The code contained in this repository stands as a didactic example presented at the Artificial Intelligence and Deep Learning course organized by Sezione Fisica Medica of Università degli Studi di Roma Tor Vergata.
This repo contains the following files:
vision_transformer.py, which is a modified version of vision_transformer.py that makes it possible to recover attention matrices.attention_rollout.py, which provides the functions to compute and display attention rollout.example.ipynb, which shows how to compute and display attention rollout.