Hugging Face's logo

STMicroelectronics
/
mobilenetv1

Image Classification
Model card Files
xet
 Community
FBAGSTM commited on
Commit
9521fe0
·
verified ·
1 Parent(s): ae28784

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +242 -6
README.md CHANGED
@@ -1,6 +1,242 @@
1
- ---
2
- license: other
3
- license_name: sla0044
4
- license_link: >-
5
- https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/LICENSE.md
6
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ license_name: sla0044
4
+ license_link: >-
5
+ https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/LICENSE.md
6
+ ---
7
+ # MobileNet v1
8
+
9
+ ## **Use case** : `Image classification`
10
+
11
+ # Model description
12
+
13
+
14
+ MobileNet is a well known architecture that can be used in multiple use cases.
15
+ Input size and width factor called `alpha` are parameters to be adapted to various use cases complexity. The `alpha` parameter is used to increase or decrease the number of filters in each layer, allowing also to reduce the number of multiply-adds and then the inference time.
16
+
17
+ The original paper demonstrates the performance of MobileNet models using `alpha` values of 1.0, 0.75, 0.5 and 0.25.
18
+
19
+ (source: https://keras.io/api/applications/mobilenet/)
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
+ | MParams alpha=1.0 | 1.3 M |
30
+ | Quantization | int8 |
31
+ | Provenance | https://www.tensorflow.org/api_docs/python/tf/keras/applications/mobilenet |
32
+ | Paper | https://arxiv.org/abs/1704.04861 |
33
+
34
+ The models are quantized using tensorflow lite converter.
35
+
36
+
37
+ ## Network inputs / outputs
38
+
39
+
40
+ For an image resolution of NxM and P classes
41
+
42
+ | Input Shape | Description |
43
+ | ----- | ----------- |
44
+ | (1, N, M, 3) | Single NxM RGB image with UINT8 values between 0 and 255 |
45
+
46
+ | Output Shape | Description |
47
+ | ----- | ----------- |
48
+ | (1, P) | Per-class confidence for P classes in FLOAT32|
49
+
50
+
51
+ ## Recommended platforms
52
+
53
+
54
+ | Platform | Supported | Recommended |
55
+ |----------|-----------|-----------|
56
+ | STM32L0 |[]|[]|
57
+ | STM32L4 |[x]|[]|
58
+ | STM32U5 |[x]|[]|
59
+ | STM32H7 |[x]|[x]|
60
+ | STM32MP1 |[x]|[x]|
61
+ | STM32MP2 |[x]|[x]|
62
+ | STM32N6 |[x]|[x]|
63
+
64
+ # Performances
65
+
66
+ ## Metrics
67
+
68
+ Measures are done with default STM32Cube.AI configuration with enabled input / output allocated option.
69
+
70
+
71
+ ### Reference **NPU** memory footprint on food-101 and ImageNet dataset (see Accuracy for details on dataset)
72
+ |Model | Dataset | Format | Resolution | Series | Internal RAM | External RAM | Weights Flash | STM32Cube.AI version | STEdgeAI Core version |
73
+ |----------|------------------|--------|-------------|------------------|------------------|---------------------|-------|----------------------|-------------------------|
74
+ | [MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_0.25_224_fft/mobilenet_v1_0.25_224_fft_int8.tflite) | food-101 | Int8 | 224x224x3 | STM32N6 | 588 | 0.0 | 321.66 | 10.0.0 | 2.0.0 |
75
+ | [MobileNet v1 0.5 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_0.5_224_fft/mobilenet_v1_0.5_224_fft_int8.tflite) | food-101 | Int8 | 224x224x3 | STM32N6 | 588 | 0.0 | 1025.63 | 10.0.0 | 2.0.0 |
76
+ | [MobileNet v1 1.0 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_1.0_224_fft/mobilenet_v1_1.0_224_fft_int8.tflite) | food-101 | Int8 | 224x224x3 | STM32N6 | 1568 | 0.0 | 3649.97 | 10.0.0 | 2.0.0 |
77
+ | [MobileNet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v1_0.25_224/mobilenet_v1_0.25_224_int8.tflite) | ImageNet | Int8 | 224x224x3 | STM32N6 | 588 | 0.0 | 549.88 | 10.0.0 | 2.0.0 |
78
+ | [MobileNet v1 0.5](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v1_0.5_224/mobilenet_v1_0.5_224_int8.tflite) | ImageNet | Int8 | 224x224x3 | STM32N6 | 588 | 0.0 | 1478.58 | 10.0.0 | 2.0.0 |
79
