Hugging Face's logo

STMicroelectronics
/
resnet50v2

Image Classification
Model card Files Community
FBAGSTM commited on
Commit
f3b63e4
·
verified ·
1 Parent(s): 972dde0

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +131 -6
README.md CHANGED
@@ -1,6 +1,131 @@
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
+ pipeline_tag: image-classification
7
+ ---
8
+ # ResNet50 v2
9
+
10
+ ## **Use case** : `Image classification`
11
+
12
+ # Model description
13
+
14
+
15
+ ResNets family is a well known architecture that uses skip connections to enable stronger gradients in much deeper networks. This variant has 50 layers.
16
+
17
+ The model is quantized in int8 using tensorflow lite converter.
18
+
19
+ ## Network information
20
+
21
+
22
+ | Network Information | Value |
23
+ |-------------------------|-----------------|
24
+ | Framework | TensorFlow Lite |
25
+ | MParams | 25.6 M |
26
+ | Quantization | int8 |
27
+ | Provenance | https://www.tensorflow.org/api_docs/python/tf/keras/applications/ResNet50V2 |
28
+ | Paper | https://arxiv.org/abs/1603.05027 |
29
+
30
+ The models are quantized using tensorflow lite converter.
31
+
32
+
33
+ ## Network inputs / outputs
34
+
35
+
36
+ For an image resolution of NxM and P classes
37
+
38
+ | Input Shape | Description |
39
+ | ----- | ----------- |
40
+ | (1, N, M, 3) | Single NxM RGB image with UINT8 values between 0 and 255 |
41
+
42
+ | Output Shape | Description |
43
+ | ----- | ----------- |
44
+ | (1, P) | Per-class confidence for P classes in FLOAT32|
45
+
46
+
47
+ ## Recommended platforms
48
+
49
+
50
+ | Platform | Supported | Recommended |
51
+ |----------|-----------|-----------|
52
+ | STM32L0 |[]|[]|
53
+ | STM32L4 |[]|[]|
54
+ | STM32U5 |[]|[]|
55
+ | STM32H7 |[x]|[]|
56
+ | STM32MP1 |[x]|[]|
57
+ | STM32MP2 |[x]|[x]|
58
+ | STM32N6 |[x]|[x]|
59
+
60
+ # Performances
61
+
62
+ ## Metrics
63
+
64
+
65
+ Measures are done with default STM32Cube.AI configuration with enabled input / output allocated option.
66
+
67
+ ### Reference **NPU** memory footprint on food-101 and ImageNet dataset (see Accuracy for details on dataset)
68
+ |Model | Dataset | Format | Resolution | Series | Internal RAM | External RAM | Weights Flash | STM32Cube.AI version | STEdgeAI Core version |
69
+ |----------|------------------|--------|-------------|------------------|------------------|---------------------|-------|----------------------|-------------------------|
70
+ | [ResNet50 v2 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/resnet50v2/ST_pretrainedmodel_public_dataset/food-101/resnet50_v2_224_fft/resnet50_v2_224_fft_int8.tflite) | food-101 | Int8 | 224x224x3 | STM32N6 | | | | 10.0.0 | 2.0.0 |
71
+ | [ResNet50 v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/resnet50v2/Public_pretrainedmodel_public_dataset/ImageNet/resnet50_v2_224/resnet50_v2_224_int8.tflite) | ImageNet | Int8 | 224x224x3 | STM32N6 | | | | 10.0.0 | 2.0.0 |
72
+
73
+ ### Reference **NPU** inference time on food-101 and ImageNet dataset (see Accuracy for details on dataset)
74
+ | Model | Dataset | Format | Resolution | Board | Execution Engine | Inference time (ms) | Inf / sec | STM32Cube.AI version | STEdgeAI Core version |
75
+ |--------|------------------|--------|-------------|------------------|------------------|---------------------|-------|----------------------|-------------------------|
76
+ | [ResNet50 v2 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/resnet50v2/ST_pretrainedmodel_public_dataset/food-101/resnet50_v2_224_fft/resnet50_v2_224_fft_int8.tflite) | food-101 | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | | | 10.0.0 | 2.0.0 |
77
+ | [ResNet50 v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/resnet50v2/Public_pretrainedmodel_public_dataset/ImageNet/resnet50_v2_224/resnet50_v2_224_int8.tflite) | ImageNet | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | | | 10.0.0 | 2.0.0 |
78
+
79
+
80
+ ### Reference **MCU** memory footprint based on Food-101 and ImageNet dataset (see Accuracy for details on dataset)
