Hugging Face's logo

STMicroelectronics
/
st_yoloxn

Object Detection
Model card Files
xet
Community
FBAGSTM commited on
Commit
12befa9
·
verified ·
1 Parent(s): ddc95ce

Release AI-ModelZoo-4.0.0

Browse files
Files changed (1) hide show
  1. README.md +198 -6
README.md CHANGED
@@ -1,6 +1,198 @@
1
- ---
2
- license: other
3
- license_name: sla0044
4
- license_link: >-
5
- https://github.com/STMicroelectronics/stm32ai-modelzoo/blob/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/LICENSE.md
6
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ license_name: sla0044
4
+ license_link: >-
5
+ https://github.com/STMicroelectronics/stm32ai-modelzoo/blob/main/object_detection/st_yoloxn/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 | |
25
+ | Paper | |
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
+ | | |
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 | Hyperparameters (depth_width) | Dataset | Format | Resolution | Series | Internal RAM (KiB) | External RAM (KiB) | Weights Flash (KiB) | STEdgeAI Core version |
63
+ |-----------------------------------------------------------------------------------------------------------------------------------|-------------------------------|------------------|----------|--------------|----------|----------------------|----------------------|------------------------|-------------------------|
64
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_192/st_yoloxn_d033_w025_192_int8.tflite) | d033_w025 | COCO-Person | Int8 | 192x192x3 | STM32N6 | 333 | 0 | 877.66 | 3.0.0 |
65
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_256/st_yoloxn_d033_w025_256_int8.tflite) | d033_w025 | COCO-Person | Int8 | 256x256x3 | STM32N6 | 624 | 0 | 884.91 | 3.0.0 |
66
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_320/st_yoloxn_d033_w025_320_int8.tflite) | d033_w025 | COCO-Person | Int8 | 320x320x3 | STM32N6 | 1125 | 0 | 895.03 | 3.0.0 |
67
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_416/st_yoloxn_d033_w025_416_int8.tflite) | d033_w025 | COCO-Person | Int8 | 416x416x3 | STM32N6 | 2676.12| 0 | 904.03 | 3.0.0 |
68
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d050_w040_256/st_yoloxn_d050_w040_256_int8.tflite) | d050_w040 | COCO-Person | Int8 | 256x256x3 | STM32N6 | 833 | 0 | 2414.64 | 3.0.0 |
69
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d100_w025_480/st_yoloxn_d100_w025_480_int8.tflite) | d100_w025 | COCO-Person | Int8 | 480x480x3 | STM32N6 | 2707.5 | 0 | 1173.25 | 3.0.0 |
70
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_custom_dataset/st_person/st_yoloxn_d033_w025_416/st_yoloxn_d033_w025_416_int8.tflite) | d033_w025 | ST-Person | Int8 | 416x416x3 | STM32N6 | 2676.12| 0 | 906.28 | 3.0.0 |
71
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_custom_dataset/st_person/st_yoloxn_d067_w025_416/st_yoloxn_d067_w025_416_int8.tflite) | d067_w025 | ST-Person | Int8 | 416x416x3 | STM32N6 | 2681.41| 0 | 1039.78 | 3.0.0 |
72
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_custom_dataset/st_person/st_yoloxn_d100_w025_416/st_yoloxn_d100_w025_416_int8.tflite) | d100_w025 | ST-Person | Int8 | 416x416x3 | STM32N6 | 2676.12| 0 | 1173.28 | 3.0.0 |
73
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_custom_dataset/st_person/st_yoloxn_d100_w025_480/st_yoloxn_d100_w025_480_int8.tflite) | d100_w025 | ST-Person | Int8 | 480x480x3 | STM32N6 | 2707.5 | 0 | 1173.25 | 3.0.0 |
