Hugging Face's logo

STMicroelectronics
/
efficientnet

Image Classification
Model card Files Community
FBAGSTM commited on
Commit
7615412
·
verified ·
1 Parent(s): ac1726c

Update ST Model Zoo

Browse files
Files changed (1) hide show
  1. README.md +23 -29
README.md CHANGED
@@ -1,10 +1,3 @@
1
- ---
2
- license: other
3
- license_name: sla0044
4
- license_link: >-
5
- https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/efficientnet/ST_pretrainedmodel_public_dataset/LICENSE.md
6
- pipeline_tag: image-classification
7
- ---
8
  # EfficientNet
9
 
10
  ## **Use case** : `Image classification`
@@ -64,48 +57,49 @@ For an image resolution of NxM and P classes :
64
  # Performances
65
 
66
  ## Metrics
 
67
  * Measures are done with default STM32Cube.AI configuration with enabled input / output allocated option.
68
 
69
  * `tfs` stands for "training from scratch", meaning that the model weights were randomly initialized before training.
70
 
71
  ### Reference **NPU** memory footprint on food-101 dataset (see Accuracy for details on dataset)
72
- |Model | Format | Resolution | Series | Internal RAM (KiB) | External RAM (KiB) | Weights Flash (KiB)| STM32Cube.AI version | STEdgeAI Core version |
73
- |----------|--------|-------------|------------------|------------------|---------------------|-------|----------------------|-------------------------|
74
- | [ST EfficientNet LC v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/efficientnet/ST_pretrainedmodel_public_dataset/food-101/st_efficientnet_lc_v1_128_tfs/st_efficientnet_lc_v1_128_tfs_int8.tflite) | Int8 | 128x128x3 | STM32N6 | 256 | 0 | 625.8 | 10.0.0 | 2.0.0 |
75
- | [ST EfficientNet LC v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/efficientnet/ST_pretrainedmodel_public_dataset/food-101/st_efficientnet_lc_v1_224_tfs/st_efficientnet_lc_v1_224_tfs_int8.tflite) | Int8 | 224x224x3 | STM32N6 | 784.02 | 0 | 632.55 | 10.0.0 | 2.0.0 |
76
 
77
  ### Reference **NPU** inference time on food-101 dataset (see Accuracy for details on dataset)
78
- | Model | Format | Resolution | Board | Execution Engine | Inference time (ms) | Inf / sec | STM32Cube.AI version | STEdgeAI Core version |
79
- |--------|--------|-------------|------------------|------------------|---------------------|-------|----------------------|-------------------------|
80
- | [ST EfficientNet LC v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/efficientnet/ST_pretrainedmodel_public_dataset/food-101/st_efficientnet_lc_v1_128_tfs/st_efficientnet_lc_v1_128_tfs_int8.tflite)| Int8 | 128x128x3 | STM32N6570-DK | NPU/MCU | 6.87 | 145.55 | 10.0.0 | 2.0.0 |
81
- | [ST EfficientNet LC v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/efficientnet/ST_pretrainedmodel_public_dataset/food-101/st_efficientnet_lc_v1_224_tfs/st_efficientnet_lc_v1_224_tfs_int8.tflite) | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 15.8 | 63.29 | 10.0.0 | 2.0.0 |
82
 
83
 
84
  ### Reference **MCU** memory footprints based on Flowers dataset (see Accuracy for details on dataset)
85
  | Model | Format | Resolution | Series | Activation RAM | Runtime RAM | Weights Flash | Code Flash | Total RAM | Total Flash | STM32Cube.AI version |
86
  |---------------------------|--------|--------------|---------|----------------|-------------|---------------|------------|------------|-------------|----------------------|
87
- | ST EfficientNet LC v1 tfs | Int8 | 224x224x3 | STM32H7 | 430.78 KiB | 58.19 KiB | 505.41 KiB | 158.4 KiB | 488.97 KiB | 663.81 KiB | 10.0.0 |
88
- | ST EfficientNet LC v1 tfs | Int8 | 128x128x3 | STM32H7 | 166.78 KiB | 57.86 KiB | 505.41 KiB | 157.68 KiB| 224.64 KiB | 663.09 KiB | 10.0.0 |
89
 
90
 
91
  ### Reference **MCU** inference time based on Flowers dataset (see Accuracy for details on dataset)
92
  | Model | Format | Resolution | Board | Execution Engine | Frequency | Inference time (ms) | STM32Cube.AI version |
93
  |---------------------------|--------|------------|-------------------|------------------|-----------|---------------------|----------------------|
94
- | ST EfficientNet LC v1 tfs | Int8 | 224x224x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 438.33 ms | 10.0.0 |
95
- | ST EfficientNet LC v1 tfs | Int8 | 128x128x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 144.96 ms | 10.0.0 |
96
- | ST EfficientNet LC v1 tfs | Int8 | 224x224x3 | STM32F769I-DISCO | 1 CPU | 216 MHz | 871.7 ms | 10.0.0 |
97
- | ST EfficientNet LC v1 tfs | Int8 | 128x128x3 | STM32F769I-DISCO | 1 CPU | 216 MHz | 259.5 ms | 10.0.0 |
98
 