81
+
82
+ | Model | Format | Resolution | Series | Activation RAM | Runtime RAM | Weights Flash | Code Flash | Total RAM | Total Flash | STM32Cube.AI version |
83
+ |--------------------------------------------------------------------------------------------------------------------------------------|--------|------------|---------|----------------|-------------|---------------|------------|-------------|-------------|-----------------------|
84
+ | [ResNet50 v2 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/resnet50v2/ST_pretrainedmodel_public_dataset/food-101/resnet50_v2_224_fft/resnet50_v2_224_fft_int8.tflite) | Int8 | 224x224x3 | STM32H7 | 2142.07 KiB | 41.02 KiB | 23240.96 KiB | 226.05 KiB | 2183.09 KiB | 23467.01 KiB | 10.0.0 |
85
+ | [ResNet50 v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/resnet50v2/Public_pretrainedmodel_public_dataset/ImageNet/resnet50_v2_224/resnet50_v2_224_int8.tflite) | Int8 | 224x224x3 | STM32H7 | 2142.07 KiB | 41.02 KiB | 25042.47 KiB | 226.05 KiB | 2183.09 KiB | 25268.52 KiB | 10.0.0 |
86
+
87
+
88
+ ### Reference **MCU** inference time based on Food-101 and ImageNet dataset (see Accuracy for details on dataset)
89
+
90
+
91
+ | Model | Format | Resolution | Board | Execution Engine | Frequency | Inference time (ms) | STM32Cube.AI version |
92
+ |-------------------|--------|------------|------------------|------------------|-----------|------------------|-----------------------|
93
+ | [ResNet50 v2 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/resnet50v2/ST_pretrainedmodel_public_dataset/food-101/resnet50_v2_224_fft/resnet50_v2_224_fft_int8.tflite) | Int8 | 224x224x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 11354.82 ms | 10.0.0 |
94
+ | [ResNet50 v2](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/resnet50v2/Public_pretrainedmodel_public_dataset/ImageNet/resnet50_v2_224/resnet50_v2_224_int8.tflite) | Int8 | 224x224x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 11368.81 ms | 10.0.0 |
95
+
96
+
97
+
98
+ ### Accuracy with Food-101 dataset
99
+
100
+ Dataset details: [link](https://data.vision.ee.ethz.ch/cvl/datasets_extra/food-101/) , License [-](), Quotation[[1]](#1) , Number of classes: 101 , Number of images: 101 000
101
+
102
+ | Model | Format | Resolution | Top 1 Accuracy |
103
+ |-------|--------|------------|----------------|
104
+ | [ResNet50 v2 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/resnet50v2/ST_pretrainedmodel_public_dataset/food-101/resnet50_v2_224_fft/resnet50_v2_224_fft.h5) | Float | 224x224x3 | 71.53 % |
105
+ | [ResNet50 v2 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/resnet50v2/ST_pretrainedmodel_public_dataset/food-101/resnet50_v2_224_fft/resnet50_v2_224_fft_int8.tflite) | Int8 | 224x224x3 | 70.07 % |
106
+
107
+
108
+ ### Accuracy with ImageNet dataset
109
+
110
+ Dataset details: [link](https://www.image-net.org), License: BSD-3-Clause, Quotation[[4]](#4)
111
+ Number of classes: 1000.
112
+ To perform the quantization, we calibrated the activations with a random subset of the training set.
113
+ For the sake of simplicity, the accuracy reported here was estimated on the 50000 labelled images of the validation set.
114
+
115
+ |model | Format | Resolution | Top 1 Accuracy |
116
+ |---------|--------|------------|----------------|
117
+ | [ResNet50 v2 ](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/resnet50v2/Public_pretrainedmodel_public_dataset/ImageNet/resnet50_v2_224/resnet50_v2_224.h5) | Float | 224x224x3 | 66.38 % |
118
+ | [ResNet50 v2 ](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/resnet50v2/Public_pretrainedmodel_public_dataset/ImageNet/resnet50_v2_224/resnet50_v2_224_int8.tflite) | Int8 | 224x224x3 | 65.99 % |
119
+
120
+
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
+
128
+ # References
129
+
130
+ <a id="1">[1]</a>
131
+ L. Bossard, M. Guillaumin, and L. Van Gool, "Food-101 -- Mining Discriminative Components with Random Forests." European Conference on Computer Vision, 2014.