74
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_custom_dataset/st_person/st_yoloxn_d100_w025_480/st_yoloxn_d100_w025_480_qdq_w4_78.84%_w8_21.16%_a8_100%_map_47.33.onnx) | d100_w025 | ST-Person |W4A8 | 480x480x3 | STM32N6 | 2593 | 0 | 738.53 | 3.0.0 |
75
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_80_classes/st_yoloxn_d100_w025_480/st_yoloxn_d100_w025_480_qdq_int8.onnx) | d100_w025 | COCO-80-classes | W4A8 | 480x480x3 | STM32N6 | 3491.48| 0 | 1217.59 | 3.0.0 |
76
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_192/st_yoloxn_d033_w025_192_qdq_w4_83.16%_w8_16.84%_a8_100%_map_37.34.onnx) | d033_w025 | COCO-Person | W4A8 | 192x192x3 | STM32N6 | 360 | 0 | 519.06 | 3.0.0 |
77
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_256/st_yoloxn_d033_w025_256_qdq_w4_83.16%_w8_16.84%_a8_100%_map_44.43.onnx) | d033_w025 | COCO-Person | W4A8 | 256x256x3 | STM32N6 | 624 | 0 | 526.31 | 3.0.0 |
78
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_320/st_yoloxn_d033_w025_320_qdq_w4_59.47%_w8_40.53%_a8_100%_map_50.61.onnx) | d033_w025 | COCO-Person | W4A8 | 320x320x3 | STM32N6 | 1125 | 0 | 638.67 | 3.0.0 |
79
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_416/st_yoloxn_d033_w025_416_qdq_w4_76.19%_w8_23.81%_a8_100%_map_53.97.onnx) | d033_w025 | COCO-Person | W4A8 | 416x416x3 | STM32N6 | 2670.84| 0 | 575.53 | 3.0.0 |
80
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d050_w040_256/st_yoloxn_d050_w040_256_qdq_w4_62.53%_w8_37.47%_a8_100%_map_49.2.onnx) | d050_w040 | COCO-Person | W4A8 | 256x256x3 | STM32N6 | 835 | 0 | 1671.08 | 3.0.0 |
81
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d100_w025_480/st_yoloxn_d100_w025_480_qdq_w4_55.23%_w8_44.77%_a8_100%_map_59.9.onnx) | d100_w025 | COCO-Person | W4A8 | 480x480x3 | STM32N6 | 2707.5 | 0 | 868.8 | 3.0.0 |
82
+
83
+
84
+ ### Reference **NPU** inference time based on COCO Person dataset (see Accuracy for details on dataset)
85
+ | Model | Hyperparameters (depth_width) | Dataset | Format | Resolution | Board | Execution Engine | Inference time (ms) | Inf / sec | STEdgeAI Core version |
86
+ |-----------------------------------------------------------------------------------------------------------------------------------|-------------------------------|----------------------|----------|--------------|---------------|--------------------|-----------------------|-------------|-------------------------|
87
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_192/st_yoloxn_d033_w025_192_int8.tflite) | d033_w025 | COCO-Person | Int8 | 192x192x3 | STM32N6570-DK | NPU/MCU | 6.63 | 150.74 | 3.0.0 |
88
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_256/st_yoloxn_d033_w025_256_int8.tflite) | d033_w025 | COCO-Person | Int8 | 256x256x3 | STM32N6570-DK | NPU/MCU | 9.42 | 106.17 | 3.0.0 |
89
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_320/st_yoloxn_d033_w025_320_int8.tflite) | d033_w025 | COCO-Person | Int8 | 320x320x3 | STM32N6570-DK | NPU/MCU | 13.29 | 75.26 | 3.0.0 |
90
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_416/st_yoloxn_d033_w025_416_int8.tflite) | d033_w025 | COCO-Person | Int8 | 416x416x3 | STM32N6570-DK | NPU/MCU | 21.3 | 46.95 | 3.0.0 |
91
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d050_w040_256/st_yoloxn_d050_w040_256_int8.tflite) | d050_w040 | COCO-Person | Int8 | 256x256x3 | STM32N6570-DK | NPU/MCU | 20.12 | 49.70 | 3.0.0 |
