Class Activation Values: Lucid and Faithful Visual Interpretations for CLIP-based Text-Image Retrievals
Pytorch implementation of Class Activation Values: Lucid and Faithful Visual Interpretations for CLIP-based Text-Image Retrievals, which is accepted by SIGIR 2025.
python 3.X
jupyter notebook
pytorch >= 1.5 (including torchvision)
matplotlib
opencv-python
run demo.ipynb
@inproceedings{chen2025cav,
title={Class Activation Values: Lucid and Faithful Visual Interpretations for CLIP-based Text-Image Retrievals},
author={Pengxu Chen and Huazhong Liu and Jihong Ding and Xinghao Huang and Shaojun Zou and Laurence T. Yang},
booktitle={Proceeding of the ACM SIGIR International Conference on Research and Development in Information Retrieval},
year={2025},
pages={844-853},
doi={10.1145/3726302.3729923}
}Any questions can be asked at [email protected]