Vision-by-Language for Training-Free Compositional Image Retrieval

Open paper
Benchmark Model Rank Results
zero-shot-composed-image-retrieval-zs-cir-onCIReVL (CLIP G/14)#13mAP@10: 27.59
zero-shot-composed-image-retrieval-zs-cir-onCIReVL (CLIP L/14)#23mAP@10: 19.01
zero-shot-composed-image-retrieval-zs-cir-onCIReVL (CLIP B/32)#27mAP@10: 15.42
zero-shot-composed-image-retrieval-zs-cir-on-1CIReVL (CLIP G/14)#8R@1: 34.65R@5: 64.29
zero-shot-composed-image-retrieval-zs-cir-on-1CIReVL (CLIP L/14)#18R@1: 24.55R@5: 52.31
zero-shot-composed-image-retrieval-zs-cir-on-1CIReVL (CLIP B/32)#19R@1: 23.94R@5: 52.51
zero-shot-composed-image-retrieval-zs-cir-on-11CIReVL (CLIP B/32)#9A-R@1: 15.9
zero-shot-composed-image-retrieval-zs-cir-on-11CIReVL (CLIP L/14)#10A-R@1: 15.9
zero-shot-composed-image-retrieval-zs-cir-on-11CIReVL (CLIP G/14)#11A-R@1: 17.4
zero-shot-composed-image-retrieval-zs-cir-on-2CIReVL (CLIP G/14)#19(Recall@10+Recall@50)/2: 42.28
zero-shot-composed-image-retrieval-zs-cir-on-2CIReVL (CLIP B/32)#26(Recall@10+Recall@50)/2: 38.82
zero-shot-composed-image-retrieval-zs-cir-on-2CIReVL (CLIP L/14)#27(Recall@10+Recall@50)/2: 38.56

Rank counts only results with a code link.