92
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d100_w025_480/st_yoloxn_d100_w025_480_int8.tflite) | d100_w025 | COCO-Person | Int8 | 480x480x3 | STM32N6570-DK | NPU/MCU | 35.8 | 27.93 | 3.0.0 |
93
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_custom_dataset/st_person/st_yoloxn_d033_w025_416/st_yoloxn_d033_w025_416_int8.tflite) | d033_w025 | ST-Person | Int8 | 416x416x3 | STM32N6570-DK | NPU/MCU | 21.66 | 46.16 | 3.0.0 |
94
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_custom_dataset/st_person/st_yoloxn_d067_w025_416/st_yoloxn_d067_w025_416_int8.tflite) | d067_w025 | ST-Person | Int8 | 416x416x3 | STM32N6570-DK | NPU/MCU | 24.37 | 41.02 | 3.0.0 |
95
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_custom_dataset/st_person/st_yoloxn_d100_w025_416/st_yoloxn_d100_w025_416_int8.tflite) | d100_w025 | ST-Person | Int8 | 416x416x3 | STM32N6570-DK | NPU/MCU | 27.02 | 36.99 | 3.0.0 |
96
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_custom_dataset/st_person/st_yoloxn_d100_w025_480/st_yoloxn_d100_w025_480_int8.tflite) | d100_w025 | ST-Person | Int8 | 480x480x3 | STM32N6570-DK | NPU/MCU | 35.8 | 27.93 | 3.0.0 |
97
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_custom_dataset/st_person/st_yoloxn_d100_w025_480/st_yoloxn_d100_w025_480_qdq_w4_78.84%_w8_21.16%_a8_100%_map_47.33.onnx) | d100_w025 | ST-Person | W4A8 | 480x480x3 | STM32N6570-DK | NPU/MCU | 34.27 | 29.18 | 3.0.0 |
98
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_80_classes/st_yoloxn_d100_w025_480/st_yoloxn_d100_w025_480_qdq_int8.onnx) | d100_w025 | COCO-80-classes | W4A8 | 480x480x3 | STM32N6570-DK | NPU/MCU | 49.34 | 20.27 | 3.0.0 |
99
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_192/st_yoloxn_d033_w025_192_qdq_w4_83.16%_w8_16.84%_a8_100%_map_37.34.onnx) | d033_w025 | COCO-Person | W4A8 | 192x192x3 | STM32N6570-DK | NPU/MCU | 6.33 | 157.96 | 3.0.0 |
100
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_256/st_yoloxn_d033_w025_256_qdq_w4_83.16%_w8_16.84%_a8_100%_map_44.43.onnx) | d033_w025 | COCO-Person | W4A8 | 256x256x3 | STM32N6570-DK | NPU/MCU | 8.20 | 121.95 | 3.0.0 |
101
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_320/st_yoloxn_d033_w025_320_qdq_w4_59.47%_w8_40.53%_a8_100%_map_50.61.onnx) | d033_w025 | COCO-Person | W4A8 | 320x320x3 | STM32N6570-DK | NPU/MCU | 11.82 | 84.66 | 3.0.0 |
102
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_416/st_yoloxn_d033_w025_416_qdq_w4_76.19%_w8_23.81%_a8_100%_map_53.97.onnx) | d033_w025 | COCO-Person | W4A8 | 416x416x3 | STM32N6570-DK | NPU/MCU | 19.87 | 50.32 | 3.0.0 |
103
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d050_w040_256/st_yoloxn_d050_w040_256_qdq_w4_62.53%_w8_37.47%_a8_100%_map_49.2.onnx) | d050_w040 | COCO-Person | W4A8 | 256x256x3 | STM32N6570-DK | NPU/MCU | 18.34 | 54.52 | 3.0.0 |
104
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d100_w025_480/st_yoloxn_d100_w025_480_qdq_w4_55.23%_w8_44.77%_a8_100%_map_59.9.onnx) | d100_w025 | COCO-Person | W4A8 | 480x480x3 | STM32N6570-DK | NPU/MCU | 34.71 | 28.80 | 3.0.0 |
105
+
106
+
107
+ ### Reference **MCU** memory footprint based on COCO Person dataset (see Accuracy for details on dataset)
108
+
109
+ | Model | Hyperparameters (depth_width) | Format | Resolution | Series | Activation RAM (KiB) | Runtime RAM (KiB) | Weights Flash (KiB) | Code Flash (KiB) | Total RAM | Total Flash | STEdgeAI Core version |
