Hugging Face's logo

STMicroelectronics
/
st_mnist

Image Classification
Model card Files Community
FBAGSTM commited on
Commit
07bbf0b
·
verified ·
1 Parent(s): 1ce3a09

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +10 -10
README.md CHANGED
@@ -2,7 +2,7 @@
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
  # ST MNIST v1
@@ -67,7 +67,7 @@ Measures are done with default STM32Cube.AI configuration with enabled input / o
67
 
68
  | Model | Format | Resolution | Series | Activation RAM | Runtime RAM | Weights Flash | Code Flash | Total RAM | Total Flash | STM32Cube.AI version |
69
  |-------------------|--------|------------|---------|----------------|-------------|---------------|------------|-------------|-------------|-----------------------|
70
- | [ST MNIST Byclass v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/st_mnist/ST_pretrainedmodel_public_dataset/emnist_byclass/st_mnist_v1_28_tfs/st_mnist_v1_28_tfs_int8.tflite) | Int8 | 28x28x1 | STM32H7 | 17.21 KiB | 4.49 KiB | 10.08 KiB | 46.8 KiB | 21.7 KiB | 56.88 KiB | 10.0.0 |
71
 
72
 
73
  ### Reference **MCU** inference time based on EMNIST-Byclass dataset (see Accuracy for details on dataset)
@@ -75,15 +75,15 @@ Measures are done with default STM32Cube.AI configuration with enabled input / o
75
 
76
  | Model | Format | Resolution | Board | Frequency | Inference time (ms) | STM32Cube.AI version |
77
  |-------------------|--------|------------|------------------|---------------|---------------------|-----------------------|
78
- | [ST MNIST Byclass v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/st_mnist/ST_pretrainedmodel_public_dataset/emnist_byclass/st_mnist_v1_28_tfs/st_mnist_v1_28_tfs_int8.tflite) | Int8 | 28x28x1 | STM32H747I-DISCO | 400 MHz | 3.41 ms | 10.0.0 |
79
 
80
 
81
  ### Reference **MPU** inference time based on EMNIST-Byclass dataset (see Accuracy for details on dataset)
82
  | Model | Format | Resolution | Quantization | Board | Execution Engine | Frequency | Inference time (ms) | %NPU | %GPU | %CPU | X-LINUX-AI version | Framework |
83
  |---------------------------------------------------------------------------------------------------------------------------------|----------|------------|---------------|-------------------|------------------|-----------|---------------------|-------|-------|------|--------------------|-----------------------|
84
- | [ST MNIST Byclass v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/st_mnist/ST_pretrainedmodel_public_dataset/emnist_byclass/st_mnist_v1_28_tfs/st_mnist_v1_28_tfs_int8.tflite) | Int8 | 28x28x1 | per-channel** | STM32MP257F-DK2 | 2 CPU | 1500 MHz | 0.31 ms | 0 | 0 | 100 | v5.1.0 | TensorFlowLite 2.11.0 |
85
- | [ST MNIST Byclass v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/st_mnist/ST_pretrainedmodel_public_dataset/emnist_byclass/st_mnist_v1_28_tfs/st_mnist_v1_28_tfs_int8.tflite) | Int8 | 28x28x1 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 0.69 ms | NA | NA | 100 | v5.1.0 | TensorFlowLite 2.11.0 |
86
- | [ST MNIST Byclass v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/st_mnist/ST_pretrainedmodel_public_dataset/emnist_byclass/st_mnist_v1_28_tfs/st_mnist_v1_28_tfs_int8.tflite) | Int8 | 28x28x1 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 1.070 ms | NA | NA | 100 | v5.1.0 | TensorFlowLite 2.11.0 |
87
 
88
  ** **To get the most out of MP25 NPU hardware acceleration, please use per-tensor quantization**
89
 
@@ -94,16 +94,16 @@ Dataset details: [link](https://www.nist.gov/itl/products-and-services/emnist-da
94
 
95
  | Model | Format | Resolution | Top 1 Accuracy |
96
  |-------|--------|------------|----------------|
97
- | [ST MNIST Byclass v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/st_mnist/ST_pretrainedmodel_public_dataset/emnist_byclass/st_mnist_v1_28_tfs/st_mnist_v1_28_tfs.h5) | Float | 28x28x1 | 91.89 % |
98
- | [ST MNIST Byclass v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/st_mnist/ST_pretrainedmodel_public_dataset/emnist_byclass/st_mnist_v1_28_tfs/st_mnist_v1_28_tfs_int8.tflite) | Int8 | 28x28x1 | 91.47 % |
99
 
100
  Following we provide the confusion matrix for the model with Float32 weights.
101
 
102
- ![plot](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/st_mnist/doc/img/st_emnist_by_class_confusion_matrix.png)
103
 