+ | [MobileNet v1 1.0](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v1_1.0_224/mobilenet_v1_1.0_224_int8.tflite) | ImageNet | Int8 | 224x224x3 | STM32N6 | 1568 | 0.0 | 4552.42 | 10.0.0 | 2.0.0 |
80
+
81
+ ### Reference **NPU** inference time on food-101 and ImageNet dataset (see Accuracy for details on dataset)
82
+
83
+
84
+ | Model | Dataset | Format | Resolution | Board | Execution Engine | Inference time (ms) | Inf / sec | STM32Cube.AI version | STEdgeAI Core version |
85
+ |--------|------------------|--------|-------------|------------------|------------------|---------------------|-------|----------------------|-------------------------|
86
+ | [MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_0.25_224_fft/mobilenet_v1_0.25_224_fft_int8.tflite) | food-101 | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 2.83 | 353.36 | 10.0.0 | 2.0.0 |
87
+ | [MobileNet v1 0.5 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_0.5_224_fft/mobilenet_v1_0.5_224_fft_int8.tflite) | food-101 | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 6.06 | 165.02 | 10.0.0 | 2.0.0 |
88
+ | [MobileNet v1 1.0 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_1.0_224_fft/mobilenet_v1_1.0_224_fft_int8.tflite) | food-101 | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 16.94 | 59.03| 10.0.0 | 2.0.0 |
89
+ | [MobileNet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v1_0.25_224/mobilenet_v1_0.25_224_int8.tflite) | food-101 | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 3.57 | 280.11 | 10.0.0 | 2.0.0 |
90
+ | [MobileNet v1 0.5](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v1_0.5_224/mobilenet_v1_0.5_224_int8.tflite) | food-101 | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 7.38 | 135.50 | 10.0.0 | 2.0.0 |
91
+ | [MobileNet v1 1.0](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v1_1.0_224/mobilenet_v1_1.0_224_int8.tflite) | food-101 | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 19.41 | 51.53 | 10.0.0 | 2.0.0 |
92
+
93
+
94
+ ### Reference **MCU** memory footprint based on Flowers dataset and ImageNet dataset (see Accuracy for details on dataset)
95
+
96
+ | Model | Format | Resolution | Series | Activation RAM | Runtime RAM | Weights Flash | Code Flash | Total RAM | Total Flash | STM32Cube.AI version |
97
+ |--------------------------------------------------------------------------------------------------------------------------------------|--------|------------|---------|----------------|-------------|---------------|------------|-------------|-------------|-----------------------|
98
+ | [MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_224_fft/mobilenet_v1_0.25_224_fft_int8.tflite) | Int8 | 224x224x3 | STM32H7 | 272.96 KiB | 16.38 KiB | 214.69 KiB | 68.05 KiB | 289.34 KiB | 282.74 KiB | 10.0.0 |
99
+ | [MobileNet v1 0.5 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.5_224_fft/mobilenet_v1_0.5_224_fft_int8.tflite) | Int8 | 224x224x3 | STM32H7 | 449.58 KiB | 16.38 KiB | 812.61 KiB | 81.46 KiB | 465.96 KiB | 894.02 KiB | 10.0.0 |
100
+ | [MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_96_fft/mobilenet_v1_0.25_96_fft_int8.tflite) | Int8 | 96x96x3 | STM32H7 | 66.96 KiB | 16.33 KiB | 214.69 KiB | 68 KiB | 83.29 KiB | 282.69 KiB | 10.0.0 |
101
+ | [MobileNet v1 0.25 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_96_grayscale_tfs/mobilenet_v1_0.25_96_grayscale_tfs_int8.tflite) | Int8 | 96x96x1 | STM32H7 | 52.8 KiB | 16.33 KiB | 214.55 KiB | 70.27 KiB | 69.13 KiB | 284.82 KiB | 10.0.0 |
102
+ | [MobileNet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v1_0.25_224/mobilenet_v1_0.25_224_int8.tflite) | Int8 | 224x224x3 | STM32H7 | 272.96 KiB | 16.43 KiB | 467.33 KiB | 70.02 KiB | 283.63 KiB | 537.35 KiB | 10.0.0 |
103
+ | [MobileNet v1 0.5](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v1_0.5_224/mobilenet_v1_0.5_224_int8.tflite) | Int8 | 224x224x3 | STM32H7 | 431.07 KiB | 16.43 KiB | 1314 KiB | 83.38 KiB | 447.5 KiB | 1397.38 KiB | 10.0.0 |