110
+ |-----------------------------------------------------------------------------------------------------------------------------------|-------------------------------|----------|--------------|----------|------------------------|---------------------|-----------------------|--------------------|-------------|---------------|-------------------------|
111
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_192/st_yoloxn_d033_w025_192_int8.tflite) | d033_w025 | Int8 | 192x192x3 | STM32H7 | 184.92 | 12.54 | 891.18 | 108.38 | 197.46 | 999.56 | 3.0.0 |
112
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_256/st_yoloxn_d033_w025_256_int8.tflite) | d033_w025 | Int8 | 256x256x3 | STM32H7 | 324.92 | 12.54 | 891.18 | 108.38 | 337.46 | 999.84 | 3.0.0 |
113
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d050_w040_256/st_yoloxn_d050_w040_256_int8.tflite) | d050_w040 | Int8 | 256x256x3 | STM32H7 | 451.4 | 15.63 | 2435.76 | 151.42 | 467.03 | 2587.18 | 3.0.0 |
114
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_320/st_yoloxn_d033_w025_320_int8.tflite) | d033_w025 | Int8 | 320x320x3 | STM32H7 | 504.92 | 12.54 | 891.18 | 108.38 | 517.46 | 1000.04 | 3.0.0 |
115
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_416/st_yoloxn_d033_w025_416_int8.tflite) | d033_w025 | Int8 | 416x416x3 | STM32H7 | 849.92 | 12.54 | 891.18 | 108.38 | 862.46 | 999.99 | 3.0.0 |
116
+
117
+
118
+ ### Reference **MCU** inference time based on COCO Person dataset (see Accuracy for details on dataset)
119
+
120
+ | Model | Hyperparameters (depth_width) | Format | Resolution | Board | Execution Engine | Frequency | Inference time (ms) | STEdgeAI Core version |
121
+ |-----------------------------------------------------------------------------------------------------------------------------------|-------------------------------|----------|--------------|------------------|--------------------|-------------|-----------------------|-------------------------|
122
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_192/st_yoloxn_d033_w025_192_int8.tflite) | d033_w025 | Int8 | 192x192x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 357.79| 3.0.0 |
123
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_256/st_yoloxn_d033_w025_256_int8.tflite) | d033_w025 | Int8 | 256x256x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 641.66| 3.0.0 |
124
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d050_w040_256/st_yoloxn_d050_w040_256_int8.tflite) | d050_w040 | Int8 | 256x256x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 1698.36| 3.0.0 |
125
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_320/st_yoloxn_d033_w025_320_int8.tflite) | d033_w025 | Int8 | 320x320x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 1026.35| 3.0.0 |
126
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_416/st_yoloxn_d033_w025_416_int8.tflite) | d033_w025 | Int8 | 416x416x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 1797.73| 3.0.0 |
127
+
128
+
129
+ ### AP on COCO Person dataset
130
+
131
+ 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
132
+
133
+ | Model | Hyperparameters (depth_width) | Format | Resolution | Depth Multiplier | Width Multiplier | Anchors | AP |
134
