This is the code for "AI-Generated Images Introduce Invisible Relevance Bias to Text-Image Retrieval" accepted by SIGIR 2024 (Invisible Relevance Bias: Text-Image Retrieval Models Prefer AI-Generated Images)
@misc{xu2024aigenerated,
title={AI-Generated Images Introduce Invisible Relevance Bias to Text-Image Retrieval},
author={Shicheng Xu and Danyang Hou and Liang Pang and Jingcheng Deng and Jun Xu and Huawei Shen and Xueqi Cheng},
year={2024},
eprint={2311.14084},
archivePrefix={arXiv},
primaryClass={cs.IR}
}
- Use ChatGPT to merge the captions:
python image_generation/call_five_captions.py
- Image generation for multiple times:
python image_generation/image_generation_by_merged_caption.py
- Selecte the most similar image:
python image_generation/select_most_similar_image.py
FLAVA, ALIGIN and BEIT-3
python test/comparison_{flava,aligin,beit}_5.py