104
+ | [MobileNet v1 1.0](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v1_1.0_224/mobilenet_v1_1.0_224_int8.tflite) | Int8 | 224x224x3 | STM32H7 | 1331.13 KiB | 16.48 KiB | 4157.09 KiB | 110.11 KiB | 1347.61 KiB | 4267.2 KiB | 10.0.0 |
105
+
106
+
107
+ ### Reference **MCU** inference time based on Flowers dataset and ImageNet dataset (see Accuracy for details on dataset)
108
+
109
+
110
+ | Model | Format | Resolution | Board | Execution Engine | Frequency | Inference time (ms) | STM32Cube.AI version |
111
+ |-------------------|--------|------------|------------------|------------------|-----------|------------------|-----------------------|
112
+ | [MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_224_fft/mobilenet_v1_0.25_224_fft_int8.tflite) | Int8 | 224x224x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 163.78 ms | 10.0.0 |
113
+ | [MobileNet v1 0.5 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.5_224_fft/mobilenet_v1_0.5_224_fft_int8.tflite) | Int8 | 224x224x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 485.79 ms | 10.0.0 |
114
+ | [MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_96_fft/mobilenet_v1_0.25_96_fft_int8.tflite) | Int8 | 96x96x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 29.94 ms | 10.0.0 |
115
+ | [MobileNet v1 0.25 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_96_grayscale_tfs/mobilenet_v1_0.25_96_grayscale_tfs_int8.tflite) | Int8 | 96x96x1 | STM32H747I-DISCO | 1 CPU | 400 MHz | 28.34 ms | 10.0.0 |
116
+ | [MobileNet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v1_0.25_224/mobilenet_v1_0.25_224_int8.tflite) | Int8 | 224x224x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 166.75 ms | 10.0.0 |
117
+ | [MobileNet v1 0.5](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v1_0.5_224/mobilenet_v1_0.5_224_int8.tflite) | Int8 | 224x224x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 504.37 ms | 10.0.0 |
118
+ | [MobileNet v1 1.0](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v1_1.0_224/mobilenet_v1_1.0_224_int8.tflite) | Int8 | 224x224x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 1641.84 ms | 10.0.0 |
119
+
120
+
121
+ ### Reference **MPU** inference time based on Flowers dataset (see Accuracy for details on dataset)
122
+ | Model | Format | Resolution | Quantization | Board | Execution Engine | Frequency | Inference time (ms) | %NPU | %GPU | %CPU | X-LINUX-AI version | Framework |
123
+ |-----------------------|--------|------------|---------------|-------------------|------------------|-----------|---------------------|-------|-------|------|--------------------|-----------------------|
124
+ | [MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_224_fft/mobilenet_v1_0.25_224_fft_int8.tflite) | Int8 | 224x224x3 | per-channel** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 14.29 ms | 6.04 | 93.96 | 0 | v5.1.0 | OpenVX |
125
+ | [MobileNet v1 0.5 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.5_224_fft/mobilenet_v1_0.5_224_fft_int8.tflite) | Int8 | 224x224x3 | per-channel** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 32.74 ms | 3.41 | 96.59 | 0 | v5.1.0 | OpenVX |
126
+ | [MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_96_fft/mobilenet_v1_0.25_96_fft_int8.tflite) | Int8 | 96x96x3 | per-channel** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 3.740 ms | 14.20 | 85.80 | 0 | v5.1.0 | OpenVX |
127
+ | [MobileNet v1 0.25 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_96_grayscale_tfs/mobilenet_v1_0.25_96_grayscale_tfs_int8.tflite) | Int8 | 96x96x1 | per-channel** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 3.68 ms | 11.47 | 88.53 | 0 | v5.1.0 | OpenVX |
128
+ | [MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_224_fft/mobilenet_v1_0.25_224_fft_int8.tflite) | Int8 | 224x224x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 33.97 ms | NA | NA | 100 | v5.1.0 | TensorFlowLite 2.11.0 |
129
