Language-only Efficient Training of Zero-shot Composed Image Retrieval

Open paper
Benchmark Model Rank Results
zero-shot-composed-image-retrieval-zs-cir-onLinCIR (CLIP G/14)#19mAP@10: 21.01
zero-shot-composed-image-retrieval-zs-cir-onLinCIR (CLIP L/14)#30mAP@10: 13.58
zero-shot-composed-image-retrieval-zs-cir-on-1LinCIR (CLIP G/14)#23R@5: 64.72
zero-shot-composed-image-retrieval-zs-cir-on-1LinCIR (CLIP L/14)#41R@5: 53.25
zero-shot-composed-image-retrieval-zs-cir-on-11LinCIR (CLIP G/14)#6A-R@1: 13.7
zero-shot-composed-image-retrieval-zs-cir-on-11LinCIR (CLIP L/14)#8A-R@1: 12.2
zero-shot-composed-image-retrieval-zs-cir-on-2LinCIR (CLIP G/14)#2(Recall@10+Recall@50)/2: 55.40
zero-shot-composed-image-retrieval-zs-cir-on-2LinCIR (CLIP L/14)#32(Recall@10+Recall@50)/2: 36.39
zero-shot-composed-image-retrieval-zs-cir-on-5LinCIR (CLIP L/14)#3Average Recall: 21.64
zero-shot-composed-image-retrieval-zs-cir-on-6LinCIR (CLIP L/14)#6(Recall@10+Recall@50)/2: 21.64

Rank counts only results with a code link.