Text-to-Image Retrieval on Flickr
98.9R@1U-MARVEL+
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| U-MARVEL+Reranker=true2025.07 | 98.9 | — | — | |
| U-MARVELReranker=false2025.07 | 97.2 | — | — | |
| LamRAReranker=true2025.07 | 96.5 | — | — | |
| UniMEReranker=false2025.07 | 93.9 | — | — | |
| MiCoZero-shot=false2024.06 | 91.1 | — | — | |
| SOTAZero-shot=false2024.06 | 90.3 | — | — | |
| MiCoZero-shot=true2024.06 | 90.1 | — | — | |
| SOTAZero-shot=true2024.06 | 89.7 | — | — | |
| LamRAreranking=pointwise/listwise2024.12 | 88.1 | — | — | |
| LamRA-RetReranker=false2025.07 | 88.1 | — | — | |
| VLM2VecReranker=false2025.07 | 85.8 | — | — | |
| SigLIP 2Setting=Zero-shot2026.02 | 85.7 | — | — | |
| UniIR-CLIP_SFReranker=false2025.07 | 84.1 | — | — | |
| E5-VReranker=false2025.07 | 84 | — | — | |
| EVA-CLIPparameters=18B2024.12 | 83.3 | — | — | |
| LamRA-Ret2024.12 | 82.8 | — | — | |
| VersaViTVision backbone state=Unfrozen, Projection layer fine-tuning=true, Training data alignment=30M image-text pairs2026.02 | 82.2 | — | — | |
| CLIP-LReranker=false2025.07 | 81.8 | — | — | |
| EVA-CLIPparameters=8B2024.12 | 80.8 | — | — | |
| EVA-CLIP-8BSetting=Zero-shot2026.02 | 80.8 | — | — | |
| SigLIPSetting=Zero-shot2026.02 | 80.5 | — | — | |
| E5-V2024.12 | 79.5 | — | — | |
| UniIR-CLIP2024.12 | 78.7 | — | — | |
| VersaViTVision backbone state=Frozen, Projection layer fine-tuning=true, Training data alignment=30M image-text pairs2026.02 | 77 | — | — | |
| Long-CLIP-L2024.12 | 76.1 | — | — | |
| ViT-B/16Architecture=ViT-B/16, Zero-shot evaluation=true2026.03 | 75.5 | — | — | |
| MagicLens-L2024.12 | 72.5 | — | — | |
| CLIP-L2024.12 | 67.3 | — | — | |
| CLIP-LSetting=Zero-shot2026.02 | 65.2 | — | — | |
| MagicLens-LReranker=false2025.07 | 60.9 | — | — | |
| CLIP-RDArchitecture=ViT-T/16, Zero-shot evaluation=true2026.03 | 58.6 | — | — | |
| TinyCLIPArchitecture=TinyCLIP, Zero-shot evaluation=true2026.03 | 58.5 | — | — | |
| CLIP-KD*Architecture=ViT-T/16, Zero-shot evaluation=true2026.03 | 56.4 | — | — | |
| ViT-T/16Architecture=ViT-T/16, Zero-shot evaluation=true2026.03 | 42 | — | — | |
| ItemizedCLIPEvaluation Protocol=Zero-shot, Training Data=Itemized-cc0.3M2025.12 | 14.4 | 38.8 | — | |
| FLAIREvaluation Protocol=Zero-shot, Training Data=Itemized-cc0.3M2025.12 | 11.9 | 36.1 | — | |
| DreamLIPEvaluation Protocol=Zero-shot, Training Data=Itemized-cc0.3M2025.12 | 9.6 | 30 | — | |
| Multi-positive SigLIPEvaluation Protocol=Zero-shot, Training Data=Itemized-cc0.3M2025.12 | 9.2 | 30.3 | — | |
| Vanilla SigLIPEvaluation Protocol=Zero-shot, Training Data=Itemized-cc0.3M2025.12 | 4.7 | 16.5 | — | |
| Vanilla CLIPEvaluation Protocol=Zero-shot, Training Data=Itemized-cc0.3M2025.12 | 3.1 | 11 | — | |
| CLIPArchitecture=L/142026.04 | — | — | 65.2 | |
| CLIP-KDBackbone=ResNet-50, Zero-shot=true2026.03 | — | — | 69.4 | |
| CLIP-KDBackbone=ResNet-18, Zero-shot=true2026.03 | — | — | 52.4 | |
| CLIP-RDBackbone=ResNet-50, Zero-shot=true2026.03 | — | — | 70.4 | |
| CLIP-RDBackbone=ResNet-18, Zero-shot=true2026.03 | — | — | 52.7 | |
| CLIP-RDBackbone=EfficientNet-B0, Zero-shot=true2026.03 | — | — | 62.7 | |
| DINOv2 (dino.txt)Architecture=L/142026.04 | — | — | 77.1 | |
| EfficientNet-B0Backbone=EfficientNet-B0, Zero-shot=true, Role=Student2026.03 | — | — | 47.2 | |
| OpenCLIPArchitecture=G/142026.04 | — | — | 79.5 | |
| PE-coreArchitecture=G/142026.04 | — | — | 85.7 | |
| ResNet-18Backbone=ResNet-18, Zero-shot=true, Role=Student2026.03 | — | — | 41.5 | |
| ResNet-50Backbone=ResNet-50, Zero-shot=true, Role=Student2026.03 | — | — | 52.7 | |
| SigLIP2Architecture=g/162026.04 | — | — | 86 | |
| TIPSArchitecture=g/142026.04 | — | — | 84.5 | |
| TIPSv2Architecture=g/142026.04 | — | — | 85.9 | |
| ViT-B/16Backbone=ViT-B/16, Zero-shot=true, Role=Teacher2026.03 | — | — | 75.5 |