zhgsxf / CLIP-knows-IA

CLIP-knows-IA

In this repository we provide the source code neccessary to recreate our experiments. We mainly use PyTorch for our experiments.

Code

This repository contains code for in total three different experiment steps.

Fine-Tuning/:

This directory contains the scripts and files for fine-tuning the CLIP model for IAA on the AVA Dataset.

Linear-Regression/:

This directory contains the scripts and files for performing linear probing experiments.

Probing/:

This directory contains the scripts and files for performing zero-shot probing for IAA on the AVA Dataset.

Additional Code

MLSP/:

In this folder we provide the source code for the MLSP model, which has been adapted from the original code to work with Python3 and TensorFlow2. This code was used to calculate the performance of the MLSP model on our split of the AVA dataset.