+ |-------|------------------------------|--------|------------|------------------|------------------|---------|----|
135
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_192/st_yoloxn_d033_w025_192_int8.tflite) | d033_w025 | Int8 | 192x192x3 | 0.33 | 0.25 | 1 | 38.16 % |
136
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_192/st_yoloxn_d033_w025_192.keras) | d033_w025 | Float | 192x192x3 | 0.33 | 0.25 | 1 | 38.82 % |
137
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_192/st_yoloxn_d033_w025_192_qdq_w4_83.16%_w8_16.84%_a8_100%_map_37.34.onnx) | d033_w025 | W4A8 | 192x192x3 | 0.33 | 0.25 | 1 | 37.34 % |
138
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_256/st_yoloxn_d033_w025_256_int8.tflite) | d033_w025 | Int8 | 256x256x3 | 0.33 | 0.25 | 1 | 44.85 % |
139
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_256/st_yoloxn_d033_w025_256.keras) | d033_w025 | Float | 256x256x3 | 0.33 | 0.25 | 1 | 45.12 % |
140
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_256/st_yoloxn_d033_w025_256_qdq_w4_83.16%_w8_16.84%_a8_100%_map_44.43.onnx) | d033_w025 | W4A8 | 192x192x3 | 0.33 | 0.25 | 1 | 44.43 % |
141
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d050_w040_256/st_yoloxn_d050_w040_256_int8.tflite) | d050_w040 | Int8 | 256x256x3 | 0.5 | 0.4 | 1 | 50.05 % |
142
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d050_w040_256/st_yoloxn_d050_w040_256.keras) | d050_w040 | Float | 256x256x3 | 0.5 | 0.4 | 1 | 51.07 % |
143
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d050_w040_256/st_yoloxn_d050_w040_256_qdq_w4_62.53%_w8_37.47%_a8_100%_map_49.2.onnx) | d050_w040 | W4A8 | 256x256x3 | 0.33 | 0.25 | 1 | 49.2 % |
144
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_320/st_yoloxn_d033_w025_320_int8.tflite) | d033_w025 | Int8 | 320x320x3 | 0.33 | 0.25 | 1 | 51.17 % |
145
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_320/st_yoloxn_d033_w025_320.keras) | d033_w025 | Float | 320x320x3 | 0.33 | 0.25 | 1 | 51.69 % |
146
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_320/st_yoloxn_d033_w025_320_qdq_w4_59.47%_w8_40.53%_a8_100%_map_50.61.onnx) | d033_w025 | W4A8 | 320x320x3 | 0.33 | 0.25 | 1 | 50.61 % |
147
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_416/st_yoloxn_d033_w025_416_int8.tflite) | d033_w025 | Int8 | 416x416x3 | 0.33 | 0.25 | 1 | 54.8 % |
148
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_416/st_yoloxn_d033_w025_416.keras) | d033_w025 | Float | 416x416x3 | 0.33 | 0.25 | 1 | 54.82 % |
149
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d033_w025_416/st_yoloxn_d033_w025_416_qdq_w4_76.19%_w8_23.81%_a8_100%_map_53.97.onnx) | d033_w025 | W4A8 | 416x416x3 | 0.33 | 0.25 | 1 | 53.97 % |
150
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d100_w025_480/st_yoloxn_d100_w025_480_int8.tflite) | d100_w025 | Int8 | 480x480x3 | 1.0 | 0.25 | 3 | 61.1 % |
151
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d100_w025_480/st_yoloxn_d100_w025_480.keras) | d100_w025 | Float | 480x480x3 | 1.0 | 0.25 | 3 | 61.7 % |
152
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_person/st_yoloxn_d100_w025_480/st_yoloxn_d100_w025_480_qdq_w4_55.23%_w8_44.77%_a8_100%_map_59.9.onnx) | d100_w025 | W4A8 | 480x480x3 | 0.33 | 0.25 | 1 | 59.9 % |
153
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_custom_dataset/st_person/st_yoloxn_d033_w025_416/st_yoloxn_d033_w025_416_int8.tflite) | d033_w025 | Int8 | 416x416x3 | 0.33 | 0.25 | 1 | 42.58 % |
