This repository contains a Python implementation of the Contrastive Language Image Pre-Training (CLIP) model built on a simplified version of the Radiology Objects in COntext (ROCO) dataset. The dataset is not shared in this folder.
We test the zero-shot and few-shot learning of the model on the Chest X-Ray dataset.
The file report.pdf contains information about the results obtained with this model, its limitations and further developments.
This implementation of the algorithm is part of the Ph.D. course "058623 - ADVANCED TOPICS IN DEEP LEARNING: THE RISE OF TRANSFORMERS" by Politecnico di Milano university.