Code release of paper: Training-free Zero-shot Composed Image Retrieval with Local Concept Reranking
We generate caption for the reference image, as "reference_caption", and "target_caption" with ChatGPT4-turbo. Both the "reference_caption" and "target_caption" are saved in directory: here
- Update datasets
- Demo and Codes
If you find our work useful in your research, please consider citing:
@article{sun2023training,
title={Training-free zero-shot composed image retrieval with local concept reranking},
author={Sun, Shitong and Ye, Fanghua and Gong, Shaogang},
journal={arXiv preprint arXiv:2312.08924},
year={2023}
}