Text-to-Image Retrieval on Flickr30K
94.9R@1EVA-CLIP-E
Evaluation results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| EVA-CLIP-E#Params=4.35B, Zero-shot=true, Strategy=Original2026.03 | 94.9 | 99.3 | 99.7 | — | — | — | — | — | 85.2 | — | — | — | — | — | |
| EVA-CLIP-E#Params=2.50B, Zero-shot=true, Strategy=Distillation2026.03 | 94.4 | 99.3 | 99.8 | — | — | — | — | — | 85.3 | — | — | — | — | — | |
| EVA-CLIP-8B#Params=7.53B, Zero-shot=true, Strategy=Original2026.03 | 94.4 | 99.4 | 99.7 | — | — | — | — | — | 85.7 | — | — | — | — | — | |
| EVA-CLIP-8B#Params=4.59B, Zero-shot=true, Strategy=Distillation2026.03 | 94.3 | 99.6 | 99.8 | — | — | — | — | — | 86.1 | — | — | — | — | — | |
| OpenCLIP-G#Params=1.84B, Zero-shot=true, Strategy=Original2026.03 | 92.6 | 99.4 | 99.8 | — | — | — | — | — | 85 | — | — | — | — | — | |
| M2-EncoderModel scale=10B, Zero-shot evaluation=true2024.01 | 92.2 | 99.5 | 99.7 | — | — | — | — | — | — | — | — | — | — | — | |
| OpenCLIP-G#Params=1.17B, Zero-shot=true, Strategy=Distillation2026.03 | 92 | 99.6 | 99.8 | — | — | — | — | — | 85 | — | — | — | — | — | |
| OpenCLIP-g#Params=0.62B, Zero-shot=true, Strategy=Distillation2026.03 | 91.6 | 99.1 | 99.6 | — | — | — | — | — | 84 | — | — | — | — | — | |
| OpenCLIP-g#Params=1.01B, Zero-shot=true, Strategy=Original2026.03 | 91.5 | 98.9 | 99.5 | — | — | — | — | — | 83.8 | — | — | — | — | — | |
| COSAExample=415M, parameters=1.2B, Zero-shot=true2023.06 | 90.2 | 98.4 | 99.4 | — | — | — | — | — | — | — | — | — | — | — | |
| BLIP-2Example=129M, Zero-shot=true2023.06 | 89.7 | 98.1 | 98.9 | — | — | — | — | — | — | — | — | — | — | — | |
| Arbitrary Ratio Feature Compression (ARFC)Backbone=SigLip 2 (ViT-B/16), CR=50%, Setting=Finetuning2026.02 | 89.7 | 98.8 | 99.8 | — | — | — | — | — | — | — | — | — | — | — | |
| BaselineBackbone=SigLip 2 (ViT-B/16), CR=0, Setting=Finetuning2026.02 | 89.6 | 98.8 | 99.7 | — | — | — | — | — | — | — | — | — | — | — | |
| Arbitrary Ratio Feature Compression (ARFC)Backbone=SigLip 2 (ViT-B/16), CR=75%, Setting=Finetuning2026.02 | 89.5 | 98.8 | 99.6 | — | — | — | — | — | — | — | — | — | — | — | |
| AutoEncoder-VAEBackbone=SigLip 2 (ViT-B/16), CR=50%, Setting=Finetuning2026.02 | 88.9 | 98.5 | 99.5 | — | — | — | — | — | — | — | — | — | — | — | |
| M2-EncoderModel scale=1B, Zero-shot evaluation=true2024.01 | 88.8 | 98.9 | 99.5 | — | — | — | — | — | — | — | — | — | — | — | |
| Arbitrary Ratio Feature Compression (ARFC)Backbone=SigLip 2 (ViT-B/16), CR=87.5%, Setting=Finetuning2026.02 | 88.8 | 98.6 | 99.6 | — | — | — | — | — | — | — | — | — | — | — | |
| Q-FormerBackbone=SigLip 2 (ViT-B/16), CR=50%, Setting=Finetuning2026.02 | 88.5 | 98.3 | 99.5 | — | — | — | — | — | — | — | — | — | — | — | |
| Florencetuning_mode=fine-tuning (FT100%), resolution=3842022.06 | 87.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AutoEncoder-VAEBackbone=SigLip 2 (ViT-B/16), CR=75%, Setting=Finetuning2026.02 | 87.8 | 98.1 | 99.3 | — | — | — | — | — | — | — | — | — | — | — | |
| OpenCLIP-G#Params=1.17B, Zero-shot=true, Strategy=Pruning2026.03 | 87.8 | 97.9 | 99.2 | — | — | — | — | — | 80.8 | — | — | — | — | — | |
| BLIPTraining Protocol=full fine-tuning, # Tunable=223M2023.05 | 87.3 | 97.6 | 98.9 | — | — | — | — | — | — | — | — | — | — | — | |
