iSEARLE: Improving Textual Inversion for Zero-Shot Composed Image Retrieval

Benchmark Model Rank Results
zero-shot-composed-image-retrieval-zs-cir-on-circoiSEARLE-XL (CLIP L/14)#29mAP@10: 13.61
zero-shot-composed-image-retrieval-zs-cir-on-circoiSEARLE-XL-OTI (CLIP L/14)#33mAP@10: 12.67
zero-shot-composed-image-retrieval-zs-cir-on-circoiSEARLE (CLIP B/32)#35mAP@10: 11.24
zero-shot-composed-image-retrieval-zs-cir-on-circoiSEARLE-OTI (CLIP B/32)#36mAP@10: 10.94
zero-shot-composed-image-retrieval-zs-cir-on-cirriSEARLE (CLIP B/32)#31R@5: 55.69
zero-shot-composed-image-retrieval-zs-cir-on-cirriSEARLE-OTI (CLIP B/32)#33R@5: 55.18
zero-shot-composed-image-retrieval-zs-cir-on-cirriSEARLE-XL-OTI (CLIP L/14)#38R@5: 54.05
zero-shot-composed-image-retrieval-zs-cir-on-cirriSEARLE-XL (CLIP L/14)#39R@5: 54.00
zero-shot-composed-image-retrieval-zs-cir-on-cocoiSEARLE-XL-OTI (CLIP L/14)#1Actions Recall@5: 32.55
zero-shot-composed-image-retrieval-zs-cir-on-cocoiSEARLE-XL (CLIP L/14)#3Actions Recall@5: 30.05
zero-shot-composed-image-retrieval-zs-cir-on-cocoiSEARLE-OTI (CLIP B/32)#6Actions Recall@5: 26.63
zero-shot-composed-image-retrieval-zs-cir-on-cocoiSEARLE (CLIP B/32)#7Actions Recall@5: 26.40
zero-shot-composed-image-retrieval-zs-cir-on-fashion-iqiSEARLE-XL-OTI (CLIP L/14)#25(Recall@10+Recall@50)/2: 39.39
zero-shot-composed-image-retrieval-zs-cir-on-fashion-iqiSEARLE-XL (CLIP L/14)#29(Recall@10+Recall@50)/2: 38.24
zero-shot-composed-image-retrieval-zs-cir-on-fashion-iqiSEARLE-OTI (CLIP B/32)#34(Recall@10+Recall@50)/2: 34.93
zero-shot-composed-image-retrieval-zs-cir-on-fashion-iqiSEARLE (CLIP B/32)#35(Recall@10+Recall@50)/2: 34.60
zero-shot-composed-image-retrieval-zs-cir-on-imagenetiSEARLE-XL (CLIP L/14)#1Average Recall: 24.46
zero-shot-composed-image-retrieval-zs-cir-on-imagenetiSEARLE-XL-OTI (CLIP L/14)#2Average Recall: 22.59
zero-shot-composed-image-retrieval-zs-cir-on-imagenetiSEARLE (CLIP B/32)#8Average Recall: 16.01
zero-shot-composed-image-retrieval-zs-cir-on-imagenetiSEARLE-OTI (CLIP B/32)#9Average Recall: 15.62
zero-shot-composed-image-retrieval-zs-cir-on-imagenet-riSEARLE-XL (CLIP L/14)#5(Recall@10+Recall@50)/2: 24.46
zero-shot-composed-image-retrieval-zs-cir-on-imagenet-riSEARLE (CLIP B/32)#11(Recall@10+Recall@50)/2: 16.01
zero-shot-composed-image-retrieval-zs-cir-on-imagenet-riSEARLE-OTI (CLIP B/32)#12(Recall@10+Recall@50)/2: 15.62