99
 
100
  ### Reference **MPU** inference time based on Flowers dataset (see Accuracy for details on dataset)
101
  | Model | Format | Resolution | Quantization | Board | Execution Engine | Frequency | Inference time (ms) | %NPU | %GPU | %CPU | X-LINUX-AI version | Framework |
102
  |---------------------------|--------|------------|---------------|-------------------|------------------|-----------|---------------------|-------|-------|------|--------------------|-----------------------|
103
- | ST EfficientNet LC v1 tfs | Int8 | 224x224x3 | per-channel** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 36.75 ms | 16.89 | 83.11 | 0 | v5.1.0 | OpenVX |
104
- | ST EfficientNet LC v1 tfs | Int8 | 128x128x3 | per-channel** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 14.67 ms | 32.55 | 67.45 | 0 | v5.1.0 | OpenVX |
105
- | ST EfficientNet LC v1 tfs | Int8 | 224x224x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 140.6 ms | NA | NA | 100 | v5.1.0 | TensorFlowLite 2.11.0 |
106
- | ST EfficientNet LC v1 tfs | Int8 | 128x128x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 47.50 ms | NA | NA | 100 | v5.1.0 | TensorFlowLite 2.11.0 |
107
- | ST EfficientNet LC v1 tfs | Int8 | 224x224x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 198.7 ms | NA | NA | 100 | v5.1.0 | TensorFlowLite 2.11.0 |
108
- | ST EfficientNet LC v1 tfs | Int8 | 128x128x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 63.84 ms | NA | NA | 100 | v5.1.0 | TensorFlowLite 2.11.0 |
109
 
110
  ** **To get the most out of MP25 NPU hardware acceleration, please use per-tensor quantization**
111
 
@@ -139,9 +133,9 @@ Number of classes: 101, number of files: 101000
139
 
140
  | Model | Format | Resolution | Top 1 Accuracy (%) |
141
  |---------------------------|--------|------------|--------------------|
142
- | ST EfficientNet LC v1 tfs | Float | 224x224x3 | 74.84 |
143
  | ST EfficientNet LC v1 tfs | Int8 | 224x224x3 | 74.44 |
144
- | ST EfficientNet LC v1 tfs | Float | 128x128x3 | 63.58 |
145
  | ST EfficientNet LC v1 tfs | Int8 | 128x128x3 | 63.07 |
146
 
147
 
 
 
 
 
 
 
 
 
1
  # EfficientNet
2
 
3
  ## **Use case** : `Image classification`
 
57
  # Performances
58
 
59
  ## Metrics
60
+
61
  * Measures are done with default STM32Cube.AI configuration with enabled input / output allocated option.
62
 
63
  * `tfs` stands for "training from scratch", meaning that the model weights were randomly initialized before training.
64
 
65
  ### Reference **NPU** memory footprint on food-101 dataset (see Accuracy for details on dataset)
66
+ |Model | Format | Resolution | Series | Internal RAM (KiB) | External RAM (KiB) | Weights Flash (KiB) | STM32Cube.AI version | STEdgeAI Core version |
67
+ |----------|--------|-------------|------------------|------------------|---------------------|---------------------|----------------------|-------------------------|
68
+ | [ST EfficientNet LC v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/efficientnet/ST_pretrainedmodel_public_dataset/food-101/st_efficientnet_lc_v1_128_tfs/st_efficientnet_lc_v1_128_tfs_int8.tflite) | Int8 | 128x128x3 | STM32N6 | 256 | 0 | 579.69 | 10.2.0 | 2.2.0 |
69
+ | [ST EfficientNet LC v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/efficientnet/ST_pretrainedmodel_public_dataset/food-101/st_efficientnet_lc_v1_224_tfs/st_efficientnet_lc_v1_224_tfs_int8.tflite) | Int8 | 224x224x3 | STM32N6 | 784.02 | 0 | 586.44 | 10.2.0 | 2.2.0 |
70
 
