Hugging Face's logo

STMicroelectronics
/
tiny_yolo_v2

Object Detection
Model card Files Community
FBAGSTM commited on
Commit
95f29f0
·
verified ·
1 Parent(s): 041601d

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +171 -6
README.md CHANGED
@@ -1,6 +1,171 @@
1
- ---
2
- license: other
3
- license_name: sla0044
4
- license_link: >-
5
- https://github.com/STMicroelectronics/stm32aimodelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_custom_dataset/LICENSE.md
6
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ license_name: sla0044
4
+ license_link: >-
5
+ https://github.com/STMicroelectronics/stm32aimodelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_custom_dataset/LICENSE.md
6
+ pipeline_tag: object-detection
7
+ ---
8
+ # Tiny Yolo v2 quantized
9
+
10
+ ## **Use case** : `Object detection`
11
+
12
+ # Model description
13
+
14
+
15
+ Tiny Yolo v2 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
+ | Provenance | https://github.com/AlexeyAB/darknet
27
+ | Paper | https://pjreddie.com/media/files/papers/YOLO9000.pdf |
28
+
29
+ The models are quantized using tensorflow lite converter.
30
+
31
+
32
+ ## Network inputs / outputs
33
+
34
+
35
+ For an image resolution of NxM and NC classes
36
+
37
+ | Input Shape | Description |
38
+ | ----- | ----------- |
39
+ | (1, W, H, 3) | Single NxM RGB image with UINT8 values between 0 and 255 |
40
+
41
+ | Output Shape | Description |
42
+ | ----- | ----------- |
43
+ | (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|
44
+
45
+
46
+ ## Recommended Platforms
47
+
48
+
49
+ | Platform | Supported | Recommended |
50
+ |----------|-----------|-------------|
51
+ | STM32L0 | [] | [] |
52
+ | STM32L4 | [] | [] |
53
+ | STM32U5 | [] | [] |
54
+ | STM32H7 | [x] | [] |
55
+ | STM32MP1 | [x] | [x] |
56
+ | STM32MP2 | [x] | [x] |
57
+ | STM32N6 | [x] | [x] |
58
+
59
+ # Performances
60
+
61
+ ## Metrics
62
+
63
+
64
+ Measures are done with default STM32Cube.AI configuration with enabled input / output allocated option.
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
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_224/tiny_yolo_v2_224_int8.tflite) | COCO-Person | Int8 | 224x224x3 | STM32N6 | 392 | 0.0 | 10804.81 | 10.0.0 | 2.0.0 |
70
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_custom_dataset/st_person/tiny_yolo_v2_224/tiny_yolo_v2_224_int8.tflite) | ST-Person | Int8 | 224x224x3 | STM32N6 | 392 | 0.0 | 10804.81 | 10.0.0 | 2.0.0 |
71
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_416/tiny_yolo_v2_416_int8.tflite) | COCO-Person | Int8 | 416x416x3 | STM32N6 | 1880.12 | 0.0 | 10829 | 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
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_224/tiny_yolo_v2_224_int8.tflite) | COCO-Person | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 30.67 | 32.61 |10.0.0 | 2.0.0 |
78
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_custom_dataset/st_person/tiny_yolo_v2_224/tiny_yolo_v2_224_int8.tflite) | ST-Person | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 30.67 | 32.61| 10.0.0 | 2.0.0 |
79
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_416/tiny_yolo_v2_416_int8.tflite) | COCO-Person | Int8 | 416x416x3 | STM32N6570-DK | NPU/MCU | 50.91 | 19.64 | 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
+
84
+ | Model | Format | Resolution | Series | Activation RAM | Runtime RAM | Weights Flash | Code Flash | Total RAM | Total Flash | STM32Cube.AI version |
85
+ |-------------------|--------|--------------|---------|----------------|-------------|---------------|------------|-------------|--------------|-----------------------|
86
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_192/tiny_yolo_v2_192_int8.tflite) | Int8 | 192x192x3 | STM32H7 | 220.6 KiB | 7.98 KiB | 10775.98 KiB | 55.85 KiB | 228.58 KiB | 10831.83 KiB | 10.0.0 |
87
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_224/tiny_yolo_v2_224_int8.tflite) | Int8 | 224x224x3 | STM32H7 | 249.35 KiB | 7.98 KiB | 10775.98 KiB | 55.8 KiB | 257.33 KiB | 10831.78 KiB | 10.0.0 |
88
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_416/tiny_yolo_v2_416_int8.tflite) | Int8 | 416x416x3 | STM32H7 | 1263.07 KiB | 8.03 KiB | 10775.98 KiB | 55.85 KiB | 1271.1 KiB | 10831.83 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
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_192/tiny_yolo_v2_192_int8.tflite) | Int8 | 192x192x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 3006.3 ms | 10.0.0 |
97
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_224/tiny_yolo_v2_224_int8.tflite) | Int8 | 224x224x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 2742.3 ms | 10.0.0 |
98
