This is initial work Satyendra Pandit. This project is created as side project for Facebook's PyTorch Challenge Scholarship 2018/19.
How to run this note book.
-
Download the restructured HAM10000 skin cancer dataset from below Google Drive link. https://drive.google.com/open?id=1hO42zUHH6AJeWfvFacbCieiChZ72VGUv & Upload the Cancer folder into drive which comprises of train_dir, val_dir, test_dir, can_to_name.json & cancer_classifier_checkpointV1.pth.
-
Mount the drive correctly then you can able to access dataset.
-
Mention the path of data_dir as data_dir = 'drive/My Drive/Cancer' in "Data Augmentation Step",
-
After successful training, save the checkpoint. Here, I have saved the checkpoint as "cancer_classifier_checkpointV1.pth".
-
Load the checkpoint and rebuild the model again.
-
Now, in Sanity Check, you can change the path of image to see the prediction. #Give image path image_path = 'drive/My Drive/Cancer/test_dir/6/ISIC_0024309.jpg' #Change path of image to get different image.
Here is a small prototype of final skin cancer classifier app. !(https://share.proto.io/LPE8ES/)