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}
}