| BLIP# Tunable=223M, Fine-tuning strategy=full fine-tuning2025.12 | 87.3 | 97.6 | 98.9 | — | — | — | — | — | — | — | — | — | — | — | |
| M2-EncoderModel scale=0.4B, Zero-shot evaluation=true2024.01 | 87.2 | 97.9 | 99.4 | — | — | — | — | — | — | — | — | — | — | — | |
| Q-FormerBackbone=SigLip 2 (ViT-B/16), CR=75%, Setting=Finetuning2026.02 | 87.1 | 97.8 | 99.3 | — | — | — | — | — | — | — | — | — | — | — | |
| AuroraTraining Protocol=frozen backbone, # Tunable=0.2M, Rank (r)=1282023.05 | 86.8 | 97.6 | 98.9 | — | — | — | — | — | — | — | — | — | — | — | |
| BLIPExample=129M, Zero-shot=true2023.06 | 86.7 | 97.3 | 98.7 | — | — | — | — | — | — | — | — | — | — | — | |
| AuroraTraining Protocol=frozen backbone, # Tunable=0.1M, Rank (r)=642023.05 | 86.7 | 97.8 | 98.7 | — | — | — | — | — | — | — | — | — | — | — | |
| UniAdapterTraining Protocol=frozen backbone, # Tunable=4.6M, Rank (r)=1282023.05 | 86.5 | 97.4 | 98.8 | — | — | — | — | — | — | — | — | — | — | — | |
| mPLUGEvaluation Protocol=Zero-shot, Finetuned on COCO=true2022.05 | 86.4 | 97.6 | — | — | — | — | — | — | — | — | — | — | — | — | |
| UniAdapterTraining Protocol=frozen backbone, # Tunable=18.8M, Rank (r)=5122023.05 | 86.4 | 97.4 | 98.9 | — | — | — | — | — | — | — | — | — | — | — | |
| Null-LoRA# Tunable=6.0M, Fine-tuning strategy=frozen backbone2025.12 | 86.3 | 97.3 | 98.8 | — | — | — | — | — | — | — | — | — | — | — | |
| UniAdapter (r=512)# Tunable=19.5M, Fine-tuning strategy=frozen backbone, rank=5122025.12 | 86.2 | 97.3 | 98.8 | — | — | — | — | — | — | — | — | — | — | — | |
| LoRATraining Protocol=frozen backbone, # Tunable=10.6M, Rank (r)=322023.05 | 85.8 | 97.1 | 98.4 | — | — | — | — | — | — | — | — | — | — | — | |
| Aurora (r=64)# Tunable=0.6M, Fine-tuning strategy=frozen backbone, rank=642025.12 | 85.8 | 97.2 | 98.7 | — | — | — | — | — | — | — | — | — | — | — | |
| ALBEFTraining Protocol=full fine-tuning, # Tunable=210M2023.05 | 85.6 | 97.5 | 98.9 | — | — | — | — | — | — | — | — | — | — | — | |
| ALBEF# Tunable=210M, Fine-tuning strategy=full fine-tuning2025.12 | 85.6 | 97.5 | 98.9 | — | — | — | — | — | — | — | — | — | — | — | |
| DoRA (r=32)# Tunable=10.8M, Fine-tuning strategy=frozen backbone, rank=322025.12 | 85.1 | 97.2 | 98.5 | — | — | — | — | — | — | — | — | — | — | — | |
| FG-CLIP2Backbone=ViT-So/162026.02 | 85 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FG-CLIP 2Backbone=ViT-So/16, Evaluation Protocol=Zero-shot2025.10 | 85 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ALIGNtuning_mode=fine-tuning (FT100%), resolution=2892022.06 | 84.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BLIPEvaluation Protocol=Zero-shot, Finetuned on COCO=true2022.05 | 84.9 | 96.7 | — | — | — | — | — | — | — | — | — | — | — | — | |
| ALIGNTraining Protocol=full fine-tuning, # Tunable=820M2023.05 | 84.9 | 97.4 | 98.6 | — | — | — | — | — | — | — | — | — | — | — | |
| ALIGN# Tunable=820M, Fine-tuning strategy=full fine-tuning2025.12 | 84.9 | 97.4 | 98.6 | — | — | — | — | — | — | — | — | — | — | — | |
| Arbitrary Ratio Feature Compression (ARFC)Backbone=SigLip 2 (ViT-B/16), CR=50%, Setting=Zero-Shot2026.02 | 84.8 | 96 | 98.1 | — | — | — | — | — | — | — | — | — | — | — | |
| LoRA (r=32)# Tunable=10.6M, Fine-tuning strategy=frozen backbone, rank=322025.12 | 84.8 | 96.6 | 98.4 | — | — | — | — | — | — | — | — | — | — | — | |
| FG-CLIP 2Backbone=ViT-L/16, Evaluation Protocol=Zero-shot2025.10 | 84.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BaselineBackbone=SigLip 2 (ViT-B/16), CR=0, Setting=Zero-Shot2026.02 | 84.5 | 95.8 | 98 | — | — | — | — | — | — | — | — | — | — | — | |
| Arbitrary Ratio Feature Compression (ARFC)Backbone=SigLip 2 (ViT-B/16), CR=75%, Setting=Zero-Shot2026.02 | 84.3 | 95.8 | 97.9 | — | — | — | — | — | — | — | — | — | — | — | |
| Uni-Perceiver-L + Conditional MoEstuning_mode=fine-tuning (FT100%)2022.06 | 83.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| OpenCLIP-g#Params=0.62B, Zero-shot=true, Strategy=Pruning2026.03 | 83.4 | 95.7 | 97.7 | — | — | — | — | — | 77.9 | — | — | — | — | — | |
| MADTPBackbone=CLIP, GFLOPs=178.82024.03 | 83.3 | 97 | 98.5 | — | — | — | — | — | — | — | — | — | — | — | |
| AutoEncoder-VAEBackbone=SigLip 2 (ViT-B/16), CR=50%, Setting=Zero-Shot2026.02 | 83.3 | 95.1 | 97.6 | — | — | — | — | — | — | — | — | — | — | — | |
| UMG-CLIPModel scale=0.4B, Zero-shot evaluation=true2024.01 | 83.1 | 96 | 98.1 | — | — | — | — | — | — | — | — | — | — | — | |
| UMG-CLIPBackbone=ViT-L/14, Evaluation Protocol=Zero-shot2025.10 | 83.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AutoEncoder-VAEBackbone=SigLip 2 (ViT-B/16), CR=87.5%, Setting=Finetuning2026.02 | 82.9 | 96.6 | 98.4 | — | — | — | — | — | — | — | — | — | — | — | |
| ALBEFEvaluation Protocol=Zero-shot, Finetuned on COCO=true2022.05 | 82.8 | 96.3 | — | — | — | — | — | — | — | — | — | — | — | — | |
| SigLIP 2Backbone=ViT-L/16, Evaluation Protocol=Zero-shot2025.10 | 82.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SigLIP2Backbone=ViT-So/162026.02 | 82.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SigLIP 2Backbone=ViT-So/16, Evaluation Protocol=Zero-shot2025.10 | 82.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MADTPGFLOPs=74.52024.03 | 82.3 | 96.2 | 98 | — | — | — | — | — | — | — | — | — | — | — | |
| mPLUGEvaluation Protocol=Zero-shot, Finetuned on COCO=false2022.05 | 82.2 | 95.8 | — | — | — | — | — | — | — | — | — | — | — | — | |
| Uni-Perceiver-Ltuning_mode=fine-tuning (FT100%)2022.06 | 82.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Q-FormerBackbone=SigLip 2 (ViT-B/16), CR=50%, Setting=Zero-Shot2026.02 | 82.1 | 94.6 | 97.3 | — | — | — | — | — | — | — | — | — | — | — | |
| UPopGFLOPs=91.02024.03 | 82 | 95.8 | 97.6 | — | — | — | — | — | — | — | — | — | — | — | |
| Qwen3-VL-EmbeddingParameters=2B2026.02 | 81.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FG-CLIP 2Backbone=ViT-B/16, Evaluation Protocol=Zero-shot2025.10 | 81.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CARTParadigm=Coarse-Fine ID2024.06 | 81.78 | 96.14 | 98.38 | — | — | — | — | — | — | — | — | — | — | — | |
| AutoEncoder-VAEBackbone=SigLip 2 (ViT-B/16), CR=75%, Setting=Zero-Shot2026.02 | 81.7 | 94.4 | 97.3 | — | — | — | — | — | — | — | — | — | — | — | |
| BEiT-3Example=21M, Zero-shot=true2023.06 | 81.5 | 95.6 | 97.8 | — | — | — | — | — | — | — | — | — | — | — | |
| BEIT3Model scale=0.7B, Zero-shot evaluation=true2024.01 | 81.5 | 95.6 | 97.8 | — | — | — | — | — | — | — | — | — | — | — | |
| FG-CLIPBackbone=ViT-L/14, Evaluation Protocol=Zero-shot2025.10 | 81.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| UMG-CLIPModel scale=5B, Zero-shot evaluation=true2024.01 | 81.2 | 95.7 | 97.7 | — | — | — | — | — | — | — | — | — | — | — | |
| Q-FormerBackbone=SigLip 2 (ViT-B/16), CR=75%, Setting=Zero-Shot2026.02 | 81.2 | 94.3 | 97.1 | — | — | — | — | — | — | — | — | — | — | — | |
| AMoEHead=Ensemble, Resolution=512x512, Parameters=0.3B-0.6B2025.12 | 81.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GMEParameters=7B2026.02 | 81.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RADIOv2.5-HHead=Ensemble, Resolution=512x512, Parameters=0.6B2025.12 | 80.96 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BaselineBackbone=XVLM2024.10 | 80.77 | 96.67 | 98.55 | — | 0.2355 | — | — | — | — | — | — | — | — | — | |
| BaselineModel=XVLM2025.11 | 80.77 | 96.67 | 98.55 | — | 0.2355 | — | — | — | — | — | — | — | — | — | |
| Uni-Perceiver-L + Conditional MoEstuning_mode=prompt tuning (PT1%)2022.06 | 80.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ToMeGFLOPs=69.82024.03 | 80.5 | 95.6 | 97.9 | — | — | — | — | — | — | — | — | — | — | — | |
| UPopBackbone=CLIP, GFLOPs=201.12024.03 | 80.5 | 95.4 | 97.6 | — | — | — | — | — | — | — | — | — | — | — | |
| CoCa-Ltuning_mode=w/o tuning (WT), resolution=5762022.06 | 80.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CoCaModel scale=2.1B, Zero-shot evaluation=true2024.01 | 80.4 | 95.7 | 97.7 | — | — | — | — | — | — | — | — | — | — | — | |
| ObjEmbedParameters=4B2026.02 | 80.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ELIPGFLOPs=93.42024.03 | 80.3 | 96 | 98 | — | — | — | — | — | — | — | — | — | — | — | |
| Arbitrary Ratio Feature Compression (ARFC)Backbone=SigLip 2 (ViT-B/16), CR=87.5%, Setting=Zero-Shot2026.02 | 80.3 | 94.1 | 97 | — | — | — | — | — | — | — | — | — | — | — | |
| VLM2Vec-V2Parameters=2B2026.02 | 80.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Uni-Perceiver-Ltuning_mode=prompt tuning (PT1%)2022.06 | 80 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ObjEmbedParameters=2B2026.02 | 80 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Q-FormerBackbone=SigLip 2 (ViT-B/16), CR=87.5%, Setting=Finetuning2026.02 | 79.9 | 93.8 | 97.6 | — | — | — | — | — | — | — | — | — | — | — | |
| RADIOv2.5-LHead=Ensemble, Resolution=512x512, Parameters=0.3B2025.12 | 79.82 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Uni-Perceiver-B + Conditional MoEstuning_mode=fine-tuning (FT100%)2022.06 | 79.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CrossGETBackbone=CLIP2024.03 | 79.6 | 97.5 | 98 | — | — | — | — | — | — | — | — | — | — | — | |
| Flamingo-3Btuning_mode=w/o tuning (WT), resolution=3202022.06 | 79.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| OpenCLIPModel scale=2.5B, Zero-shot evaluation=true2024.01 | 79.5 | 95 | 97.1 | — | — | — | — | — | — | — | — | — | — | — | |
| RADIOv2.5-HHead=CLIP, Resolution=512x512, Parameters=0.6B2025.12 | 79.26 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SPD (Ours)Model=XVLM2025.11 | 79.13 | 96.11 | 98.25 | — | 0.1859 | — | — | — | — | — | — | 21.1 | — | — | |
| UME-R1Parameters=2B2026.02 | 79.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RADIOv2.5-HHead=SigLIP, Resolution=512x512, Parameters=0.6B2025.12 | 79.08 | — | — | — | — | — | — | — | — | — | — | — | — | — |