This repository contains the code for the study T2Distill-GradCAM: Learning Vision Transformers using Two-Teacher Knowledge Distillation with Autoencoders and Gradient-weighted Class Activation Maps in Diabetic Retinopathy Classification, accepted for publication and presentation at the 25th International Conference on Computing and Artificial Intelligence, authored by myself, Patricia Denise Poblete, and Ann Clarisse Salazar.

The implementation of the Vision Transformer architecture in this study is based on that of faustomorales. The preprocessing techniques are also based in part on sveitser's implementation.

Note that this repository does not contain the EyePACS fundus image dataset, which can be accessed through its Kaggle page. The labels are already provided in this repository, although it must be noted that the test labels were obtained from an unofficial source.

1. Install required packages with pip install -r requirements.txt.
2. Obtain images and place in the data folder under train and test. The .csv files containing the labels are preprovided.