Text-to-Image Retrieval on MS-COCO 5K
54.1R@1SoftMask++
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SoftMask++#Img=4M, Zero-shot=true2023.04 | 54.1 | 79.8 | 87.3 | — | |
| CODIS#Img=4M, Zero-shot=true2023.04 | 53.9 | 79.5 | 87.1 | — | |
| TCL#Img=4M, Zero-shot=true2023.04 | 53.5 | 79 | 87.1 | — | |
| ALBEF#Img=4M, Zero-shot=true2023.04 | 50.1 | 76.4 | 84.5 | — | |
| UNITER#Img=4M, Zero-shot=true2023.04 | 48.8 | 76.7 | 85.8 | — | |
| ALIGN#Img=1.2B, Zero-shot=true2023.04 | 45.6 | 69.8 | 78.6 | — | |
| ReConNoise=0%2025.02 | 44.4 | 73.1 | 83.1 | 441.6 | |
| SEPSFine-grained alignment=true, Zero-shot=true, Backbone=CLIP ViT-B/16, Training Dataset=Flickr2025.11 | 44.3 | 69.5 | — | — | |
| ReConNoise=20%2025.02 | 43.5 | 72.4 | 82.7 | 437.6 | |
| CRCLNoise Rate=20%, Evaluation Protocol=Fine-tune2023.10 | 42.9 | 71.9 | 82.1 | — | |
| GSCNoise level=20%, Evaluation protocol=Finetune2024.05 | 42 | 71.4 | 81.8 | 430.8 | |
| ReConNoise=50%2025.02 | 41.5 | 70.7 | 81 | 428.3 | |
| MSCNNoise Level=20%, Re-implemented=false2024.03 | 41 | 69.8 | 80.3 | 423 | |
| REPAIRNoise Level=20%, Re-implemented=false2024.03 | 41 | 70.2 | 80.7 | 425 | |
| NCRNoise Level=20%, Re-implemented=true2024.03 | 40.8 | 69.7 | 80.4 | 421.4 | |
| NCRNoise level=20%, Evaluation protocol=Finetune2024.05 | 40.6 | 69.8 | 80.1 | 422 | |
| ViLT#Img=4M, Zero-shot=true2023.04 | 40.4 | 70 | 81.1 | — | |
| GSCNoise level=50%, Evaluation protocol=Finetune2024.05 | 40 | 69.1 | 79.7 | 416.3 | |
| REPAIRNoise Level=40%, Re-implemented=false2024.03 | 39.2 | 68.6 | 79.2 | 415.4 | |
| NCRNoise Level=40%, Re-implemented=true2024.03 | 39.1 | 68.2 | 78.7 | 413.1 | |
| BiCroNoise Level=20%, Re-implemented=true2024.03 | 38.9 | 68.1 | 79.1 | 410.9 | |
| NCRNoise level=50%, Evaluation protocol=Finetune2024.05 | 37.9 | 66.6 | 77.8 | 404.6 | |
| CLIP#Img=400M, Zero-shot=true2023.04 | 37.8 | 62.4 | 72.2 | — | |
| CLIPNoise Rate=0%, Evaluation Protocol=Zero-Shot, Backbone=ViT-L/142023.10 | 37.8 | 62.4 | 72.2 | — | |
| CLIP-LNoise level=0%, Evaluation protocol=Zeroshot, Backbone=ViT-L/142024.05 | 37.8 | 62.4 | 72.2 | 400.4 | |
| CLIP-14Noise=0%, Backbone=ViT-L/142025.02 | 37.8 | 62.4 | 72.2 | 400.4 | |
| BiCroNoise Level=40%, Re-implemented=true2024.03 | 37.1 | 66.2 | 77.3 | 398.9 | |
| REPAIRNoise Level=60%, Re-implemented=false2024.03 | 36.8 | 65.9 | 77 | 400.3 | |
| NCRNoise Level=60%, Re-implemented=true2024.03 | 36.3 | 65.2 | 76.6 | 394.9 | |
| BiCroNoise Level=60%, Re-implemented=true2024.03 | 34.2 | 62.9 | 74.5 | 381.6 | |
| CLIP#Fine-grained alignment=false, Zero-shot=true, Backbone=CLIP ViT-B/16, Training Dataset=untrained2025.11 | 33.3 | 58.2 | — | — | |
| CLIPNoise Rate=0%, Evaluation Protocol=Zero-Shot, Backbone=ViT-B/322023.10 | 30.4 | 56 | 66.8 | — | |
| CLIP-BNoise level=0%, Evaluation protocol=Zeroshot, Backbone=ViT-B/322024.05 | 30.4 | 56 | 66.8 | 361.6 | |
| CLIP-32Noise=0%, Backbone=ViT-B/322025.02 | 30.4 | 56 | 66.8 | 361.6 | |
| CLIPNoise Rate=20%, Evaluation Protocol=Fine-tune, Backbone=ViT-L/142023.10 | 22.6 | 43.2 | 53.7 | — | |
| CLIP-14Noise=20%, Backbone=ViT-L/142025.02 | 22.6 | 43.2 | 53.7 | 289.4 | |
| LAPSFine-grained alignment=true, Zero-shot=true, Backbone=CLIP ViT-B/16, Training Dataset=Flickr2025.11 | 19 | 50.5 | — | — | |
| CLIPNoise Rate=20%, Evaluation Protocol=Fine-tune, Backbone=ViT-B/322023.10 | 14.8 | 37.6 | 49.6 | — | |
| CLIP-BNoise level=20%, Evaluation protocol=Finetune, Backbone=ViT-B/322024.05 | 14.8 | 37.6 | 49.6 | 236.3 | |
| CLIP-32Noise=20%, Backbone=ViT-B/322025.02 | 14.8 | 37.6 | 49.6 | 236.3 | |
| D2S-VSEFine-grained alignment=false, Zero-shot=true, Backbone=CLIP ViT-B/16, Training Dataset=Flickr2025.11 | 11.3 | 28.3 | — | — | |
| SCANFine-grained alignment=true, Zero-shot=true, Backbone=CLIP ViT-B/16, Training Dataset=Flickr2025.11 | 10.5 | 26.8 | — | — | |
| VSE++Fine-grained alignment=false, Zero-shot=true, Backbone=CLIP ViT-B/16, Training Dataset=Flickr2025.11 | 9.1 | 24.6 | — | — | |
| CLIP-BNoise level=50%, Evaluation protocol=Finetune, Backbone=ViT-B/322024.05 | 7.8 | 19.5 | 26.8 | 131.1 | |
| CLIP-32Noise=50%, Backbone=ViT-B/322025.02 | 7.8 | 19.5 | 26.8 | 131.1 |