+ | [MobileNet v1 0.5 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.5_224_fft/mobilenet_v1_0.5_224_fft_int8.tflite) | Int8 | 224x224x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 91.42 ms | NA | NA | 100 | v5.1.0 | TensorFlowLite 2.11.0 |
130
+ | [MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_96_fft/mobilenet_v1_0.25_96_fft_int8.tflite) | Int8 | 96x96x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 6.40 ms | NA | NA | 100 | v5.1.0 | TensorFlowLite 2.11.0 |
131
+ | [MobileNet v1 0.25 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_96_grayscale_tfs/mobilenet_v1_0.25_96_grayscale_tfs_int8.tflite) | Int8 | 96x96x1 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 5.83 ms | NA | NA | 100 | v5.1.0 | TensorFlowLite 2.11.0 |
132
+ |[MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_224_fft/mobilenet_v1_0.25_224_fft_int8.tflite) | Int8 | 224x224x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 52.51 ms | NA | NA | 100 | v5.1.0 | TensorFlowLite 2.11.0 |
133
+ |[MobileNet v1 0.5 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.5_224_fft/mobilenet_v1_0.5_224_fft_int8.tflite) | Int8 | 224x224x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 145.4 ms | NA | NA | 100 | v5.1.0 | TensorFlowLite 2.11.0 |
134
+ |[MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_96_fft/mobilenet_v1_0.25_96_fft_int8.tflite) | Int8 | 96x96x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 9.75 ms | NA | NA | 100 | v5.1.0 | TensorFlowLite 2.11.0 |
135
+ |[MobileNet v1 0.25 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_96_grayscale_tfs/mobilenet_v1_0.25_96_grayscale_tfs_int8.tflite) | Int8 | 96x96x1 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 9.01 ms | NA | NA | 100 | v5.1.0 | TensorFlowLite 2.11.0 |
136
+
137
+ ** **To get the most out of MP25 NPU hardware acceleration, please use per-tensor quantization**
138
+
139
+ ### Accuracy with Flowers dataset
140
+
141
+
142
+ Dataset details: [link](http://download.tensorflow.org/example_images/flower_photos.tgz) , License [CC BY 2.0](https://creativecommons.org/licenses/by/2.0/) , Quotation[[1]](#1) , Number of classes: 5, Number of images: 3 670
143
+
144
+ | Model | Format | Resolution | Top 1 Accuracy |
145
+ |-------|--------|------------|----------------|
146
+ | [MobileNet v1 0.25 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_224_tfs/mobilenet_v1_0.25_224_tfs.h5) | Float | 224x224x3 | 88.83 % |
147
+ | [MobileNet v1 0.25 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_224_tfs/mobilenet_v1_0.25_224_tfs_int8.tflite) | Int8 | 224x224x3 | 89.37 % |
148
+ | [MobileNet v1 0.25 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_224_tl/mobilenet_v1_0.25_224_tl.h5) | Float | 224x224x3 | 85.83 % |
149
+ | [MobileNet v1 0.25 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_224_tl/mobilenet_v1_0.25_224_tl_int8.tflite) | Int8 | 224x224x3 | 83.24 % |
150
+ | [MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_224_fft/mobilenet_v1_0.25_224_fft.h5) | Float | 224x224x3 | 93.05 % |
151
+ | [MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_224_fft/mobilenet_v1_0.25_224_fft_int8.tflite) | Int8 | 224x224x3 | 92.1 % |
152
+ | [MobileNet v1 0.5 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.5_224_tfs/mobilenet_v1_0.5_224_tfs.h5) | Float | 224x224x3 | 92.1 % |
153
+ | [MobileNet v1 0.5 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.5_224_tfs/mobilenet_v1_0.5_224_tfs_int8.tflite) | Int8 | 224x224x3 | 91.55 % |
154
+ | [MobileNet v1 0.5 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.5_224_tl/mobilenet_v1_0.5_224_tl.h5) | Float | 224x224x3 | 88.56 % |
155
+ | [MobileNet v1 0.5 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.5_224_tl/mobilenet_v1_0.5_224_tl_int8.tflite) | Int8 | 224x224x3 | 87.74 % |
156
+ | [MobileNet v1 0.5 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.5_224_fft/mobilenet_v1_0.5_224_fft.h5) | Float | 224x224x3 | 95.1 % |
157
