Hugging Face's logo

STMicroelectronics
/
st_yolo_x

Object Detection
Model card Files
xet
Community
FBAGSTM commited on
Commit
ffd09fd
·
verified ·
1 Parent(s): ae1a328

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +154 -6
README.md CHANGED
@@ -1,6 +1,154 @@
1
- ---
2
- license: other
3
- license_name: sla0044
4
- license_link: >-
5
- https://github.com/STMicroelectronics/stm32aimodelzoo/object_detection/st_yolo_x/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_x/ST_pretrainedmodel_public_dataset/LICENSE.md
6
+ pipeline_tag: object-detection
7
+ ---
8
+ # ST Yolo X quantized
9
+
10
+ ## **Use case** : `Object detection`
11
+
12
+ # Model description
13
+
14
+
15
+ ST Yolo X is a real-time object detection model targeted for real-time processing implemented in Tensorflow.
16
+ This is an optimized ST version of the well known yolo x, quantized in int8 format using tensorflow lite converter.
17
+
18
+ ## Network information
19
+
20
+ | Network information | Value |
21
+ |-------------------------|-----------------|
22
+ | Framework | TensorFlow Lite |
23
+ | Quantization | int8 |
24
+ | Provenance | TO DO |
25
+ | Paper | TO DO |
26
+
27
+
28
+
29
+ ## Network inputs / outputs
30
+
31
+ For an image resolution of NxM and NC classes
32
+
33
+ | Input Shape | Description |
34
+ | ----- | ----------- |
35
+ | (1, W, H, 3) | Single NxM RGB image with UINT8 values between 0 and 255 |
36
+
37
+ | Output Shape | Description |
38
+ | ----- | ----------- |
39
+ | TO DO |
40
+
41
+
42
+ ## Recommended Platforms
43
+
44
+ | Platform | Supported | Recommended |
45
+ |----------|-----------|-------------|
46
+ | STM32L0 | [] | [] |
47
+ | STM32L4 | [] | [] |
48
+ | STM32U5 | [] | [] |
49
+ | STM32H7 | [x] | [] |
50
+ | STM32MP1 | [x] | [] |
51
+ | STM32MP2 | [x] | [x] |
52
+ | STM32N6 | [x] | [x] |
53
+
54
+
55
+ # Performances
56
+
57
+ ## Metrics
58
+
59
+ Measures are done with default STM32Cube.AI configuration with enabled input / output allocated option.
60
+
61
+ ### Reference **NPU** memory footprint based on COCO Person dataset (see Accuracy for details on dataset)
62
+ |Model | Dataset | Format | Resolution | Series | Internal RAM (KiB) | External RAM (KiB)| Weights Flash (KiB)| STM32Cube.AI version | STEdgeAI Core version |
63
+ |----------|------------------|--------|-------------|------------------|------------------|---------------------|-------|----------------------|-------------------------|
64
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_192/st_yolo_x_nano_192_0.33_0.25_int8.tflite) | COCO-Person | Int8 | 192x192x3 | STM32N6 | 324 | 0.0 | 1028.08 | 10.0.0 | 2.0.0 |
65
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_256/st_yolo_x_nano_256_0.33_0.25_int8.tflite) | COCO-Person | Int8 | 256x256x3 | STM32N6 | 624 | 0.0 | 1028.08 | 10.0.0 | 2.0.0 |
66
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_256/st_yolo_x_nano_256_0.5_0.4_int8.tflite) | COCO-Person | Int8 | 256x256x3 | STM32N6 | 971.62 | 0.0 | 2547.17 | 10.0.0 | 2.0.0 |
67
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_320/st_yolo_x_nano_320_0.33_0.25_int8.tflite) | COCO-Person | Int8 | 320x320x3 | STM32N6 | 968.5 | 0.0 | 1028.08 | 10.0.0 | 2.0.0 |
68
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_416/st_yolo_x_nano_416_0.33_0.25_int8.tflite) | COCO-Person | Int8 | 416x416x3 | STM32N6 | 2640.62 | 0.0 | 1027.89 | 10.0.0 | 2.0.0 |
69
+
70
+ ### Reference **NPU** inference time based on COCO Person dataset (see Accuracy for details on dataset)
71
+ | Model | Dataset | Format | Resolution | Board | Execution Engine | Inference time (ms) | Inf / sec | STM32Cube.AI version | STEdgeAI Core version |
72
+ |--------|------------------|--------|-------------|------------------|------------------|---------------------|-------|----------------------|-------------------------|
73
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_192/st_yolo_x_nano_192_0.33_0.25_int8.tflite) | COCO-Person | Int8 | 192x192x3 | STM32N6570-DK | NPU/MCU | 5.99 | 166.94 | 10.0.0 | 2.0.0 |
74
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_256/st_yolo_x_nano_256_0.33_0.25_int8.tflite) | COCO-Person | Int8 | 256x256x3 | STM32N6570-DK | NPU/MCU | 8.5 | 117.65 | 10.0.0 | 2.0.0 |
75
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_256/st_yolo_x_nano_256_0.5_0.4_int8.tflite) | COCO-Person | Int8 | 256x256x3 | STM32N6570-DK | NPU/MCU | 21.12 | 47.35 | 10.0.0 | 2.0.0 |
76
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_320/st_yolo_x_nano_320_0.33_0.25_int8.tflite) | COCO-Person | Int8 | 320x320x3 | STM32N6570-DK | NPU/MCU | 11.59 | 86.29 | 10.0.0 | 2.0.0 |
77
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_416/st_yolo_x_nano_416_0.33_0.25_int8.tflite) | COCO-Person | Int8 | 416x416x3 |
78
+ STM32N6570-DK | NPU/MCU | 17.99 | 55.59 | 10.0.0 | 2.0.0 |
79
+
80