154
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_custom_dataset/st_person/st_yoloxn_d033_w025_416/st_yoloxn_d033_w025_416.keras) | d033_w025 | Float | 416x416x3 | 0.33 | 0.25 | 1 | 44.28 % |
155
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_custom_dataset/st_person/st_yoloxn_d067_w025_416/st_yoloxn_d067_w025_416_int8.tflite) | d067_w025 | Int8 | 416x416x3 | 0.67 | 0.25 | 1 | 46.2 % |
156
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_custom_dataset/st_person/st_yoloxn_d067_w025_416/st_yoloxn_d067_w025_416.keras) | d067_w025 | Float | 416x416x3 | 0.67 | 0.25 | 1 | 46.7 % |
157
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_custom_dataset/st_person/st_yoloxn_d100_w025_416/st_yoloxn_d100_w025_416_int8.tflite) | d100_w025 | Int8 | 416x416x3 | 1.0 | 0.25 | 1 | 47.27 % |
158
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_custom_dataset/st_person/st_yoloxn_d100_w025_416/st_yoloxn_d100_w025_416.keras) | d100_w025 | Float | 416x416x3 | 1.0 | 0.25 | 1 | 48.14 % |
159
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_custom_dataset/st_person/st_yoloxn_d100_w025_480/st_yoloxn_d100_w025_480_int8.tflite) | d100_w025 | Int8 | 480x480x3 | 1.0 | 0.25 | 1 | 48.15 % |
160
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_custom_dataset/st_person/st_yoloxn_d100_w025_480/st_yoloxn_d100_w025_480.keras) | d100_w025 | Float | 480x480x3 | 1.0 | 0.25 | 1 | 48.68 % |
161
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_custom_dataset/st_person/st_yoloxn_d100_w025_480/st_yoloxn_d100_w025_480_qdq_w4_78.84%_w8_21.16%_a8_100%_map_47.33.onnx) | d100_w025 | W4A8 | 480x480x3 | 1.0 | 0.25 | 1 | 47.3 % |
162
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_80_classes/st_yoloxn_d100_w025_480/st_yoloxn_d100_w025_480.tflite) | d100_w025 | Int8 | 480x480x3 | 1.0 | 0.25 | 1 | 34.3 % |
163
+ | [st_yoloxn](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/object_detection/st_yoloxn/ST_pretrainedmodel_public_dataset/coco_2017_80_classes/st_yoloxn_d100_w025_480/st_yoloxn_d100_w025_480.keras) | d100_w025 | Float | 480x480x3 | 1.0 | 0.25 | 1 | 35.7 % |
164
+
165
+ \* EVAL_IOU = 0.5, NMS_THRESH = 0.5, SCORE_THRESH = 0.001, MAX_DETECTIONS = 100
166
+
167
+ ## Retraining and Integration in a simple example:
168
+
169
+ Please refer to the stm32ai-modelzoo-services GitHub [here](https://github.com/STMicroelectronics/stm32ai-modelzoo-services)
170
+
171
+
172
+ # References
173
+
174
+
175
+ <a id="1">[1]</a>
176
+ “Microsoft COCO: Common Objects in Context”. [Online]. Available: https://cocodataset.org/#download.
177
+ @article{DBLP:journals/corr/LinMBHPRDZ14,
178
+ author = {Tsung{-}Yi Lin and
179
+ Michael Maire and
180
+ Serge J. Belongie and
181
+ Lubomir D. Bourdev and
182
+ Ross B. Girshick and
183
+ James Hays and
184
+ Pietro Perona and
185
+ Deva Ramanan and
186
+ Piotr Doll{'{a} }r and
187
+ C. Lawrence Zitnick},
188
+ title = {Microsoft {COCO:} Common Objects in Context},
189
+ journal = {CoRR},
190
+ volume = {abs/1405.0312},
191
+ year = {2014},
192
+ url = {http://arxiv.org/abs/1405.0312},
193
+ archivePrefix = {arXiv},
194
+ eprint = {1405.0312},
195
+ timestamp = {Mon, 13 Aug 2018 16:48:13 +0200},
196
+ biburl = {https://dblp.org/rec/bib/journals/corr/LinMBHPRDZ14},
197
+ bibsource = {dblp computer science bibliography, https://dblp.org}
198
+ }