+ | [MobileNet v1 0.5 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.5_224_fft/mobilenet_v1_0.5_224_fft_int8.tflite) | Int8 | 224x224x3 | 94.41 % |
158
+ | [MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_96_fft/mobilenet_v1_0.25_96_fft.h5) | Float | 96x96x3 | 87.47 % |
159
+ | [MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_96_fft/mobilenet_v1_0.25_96_fft_int8.tflite) | Int8 | 96x96x3 | 87.06 % |
160
+ | [MobileNet v1 0.25 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_96_grayscale_tfs/mobilenet_v1_0.25_96_grayscale_tfs.h5) | Float | 96x96x1 | 74.93 % |
161
+ | [MobileNet v1 0.25 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v1_0.25_96_grayscale_tfs/mobilenet_v1_0.25_96_grayscale_tfs_int8.tflite) | Int8 | 96x96x1 | 74.93 % |
162
+
163
+
164
+
165
+ ### Accuracy with Plant-village dataset
166
+
167
+
168
+ Dataset details: [link](https://data.mendeley.com/datasets/tywbtsjrjv/1) , License [CC0 1.0](https://creativecommons.org/publicdomain/zero/1.0/), Quotation[[2]](#2) , Number of classes: 39, Number of images: 61 486
169
+
170
+ | Model | Format | Resolution | Top 1 Accuracy |
171
+ |-------|--------|------------|----------------|
172
+ | [MobileNet v1 0.25 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v1_0.25_224_tfs/mobilenet_v1_0.25_224_tfs.h5) | Float | 224x224x3 | 99.92 % |
173
+ | [MobileNet v1 0.25 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v1_0.25_224_tfs/mobilenet_v1_0.25_224_tfs_int8.tflite) | Int8 | 224x224x3 | 99.92 % |
174
+ | [MobileNet v1 0.25 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v1_0.25_224_tl/mobilenet_v1_0.25_224_tl.h5) | Float | 224x224x3 | 85.38 % |
175
+ | [MobileNet v1 0.25 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v1_0.25_224_tl/mobilenet_v1_0.25_224_tl_int8.tflite) | Int8 | 224x224x3 | 83.7 % |
176
+ | [MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v1_0.25_224_fft/mobilenet_v1_0.25_224_fft.h5) | Float | 224x224x3 | 99.95 % |
177
+ | [MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v1_0.25_224_fft/mobilenet_v1_0.25_224_fft_int8.tflite) | Int8 | 224x224x3 | 99.82 % |
178
+ | [MobileNet v1 0.5 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v1_0.5_224_tfs/mobilenet_v1_0.5_224_tfs.h5) | Float | 224x224x3 | 99.9 % |
179
+ | [MobileNet v1 0.5 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v1_0.5_224_tfs/mobilenet_v1_0.5_224_tfs_int8.tflite) | Int8 | 224x224x3 | 99.83 % |
180
+ | [MobileNet v1 0.5 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v1_0.5_224_tl/mobilenet_v1_0.5_224_tl.h5) | Float | 224x224x3 | 93.05 % |
181
+ | [MobileNet v1 0.5 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v1_0.5_224_tl/mobilenet_v1_0.5_224_tl_int8.tflite) | Int8 | 224x224x3 | 92.7 % |
182
+ | [MobileNet v1 0.5 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v1_0.5_224_fft/mobilenet_v1_0.5_224_fft.h5) | Float | 224x224x3 | 99.94 % |
183
+ | [MobileNet v1 0.5 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v1_0.5_224_fft/mobilenet_v1_0.5_224_fft_int8.tflite) | Int8 | 224x224x3 | 99.85 % |
184
+
185
+
186
+ ### Accuracy with Food-101 dataset
187
+
188
+
189
+ Dataset details: [link](https://data.vision.ee.ethz.ch/cvl/datasets_extra/food-101/) , License [-](), Quotation[[3]](#3) , Number of classes: 101 , Number of images: 101 000
190
+
191
+ | Model | Format | Resolution | Top 1 Accuracy |
192
+ |-------|--------|------------|----------------|
193
+ | [MobileNet v1 0.25 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_0.25_224_tfs/mobilenet_v1_0.25_224_tfs.h5) | Float | 224x224x3 | 72.16 % |
194
+ | [MobileNet v1 0.25 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_0.25_224_tfs/mobilenet_v1_0.25_224_tfs_int8.tflite) | Int8 | 224x224x3 | 71.13 % |
195
+ | [MobileNet v1 0.25 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_0.25_224_tl/mobilenet_v1_0.25_224_tl.h5) | Float | 224x224x3 | 43.21 % |
196
+ | [MobileNet v1 0.25 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_0.25_224_tl/mobilenet_v1_0.25_224_tl_int8.tflite) | Int8 | 224x224x3 | 39.89 % |
