This directory includes code and data to train a Tiny VGG model (inspired by the demo CNN in Stanford CS231n class) on 10 everyday classes from the Tiny ImageNet.
First, you want to unzip data.zip. The file structure would be something like:
.
├── data
│ ├── class_10_train
│ │ ├── n01882714
│ │ │ ├── images [500 entries exceeds filelimit, not opening dir]
│ │ │ └── n01882714_boxes.txt
│ │ ├── n02165456
│ │ │ ├── images [500 entries exceeds filelimit, not opening dir]
│ │ │ └── n02165456_boxes.txt
│ │ ├── n02509815
│ │ │ ├── images [500 entries exceeds filelimit, not opening dir]
│ │ │ └── n02509815_boxes.txt
│ │ ├── n03662601
│ │ │ ├── images [500 entries exceeds filelimit, not opening dir]
│ │ │ └── n03662601_boxes.txt
│ │ ├── n04146614
│ │ │ ├── images [500 entries exceeds filelimit, not opening dir]
│ │ │ └── n04146614_boxes.txt
│ │ ├── n04285008
│ │ │ ├── images [500 entries exceeds filelimit, not opening dir]
│ │ │ └── n04285008_boxes.txt
│ │ ├── n07720875
│ │ │ ├── images [500 entries exceeds filelimit, not opening dir]
│ │ │ └── n07720875_boxes.txt
│ │ ├── n07747607
│ │ │ ├── images [500 entries exceeds filelimit, not opening dir]
│ │ │ └── n07747607_boxes.txt
│ │ ├── n07873807
│ │ │ ├── images [500 entries exceeds filelimit, not opening dir]
│ │ │ └── n07873807_boxes.txt
│ │ └── n07920052
│ │ ├── images [500 entries exceeds filelimit, not opening dir]
│ │ └── n07920052_boxes.txt
│ ├── class_10_val
│ │ ├── test_images [250 entries exceeds filelimit, not opening dir]
│ │ └── val_images [250 entries exceeds filelimit, not opening dir]
│ ├── class_dict_10.json
│ └── val_class_dict_10.json
├── data.zip
├── environment.yaml
└── tiny-vgg.py
To install all dependencies, run the following code
conda env create --file environment.yaml
To train Tiny VGG on these 10 classes, run the following code
python tiny-vgg.py
After training, you will get two saved models in Keras format: trained_tiny_vgg.h5
and trained_vgg_best.h5. The first file is the final model after training, and
trained_vgg_best.h5 is the model having the best validation performance.
You can use either one for CNN Explainer.
Before loading the model using tensorflow.js, you want to convert the model file
from Keras h5 format to tensorflow.js format.
tensorflowjs_converter --input_format keras trained_vgg_best.h5 ./
Then you can put the output file group1-shard1of1.bin in /public/data and use
tensorflow.js to load the trained model.