Efficient and Long-Tailed Generalization for Pre-trained Vision-Language Model

This is the source code for the paper: Efficient and Long-Tailed Generalization for Pre-trained Vision-Language Model (KDD 2024).

Install

This code is built on top of the codebase for CoOp. Please follow their instructions to prepare the required environments and datasets.

Run

Please refer to the running scripts under scripts/candle for different experimental settings.

Make sure to change the path of DATA in corresponding scripts and run the commands under the main directory.

In run_example.sh, we provide some examples on how to run the model.