197
+ | [MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_0.25_224_fft/mobilenet_v1_0.25_224_fft.h5) | Float | 224x224x3 | 72.36 % |
198
+ | [MobileNet v1 0.25 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_0.25_224_fft/mobilenet_v1_0.25_224_fft_int8.tflite) | Int8 | 224x224x3 | 69.52 % |
199
+ | [MobileNet v1 0.5 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_0.5_224_tfs/mobilenet_v1_0.5_224_tfs.h5) | Float | 224x224x3 | 76.97 % |
200
+ | [MobileNet v1 0.5 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_0.5_224_tfs/mobilenet_v1_0.5_224_tfs_int8.tflite) | Int8 | 224x224x3 | 76.37 % |
201
+ | [MobileNet v1 0.5 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_0.5_224_tl/mobilenet_v1_0.5_224_tl.h5) | Float | 224x224x3 | 48.78 % |
202
+ | [MobileNet v1 0.5 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_0.5_224_tl/mobilenet_v1_0.5_224_tl_int8.tflite) | Int8 | 224x224x3 | 45.89 % |
203
+ | [MobileNet v1 0.5 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_0.5_224_fft/mobilenet_v1_0.5_224_fft.h5) | Float | 224x224x3 | 76.72 % |
204
+ | [MobileNet v1 0.5 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_0.5_224_fft/mobilenet_v1_0.5_224_fft_int8.tflite) | Int8 | 224x224x3 | 74.82 % |
205
+ | [MobileNet v1 1.0 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_1.0_224_fft/mobilenet_v1_1.0_224_fft.h5) | Float | 224x224x3 | 80.38 % |
206
+ | [MobileNet v1 1.0 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v1_1.0_224_fft/mobilenet_v1_1.0_224_fft_int8.tflite) | Int8 | 224x224x3 | 79.43 % |
207
+
208
+
209
+ ### Accuracy with ImageNet dataset
210
+
211
+ Dataset details: [link](https://www.image-net.org), License: BSD-3-Clause, Quotation[[4]](#4)
212
+ Number of classes: 1000.
213
+ To perform the quantization, we calibrated the activations with a random subset of the training set.
214
+ For the sake of simplicity, the accuracy reported here was estimated on the 50000 labelled images of the validation set.
215
+
216
+ |model | Format | Resolution | Top 1 Accuracy |
217
+ |---------|--------|------------|----------------|
218
+ | [MobileNet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v1_0.25_224/mobilenet_v1_0.25_224.h5) | Float | 224x224x3 | 48.96 % |
219
+ | [MobileNet v1 0.25](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v1_0.25_224/mobilenet_v1_0.25_224_int8.tflite) | Int8 | 224x224x3 | 46.34 % |
220
+ | [MobileNet v1 0.5](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v1_0.5_224/mobilenet_v1_0.5_224.h5) | Float | 224x224x3 | 62.11 % |
221
+ | [MobileNet v1 0.5](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v1_0.5_224/mobilenet_v1_0.5_224_int8.tflite) | Int8 | 224x224x3 | 59.92 % |
222
+ | [MobileNet v1 1.0](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v1_1.0_224/mobilenet_v1_1.0_224.h5) | Float | 224x224x3 | 69.52 % |
223
+ | [MobileNet v1 1.0](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv1/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v1_1.0_224/mobilenet_v1_1.0_224_int8.tflite) | Int8 | 224x224x3 | 68.64 % |
224
+
225
+
226
+
227
+ ## Retraining and Integration in a simple example:
228
+
229
+ Please refer to the stm32ai-modelzoo-services GitHub [here](https://github.com/STMicroelectronics/stm32ai-modelzoo-services)
230
+
231
+
232
+
233
+ # References
234
+
235
+ <a id="1">[1]</a>
236
+ "Tf_flowers : tensorflow datasets," TensorFlow. [Online]. Available: https://www.tensorflow.org/datasets/catalog/tf_flowers.
237
+
238
+ <a id="2">[2]</a>
239
+ J, ARUN PANDIAN; GOPAL, GEETHARAMANI (2019), "Data for: Identification of Plant Leaf Diseases Using a 9-layer Deep Convolutional Neural Network", Mendeley Data, V1, doi: 10.17632/tywbtsjrjv.1
240
+
241
+ <a id="3">[3]</a>
242
+ L. Bossard, M. Guillaumin, and L. Van Gool, "Food-101 -- Mining Discriminative Components with Random Forests." European Conference on Computer Vision, 2014.