104
  Following we provide the confusion matrix for the quantized model with INT8 weights.
105
 
106
- ![plot](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/st_mnist/doc/img/st_emnist_by_class_confusion_matrix_int8.png)
107
 
108
 
109
  ## Retraining and Integration in a simple example:
 
2
  license: other
3
  license_name: sla0044
4
  license_link: >-
5
+ https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/LICENSE.md
6
  pipeline_tag: image-classification
7
  ---
8
  # ST MNIST v1
 
67
 
68
  | Model | Format | Resolution | Series | Activation RAM | Runtime RAM | Weights Flash | Code Flash | Total RAM | Total Flash | STM32Cube.AI version |
69
  |-------------------|--------|------------|---------|----------------|-------------|---------------|------------|-------------|-------------|-----------------------|
70
+ | [ST MNIST Byclass v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/st_mnist/ST_pretrainedmodel_public_dataset/emnist_byclass/st_mnist_v1_28_tfs/st_mnist_v1_28_tfs_int8.tflite) | Int8 | 28x28x1 | STM32H7 | 17.21 KiB | 4.49 KiB | 10.08 KiB | 46.8 KiB | 21.7 KiB | 56.88 KiB | 10.0.0 |
71
 
72
 
73
  ### Reference **MCU** inference time based on EMNIST-Byclass dataset (see Accuracy for details on dataset)
 
75
 
76
  | Model | Format | Resolution | Board | Frequency | Inference time (ms) | STM32Cube.AI version |
77
  |-------------------|--------|------------|------------------|---------------|---------------------|-----------------------|
78
+ | [ST MNIST Byclass v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/st_mnist/ST_pretrainedmodel_public_dataset/emnist_byclass/st_mnist_v1_28_tfs/st_mnist_v1_28_tfs_int8.tflite) | Int8 | 28x28x1 | STM32H747I-DISCO | 400 MHz | 3.41 ms | 10.0.0 |
79
 
80
 
81
  ### Reference **MPU** inference time based on EMNIST-Byclass dataset (see Accuracy for details on dataset)
82
  | Model | Format | Resolution | Quantization | Board | Execution Engine | Frequency | Inference time (ms) | %NPU | %GPU | %CPU | X-LINUX-AI version | Framework |
83
  |---------------------------------------------------------------------------------------------------------------------------------|----------|------------|---------------|-------------------|------------------|-----------|---------------------|-------|-------|------|--------------------|-----------------------|
84
+ | [ST MNIST Byclass v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/st_mnist/ST_pretrainedmodel_public_dataset/emnist_byclass/st_mnist_v1_28_tfs/st_mnist_v1_28_tfs_int8.tflite) | Int8 | 28x28x1 | per-channel** | STM32MP257F-DK2 | 2 CPU | 1500 MHz | 0.31 ms | 0 | 0 | 100 | v5.1.0 | TensorFlowLite 2.11.0 |
85
+ | [ST MNIST Byclass v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/st_mnist/ST_pretrainedmodel_public_dataset/emnist_byclass/st_mnist_v1_28_tfs/st_mnist_v1_28_tfs_int8.tflite) | Int8 | 28x28x1 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 0.69 ms | NA | NA | 100 | v5.1.0 | TensorFlowLite 2.11.0 |
86
+ | [ST MNIST Byclass v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/st_mnist/ST_pretrainedmodel_public_dataset/emnist_byclass/st_mnist_v1_28_tfs/st_mnist_v1_28_tfs_int8.tflite) | Int8 | 28x28x1 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 1.070 ms | NA | NA | 100 | v5.1.0 | TensorFlowLite 2.11.0 |
87
 
88
  ** **To get the most out of MP25 NPU hardware acceleration, please use per-tensor quantization**
89
 
 
94
 
95
  | Model | Format | Resolution | Top 1 Accuracy |
96
  |-------|--------|------------|----------------|
97
+ | [ST MNIST Byclass v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/st_mnist/ST_pretrainedmodel_public_dataset/emnist_byclass/st_mnist_v1_28_tfs/st_mnist_v1_28_tfs.h5) | Float | 28x28x1 | 91.89 % |
98
+ | [ST MNIST Byclass v1 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/st_mnist/ST_pretrainedmodel_public_dataset/emnist_byclass/st_mnist_v1_28_tfs/st_mnist_v1_28_tfs_int8.tflite) | Int8 | 28x28x1 | 91.47 % |
99
 
100
  Following we provide the confusion matrix for the model with Float32 weights.
101
 
102
+ ![plot](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/st_mnist/doc/img/st_emnist_by_class_confusion_matrix.png)
103
 
104
  Following we provide the confusion matrix for the quantized model with INT8 weights.
105
 
106
+ ![plot](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/image_classification/st_mnist/doc/img/st_emnist_by_class_confusion_matrix_int8.png)
107
 
108
 
109
  ## Retraining and Integration in a simple example: