Text-to-image Retrieval on MSCOCO (5K)
53.98R@1AMoE
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| AMoEHead=Ensemble, Resolution=512x512, Parameters=0.3B-0.6B2025.12 | 53.98 | — | — | |
| TCL#Images=4M, Zero-shot=true2022.02 | 53.5 | 79 | 87.1 | |
| HarmoCLIPBackbone=ViT-B/16, Extra data / VLM augmentation=false2025.11 | 53.44 | — | — | |
| RADIOv2.5-HHead=Ensemble, Resolution=512x512, Parameters=0.6B2025.12 | 53.24 | — | — | |
| VISTA-LTime (ms)=~40, Fine-tuning=true2022.03 | 52.6 | 79.6 | 87.6 | |
| RADIOv2.5-LHead=Ensemble, Resolution=512x512, Parameters=0.3B2025.12 | 52.44 | — | — | |
| RADIOv2.5-HHead=CLIP, Resolution=512x512, Parameters=0.6B2025.12 | 52.24 | — | — | |
| TeachersHead=SigLIP2, Resolution=512x5122025.12 | 52.1 | — | — | |
| AMoEHead=SigLIP2, Resolution=512x512, Parameters=0.3B-0.6B2025.12 | 51.81 | — | — | |
| VSEinftyFine-tuning=true2022.03 | 51.6 | 79.3 | — | |
| RADIOv2.5-LHead=CLIP, Resolution=512x512, Parameters=0.3B2025.12 | 51.6 | — | — | |
| RADIOv2.5-HHead=SigLIP, Resolution=512x512, Parameters=0.6B2025.12 | 50.88 | — | — | |
| SOHOFine-tuning=true2022.03 | 50.6 | 78 | 86.7 | |
| UNITER-BTime (ms)=~900, Fine-tuning=true2022.03 | 50.3 | 78.5 | 87.2 | |
| ALBEF#Images=4M, Zero-shot=true2022.02 | 50.1 | 76.4 | 84.5 | |
| Pixel-BERT-XTime (ms)=~160, Fine-tuning=true2022.03 | 50.1 | 77.6 | 86.2 | |
| RADIOv2.5-LHead=SigLIP, Resolution=512x512, Parameters=0.3B2025.12 | 49.94 | — | — | |
| UNITER#Images=4M, Zero-shot=true2022.02 | 48.8 | 76.7 | 85.8 | |
| Unicoder-VLTime (ms)=925, Fine-tuning=true2022.03 | 48.4 | 76.7 | 85.9 | |
| VISTA-BTime (ms)=~17, Fine-tuning=true2022.03 | 47.8 | 75.8 | 84.5 | |
| TeachersHead=DINOv3, Resolution=512x5122025.12 | 47.66 | — | — | |
| AMoEHead=DINOv3, Resolution=512x512, Parameters=0.3B-0.6B2025.12 | 46.67 | — | — | |
| ALIGN#Images=1.2B, Zero-shot=true2022.02 | 45.6 | 69.8 | 78.6 | |
| FG-CLIPBackbone=ViT-B/16, Extra data / VLM augmentation=true2025.11 | 45.4 | — | — | |
| ViLT-BTime (ms)=~15, Fine-tuning=true2022.03 | 42.7 | 72.9 | 83.1 | |
| SGRAFFine-tuning=true2022.03 | 41.9 | — | 81.3 | |
| Pixel-BERT-RTime (ms)=~60, Fine-tuning=true2022.03 | 41.1 | 69.7 | 80.5 | |
| VSRNFine-tuning=true2022.03 | 40.5 | 70.6 | 81.1 | |
| EVA-CLIPBackbone=ViT-B/16, Extra data / VLM augmentation=false2025.11 | 40.47 | — | — | |
| ViLT-B/32Visual Embed=Linear, Time (ms)=~15, Zero-Shot=true, Pre-training steps=200K2021.02 | 40.4 | 70 | 81.1 | |
| ViLT#Images=4M, Zero-shot=true2022.02 | 40.4 | 70 | 81.1 | |
| FineCLIPBackbone=ViT-B/16, Extra data / VLM augmentation=true2025.11 | 40.2 | — | — | |
| IMRAMFine-tuning=true2022.03 | 39.7 | 69.1 | 79.8 | |
| SCANFine-tuning=true2022.03 | 38.6 | 69.3 | 80.4 | |
| CLIP#Images=400M, Zero-shot=true2022.02 | 37.8 | 62.4 | 72.2 | |
| CLIPBackbone=ViT-B/16, Extra data / VLM augmentation=false2025.11 | 37.6 | — | — | |
| ViLT-B/32Visual Embed=Linear, Time (ms)=~15, Zero-Shot=true2021.02 | 37.3 | 67.4 | 79 | |
| ImageBERTVisual Embed=Region, Time (ms)=~925, Zero-Shot=true, Additional Pre-training Dataset=10M proprietary2021.02 | 32.3 | 59 | 70.2 | |
| ImageBERT#Images=6M, Zero-shot=true2022.02 | 32.3 | 59 | 70.2 | |
| PCMEFine-tuning=true2022.03 | 31.9 | 62.1 | 74.5 | |
| UniConBackbone=CLIP ViT-B/32, Train time (s)=11.152026.04 | 29.2 | — | 64.4 | |
| CLIPSelfBackbone=ViT-B/16, Extra data / VLM augmentation=false2025.11 | 16.1 | — | — | |
| UniConBackbone=RN-50 + SBERT, Train time (s)=11.112026.04 | 12.9 | — | 43.9 | |
| SGD–CLIPBackbone=CLIP ViT-B/32, Train time (s)=1066.602026.04 | 12.3 | — | 42.7 | |
| SGD–CLIPBackbone=RN-50 + SBERT, Train time (s)=5121.722026.04 | 6 | — | 28.6 | |
| DINORANKCLIPBackbone=ViT-B/32, Pre-training Dataset=CC3M, R=32026.05 | 4.35 | 11.8 | 16.2 | |
| RANKCLIPBackbone=ViT-B/32, Pre-training Dataset=CC3M2026.05 | 3.74 | 10.28 | 14.18 | |
| CyCLIPBackbone=ViT-B/32, Pre-training Dataset=CC3M2026.05 | 3.72 | 11.16 | 16.06 | |
| CLIPBackbone=ViT-B/32, Pre-training Dataset=CC3M2026.05 | 3.7 | 9.74 | 14.04 | |
| ALIPBackbone=ViT-B/32, Pre-training Dataset=CC3M2026.05 | 3.7 | 10.22 | 14.38 | |
| RegionCLIPBackbone=ViT-B/16, Extra data / VLM augmentation=false2025.11 | 2 | — | — |