We used sweep method of wandb to find best hyperparameters. We trained network on persianMNIST dataset. We used framework pytorch==1.8.2.
persianMNIST image dataset includes a training set of 960 samples and a test set of 240 samples. Each sample is resized to 64*64 pixels on a RGB associated with a tag of 10 classes. Please download dataset from here
To train the model please run the file:
Train.py
Show results on wandb here
To test the model please run the file:
Test.py
Please download the weights from here
to test the trained model, please run the following file:
python Inference.py --img_path data/2.jpeg --model_path model_mnist.pth --device GPU