Int8 | 256x256x3 | per-channel** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 14.43 ms | 3.20 | 96.80 |0 | v5.1.0 | OpenVX |
108
+ | st_yolo_lc_v1 | Int8 | 192x192x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 32.84 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
109
+ | st_yolo_lc_v1 | Int8 | 224x224x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 45.13 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
110
+ | st_yolo_lc_v1 | Int8 | 256x256x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 59.38 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
111
+ | st_yolo_lc_v1 | Int8 | 192x192x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 52.64 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
112
+ | st_yolo_lc_v1 | Int8 | 224x224x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 71.26 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
113
+ | st_yolo_lc_v1 | Int8 | 256x256x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 93.50 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
114
+
115
+ ** **To get the most out of MP25 NPU hardware acceleration, please use per-tensor quantization**
116
+
117
+ ### AP on COCO Person dataset
118
+
119
+
120
+ 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
121
+
122
+ | Model | Format | Resolution | AP |
123
+ |-------|--------|------------|----------------|
124
+ | st_yolo_lc_v1 | Int8 | 192x192x3 | 39.0 % |
125
+ | st_yolo_lc_v1 | Float | 192x192x3 | 39.2 % |
126
+ | st_yolo_lc_v1 | Int8 | 224x224x3 | 42.94 % |
127
+ | st_yolo_lc_v1 | Float | 224x224x3 | 41.7 % |
128
+ | st_yolo_lc_v1 | Int8 | 256x256x3 | 43.8 % |
129
+ | st_yolo_lc_v1 | Float | 256x256x3 | 44.7 % |
130
+
131
+ \* EVAL_IOU = 0.4, NMS_THRESH = 0.5, SCORE_THRESH =0.001
132
+
133
+
134
+ ## Retraining and Integration in a simple example:
135
+
136
+ Please refer to the stm32ai-modelzoo-services GitHub [here](https://github.com/STMicroelectronics/stm32ai-modelzoo-services)
137
+ # References
138
+
139
+
140
+ <a id="1">[1]</a>
141
+ “Microsoft COCO: Common Objects in Context”. [Online]. Available: https://cocodataset.org/#download.
142
+ @article{DBLP:journals/corr/LinMBHPRDZ14,
143
+ author = {Tsung{-}Yi Lin and
144
+ Michael Maire and
145
+ Serge J. Belongie and
146
+ Lubomir D. Bourdev and
147
+ Ross B. Girshick and
148
+ James Hays and
149
+ Pietro Perona and
150
+ Deva Ramanan and
151
+ Piotr Doll{'{a} }r and
152
+ C. Lawrence Zitnick},
153
+ title = {Microsoft {COCO:} Common Objects in Context},
154
+ journal = {CoRR},
155
+ volume = {abs/1405.0312},
156
+ year = {2014},
157
+ url = {http://arxiv.org/abs/1405.0312},
158
+ archivePrefix = {arXiv},
159
+ eprint = {1405.0312},
160
+ timestamp = {Mon, 13 Aug 2018 16:48:13 +0200},
161
+ biburl = {https://dblp.org/rec/bib/journals/corr/LinMBHPRDZ14},
162
+ bibsource = {dblp computer science bibliography, https://dblp.org}
163
+ }
164
+