Follow-up Differential Descriptions

Code for Follow-Up Differential Descriptions: Language Models Resolve Ambiguities for Image Classification

Reproducing the results

Install the requirements in the requirements.txt file. To install the clip module, follow the instructions in the original repository.

Follow the instructions in Dataset_preparation.md to prepare the datasets.

Set the DATA_ROOT and PROMPT_ROOT variables in the ./create_results.sh and run the script to reproduce the results reported in Table 1.

To reproduce the results in Table 2, please use the code in ./analysis_exp directory of this repo.