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_416/tiny_yolo_v2_416_int8.tflite) | Int8 | 416x416x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 10468.2 ms | 10.0.0 |
99
+
100
+
101
+ ### Reference **MPU** inference time based on COCO Person dataset (see Accuracy for details on dataset)
102
+
103
+
104
+ | Model | Format | Resolution | Quantization | Board | Execution Engine | Frequency | Inference time (ms) | %NPU | %GPU | %CPU | X-LINUX-AI version | Framework |
105
+ |--------------|--------|------------|---------------|-------------------|------------------|-----------|---------------------|-------|-------|------|--------------------|-----------------------|
106
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_224/tiny_yolo_v2_224_int8.tflite) | Int8 | 224x224x3 | per-channel** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 120.8 ms | 3.45 | 96.55 |0 | v5.1.0 | OpenVX |
107
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_416/tiny_yolo_v2_416_int8.tflite) | Int8 | 416x416x3 | per-channel** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 425.6 ms | 2.74 | 97.26 |0 | v5.1.0 | OpenVX |
108
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_224/tiny_yolo_v2_224_int8.tflite) | Int8 | 224x224x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 410.50 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
109
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_416/tiny_yolo_v2_416_int8.tflite) | Int8 | 416x416x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 1347 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
110
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_224/tiny_yolo_v2_224_int8.tflite) | Int8 | 224x224x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 619.70 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
111
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_416/tiny_yolo_v2_416_int8.tflite) | Int8 | 416x416x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 2105 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
112
+
113
+ ** **To get the most out of MP25 NPU hardware acceleration, please use per-tensor quantization**
114
+
115
+ ### AP on COCO Person dataset
116
+
117
+ 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
118
+
119
+ | Model | Format | Resolution | AP |
120
+ |-------|--------|------------|----------------|
121
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_192/tiny_yolo_v2_192_int8.tflite) | Int8 | 192x192x3 | 33.7 % |
122
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_192/tiny_yolo_v2_192.h5) | Float | 192x192x3 | 34.5 % |
123
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_224/tiny_yolo_v2_224_int8.tflite) | Int8 | 224x224x3 | 37.3 % |
124
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_224/tiny_yolo_v2_224.h5) | Float | 224x224x3 | 38.4 % |
125
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_416/tiny_yolo_v2_416_int8.tflite) | Int8 | 416x416x3 | 50.7 % |
126
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_public_dataset/coco_2017_person/tiny_yolo_v2_416/tiny_yolo_v2_416.h5) | Float | 416x416x3 | 51.5 % |
127
+
128
+ \* EVAL_IOU = 0.4, NMS_THRESH = 0.5, SCORE_THRESH =0.001
129
+
130
+ ### AP on ST Person dataset
131
+
132
+ | Model | Format | Resolution | AP |
133
+ |-------|--------|------------|----------------|
134
+ | [tiny_yolo_v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/object_detection/tiny_yolo_v2/ST_pretrainedmodel_custom_dataset/st_person/tiny_yolo_v2_224/tiny_yolo_v2_224_int8.tflite) | Int8 | 224x224x3 | 34.0 % |
135
+
136
+
137
+ \* EVAL_IOU = 0.4, NMS_THRESH = 0.5, SCORE_THRESH =0.001
138
+
139
+ ## Retraining and Integration in a simple example:
140
+
141
+ Please refer to the stm32ai-modelzoo-services GitHub [here](https://github.com/STMicroelectronics/stm32ai-modelzoo-services)
142
+
143
+
144
+ # References
145
+
146
+
147
+ <a id="1">[1]</a>
148
+ “Microsoft COCO: Common Objects in Context”. [Online]. Available: https://cocodataset.org/#download.
149
+ @article{DBLP:journals/corr/LinMBHPRDZ14,
150
+ author = {Tsung{-}Yi Lin and
151
+ Michael Maire and
152
+ Serge J. Belongie and
153
+ Lubomir D. Bourdev and
154
+ Ross B. Girshick and
155
+ James Hays and
156
+ Pietro Perona and
157
+ Deva Ramanan and
158
+ Piotr Doll{'{a} }r and
159
+ C. Lawrence Zitnick},
160
+ title = {Microsoft {COCO:} Common Objects in Context},
161
+ journal = {CoRR},
162
+ volume = {abs/1405.0312},
163
+ year = {2014},
164
+ url = {http://arxiv.org/abs/1405.0312},
165
+ archivePrefix = {arXiv},
166
+ eprint = {1405.0312},
167
+ timestamp = {Mon, 13 Aug 2018 16:48:13 +0200},
168
+ biburl = {https://dblp.org/rec/bib/journals/corr/LinMBHPRDZ14},
169
+ bibsource = {dblp computer science bibliography, https://dblp.org}
170
+ }
171
+