+ ### Reference **MCU** memory footprint based on COCO Person dataset (see Accuracy for details on dataset)
81
+
82
+ | Model | Format | Resolution | Series | Activation RAM (KiB) | Runtime RAM (KiB)| Weights Flash (KiB)| Code Flash (KiB)| Total RAM | Total Flash | STM32Cube.AI version |
83
+ |-------------------|--------|--------------|---------|----------------|-------------|---------------|------------|-------------|--------------|-----------------------|
84
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_192/st_yolo_x_nano_192_0.33_0.25_int8.tflite) | Int8 | 192x192x3 | STM32H7 | 162.42 | 64.05 | 891.18 | 166.19 | 226.47 | 1057.37 | 10.0.0 |
85
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_256/st_yolo_x_nano_256_0.33_0.25_int8.tflite) | Int8 | 256x256x3 | STM32H7 | 284.92 | 64.05 | 891.18 | 166.21 | 348.97 | 1057.39 | 10.0.0 |
86
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_256/st_yolo_x_nano_256_0.5_0.4_int8.tflite) | Int8 | 256x256x3 | STM32H7 | 463.9 | 83.8 | 2435.76 | 228.22| 547.7 |2663.98 | 10.0.0 |
87
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_320/st_yolo_x_nano_320_0.33_0.25_int8.tflite) | Int8 | 320x320x3 | STM32H7 | 442.42 | 64.05 | 891.18 | 166.25 | 506.47 | 1057.43 | 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_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_192/st_yolo_x_nano_192_0.33_0.25_int8.tflite) | Int8 | 192x192x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 352.4 | 10.0.0 |
96
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_256/st_yolo_x_nano_256_0.33_0.25_int8.tflite) | Int8 | 256x256x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 619.92 | 10.0.0 |
97
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_256/st_yolo_x_nano_256_0.5_0.4_int8.tflite) | Int8 | 256x256x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 1696.59 | 10.0.0 |
98
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_320/st_yolo_x_nano_320_0.33_0.25_int8.tflite) | Int8 | 320x320x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 988.86 | 10.0.0 |
99
+
100
+
101
+
102
+ ### AP on COCO Person dataset
103
+
104
+
105
+ 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
106
+
107
+ | Model | Format | Resolution | AP |
108
+ |-------|--------|------------|----------------|
109
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_192/st_yolo_x_nano_192_0.33_0.25_int8.tflite) | Int8 | 192x192x3 | 45.1 % |
110
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_192/st_yolo_x_nano_192_0.33_0.25.h5) | Float | 192x192x3 | 45.2 % |
111
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_256/st_yolo_x_nano_256_0.33_0.25_int8.tflite) | Int8 | 256x256x3 | 53.6 % |
112
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_256/st_yolo_x_nano_256_0.33_0.25.h5) | Float | 256x256x3 | 53.3 % |
113
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_256/st_yolo_x_nano_256_0.5_0.4_int8.tflite) | Int8 | 256x256x3 | 58.6 % |
114
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_256/st_yolo_x_nano_256_0.5_0.4.h5) | Float | 256x256x3 | 58.7 % |
115
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_320/st_yolo_x_nano_320_0.33_0.25_int8.tflite) | Int8 | 320x320x3 | 57.1 % |
116
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_320/st_yolo_x_nano_320_0.33_0.25.h5) | Float | 320x320x3 | 57.1 % |
117
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_416/st_yolo_x_nano_416_0.33_0.25_int8.tflite) | Int8 | 416x416x3 | 62.2 % |
118
+ | [st_yolo_x_nano](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/st_yolo_x/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yolo_x_nano_416/st_yolo_x_nano_416_0.33_0.25.h5) | Float | 416x416x3 | 62.5 % |
119
+
120
+ \* EVAL_IOU = 0.4, NMS_THRESH = 0.5, SCORE_THRESH =0.001
121
+
122
+ ## Retraining and Integration in a simple example:
123
+
124
+ Please refer to the stm32ai-modelzoo-services GitHub [here](https://github.com/STMicroelectronics/stm32ai-modelzoo-services)
125
+
126
+
127
+ # References
128
+
129
+
130
+ <a id="1">[1]</a>
131
+ “Microsoft COCO: Common Objects in Context”. [Online]. Available: https://cocodataset.org/#download.
132
+ @article{DBLP:journals/corr/LinMBHPRDZ14,
133
+ author = {Tsung{-}Yi Lin and
134
+ Michael Maire and
135
+ Serge J. Belongie and
136
+ Lubomir D. Bourdev and
137
+ Ross B. Girshick and
138
+ James Hays and
139
+ Pietro Perona and
140
+ Deva Ramanan and
141
+ Piotr Doll{'{a} }r and
142
+ C. Lawrence Zitnick},
143
+ title = {Microsoft {COCO:} Common Objects in Context},
144
+ journal = {CoRR},
145
+ volume = {abs/1405.0312},
146
+ year = {2014},
147
+ url = {http://arxiv.org/abs/1405.0312},
148
+ archivePrefix = {arXiv},
149
+ eprint = {1405.0312},
150
+ timestamp = {Mon, 13 Aug 2018 16:48:13 +0200},
151
+ biburl = {https://dblp.org/rec/bib/journals/corr/LinMBHPRDZ14},
152
+ bibsource = {dblp computer science bibliography, https://dblp.org}
153
+ }
154
+