Skip to content

Latest commit

 

History

History
155 lines (95 loc) · 8.76 KB

README.md

File metadata and controls

155 lines (95 loc) · 8.76 KB

Waffling for Performance Python 3.9+

Authors: Karsten Roth, Jae Myung Kim, Almut Sophia Koepke, Oriol Vinyals, Cordelia Schmid, Zeynep Akata


Table of Contents


Overview

This repository contains code to replicate key experiments from our paper Waffling around for Performance: Visual Classification with Random Words and Broad Concepts. It should also provide a good starting point for any subsequent research looking to study improved (zero-shot) transfer performance of pretrained Vision Language Models (VLM), and extends the great repository associated with the original Visual Classification via Description from Large Language Models paper.

If you find this repository useful or use it as part of your research, please consider citing it.


Setting Up

Set up environment: To get started, simply set up the correct environment using environment.yaml by running

conda env create -f environment.yaml

and activate the environment via conda activate waffle.

Ensure clip is up-to-date: The above command should install all relevant libraries. If you are not able to utilize the ViT-L/14 backbone, it is like because your version of clip is not up-to-date. In this case, consider re-installing it:

pip install git+https://github.com/openai/CLIP.git

Downloading datasets