71
  ### Reference **NPU** inference time on food-101 dataset (see Accuracy for details on dataset)
72
+ | Model | Format | Resolution | Board | Execution Engine | Inference time (ms) | Inf / sec | STM32Cube.AI version | STEdgeAI Core version |
73
+ |--------|--------|-------------|------------------|------------------|---------------------|-----------|----------------------|-------------------------|
74
+ | [ST EfficientNet LC v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/efficientnet/ST_pretrainedmodel_public_dataset/food-101/st_efficientnet_lc_v1_128_tfs/st_efficientnet_lc_v1_128_tfs_int8.tflite)| Int8 | 128x128x3 | STM32N6570-DK | NPU/MCU | 6.88 | 145.34 | 10.2.0 | 2.2.0 |
75
+ | [ST EfficientNet LC v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/efficientnet/ST_pretrainedmodel_public_dataset/food-101/st_efficientnet_lc_v1_224_tfs/st_efficientnet_lc_v1_224_tfs_int8.tflite) | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 15.76 | 63.45 | 10.2.0 | 2.2.0 |
76
 
77
 
78
  ### Reference **MCU** memory footprints based on Flowers dataset (see Accuracy for details on dataset)
79
  | Model | Format | Resolution | Series | Activation RAM | Runtime RAM | Weights Flash | Code Flash | Total RAM | Total Flash | STM32Cube.AI version |
80
  |---------------------------|--------|--------------|---------|----------------|-------------|---------------|------------|------------|-------------|----------------------|
81
+ | ST EfficientNet LC v1 tfs | Int8 | 224x224x3 | STM32H7 | 430.78 KiB | 58.19 KiB | 505.41 KiB | 158.4 KiB | 488.97 KiB | 663.81 KiB | 10.2.0 |
82
+ | ST EfficientNet LC v1 tfs | Int8 | 128x128x3 | STM32H7 | 166.78 KiB | 57.86 KiB | 505.41 KiB | 156.74 KiB | 224.64 KiB | 662.15 KiB | 10.2.0 |
83
 
84
 
85
  ### Reference **MCU** inference time based on Flowers dataset (see Accuracy for details on dataset)
86
  | Model | Format | Resolution | Board | Execution Engine | Frequency | Inference time (ms) | STM32Cube.AI version |
87
  |---------------------------|--------|------------|-------------------|------------------|-----------|---------------------|----------------------|
88
+ | ST EfficientNet LC v1 tfs | Int8 | 224x224x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 438.33 ms | 10.2.0 |
89
+ | ST EfficientNet LC v1 tfs | Int8 | 128x128x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 147.43 ms | 10.2.0 |
90
+ | ST EfficientNet LC v1 tfs | Int8 | 224x224x3 | STM32F769I-DISCO | 1 CPU | 216 MHz | 871.7 ms | 10.2.0 |
91
+ | ST EfficientNet LC v1 tfs | Int8 | 128x128x3 | STM32F769I-DISCO | 1 CPU | 216 MHz | 259.5 ms | 10.2.0 |
92
 
93
 
94
  ### Reference **MPU** inference time based on Flowers dataset (see Accuracy for details on dataset)
95
  | Model | Format | Resolution | Quantization | Board | Execution Engine | Frequency | Inference time (ms) | %NPU | %GPU | %CPU | X-LINUX-AI version | Framework |
96
  |---------------------------|--------|------------|---------------|-------------------|------------------|-----------|---------------------|-------|-------|------|--------------------|-----------------------|
97
+ | ST EfficientNet LC v1 tfs | Int8 | 224x224x3 | per-channel** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 36.82 ms | 14.72 | 85.28 | 0 | v6.1.0 | OpenVX |
98
+ | ST EfficientNet LC v1 tfs | Int8 | 128x128x3 | per-channel** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 14.81 ms | 29.68 | 70.32 | 0 | v6.1.0 | OpenVX |
99
+ | ST EfficientNet LC v1 tfs | Int8 | 224x224x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 137.34 ms | NA | NA | 100 | v6.1.0 | TensorFlowLite 2.18.0 |
100
+ | ST EfficientNet LC v1 tfs | Int8 | 128x128x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 45.80 ms | NA | NA | 100 | v6.1.0 | TensorFlowLite 2.18.0 |
101
+ | ST EfficientNet LC v1 tfs | Int8 | 224x224x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 195.25 ms | NA | NA | 100 | v6.1.0 | TensorFlowLite 2.18.0 |
102
+ | ST EfficientNet LC v1 tfs | Int8 | 128x128x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 65.14 ms | NA | NA | 100 | v6.1.0 | TensorFlowLite 2.18.0 |
103
 
104
  ** **To get the most out of MP25 NPU hardware acceleration, please use per-tensor quantization**
105
 
 
133
 
134
  | Model | Format | Resolution | Top 1 Accuracy (%) |
135
  |---------------------------|--------|------------|--------------------|
136
+ | ST EfficientNet LC v1 tfs | Float | 224x224x3 | 74.83 |
137
  | ST EfficientNet LC v1 tfs | Int8 | 224x224x3 | 74.44 |
138
+ | ST EfficientNet LC v1 tfs | Float | 128x128x3 | 63.56 |
139
  | ST EfficientNet LC v1 tfs | Int8 | 128x128x3 | 63.07 |
140
 
141