Pretrain like Your Inference: Masked Tuning Improves Zero-Shot Composed Image Retrieval

Benchmark Model Rank Results
zero-shot-composed-image-retrieval-zs-cir-on-circoMTCIR (CLIP L/14)#34mAP@10: 11.63
zero-shot-composed-image-retrieval-zs-cir-on-circoMTCIR (BLIP B/16)#39mAP@10: 8.03
zero-shot-composed-image-retrieval-zs-cir-on-cirrMTCIR (BLIP B/16)#25R@5: 58.87
zero-shot-composed-image-retrieval-zs-cir-on-cirrMTCIR (CLIP L/14)#36R@5: 54.58
zero-shot-composed-image-retrieval-zs-cir-on-fashion-iqMTCIR (CLIP L/14)#9(Recall@10+Recall@50)/2: 46.42