cross-modal-retrieval

TEVI is a framework that uses captions to guide the editing of image embeddings via sparse autoencoders, improving vision-language alignment and retrieval performance.