Text Retrieval on COCO
73.6R@1EVA-CLIP-18B
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| EVA-CLIP-18BZero-shot=true2024.02 | 73.6 | 90.9 | 95 | 87.8 | — | |
| VladVABackbone=LLAVA-1.5-7B, Zero-shot=true2024.12 | 72.9 | — | 94.4 | — | — | |
| EVA-CLIPParameters=18B, Zero-shot=true2024.12 | 72.8 | — | 94.2 | — | — | |
| DFN5B-CLIP-H/14Zero-shot=true2024.02 | 72.3 | 90.2 | 94.6 | 86.6 | — | |
| DFN5B-CLIP-H/14+Zero-shot=true2024.02 | 71.8 | 90.4 | 94.9 | 87.2 | — | |
| EVA-CLIP-8BZero-shot=true2024.02 | 70.3 | 89.3 | 93.9 | 86.2 | — | |
| InternVL-CZero-shot=true2024.02 | 70.3 | 89.2 | 93.8 | 86.6 | — | |
| EVA-CLIPParameters=8B, Zero-shot=true2024.12 | 70.1 | — | 93.1 | — | — | |
| EVA-01-CLIP-g/14+Zero-shot=true2024.02 | 69.4 | 88.3 | 93.2 | 85.3 | — | |
| EVA-02-CLIP-E/14+Zero-shot=true2024.02 | 69.4 | 88.6 | 93.3 | 85.7 | — | |
| OpenCLIP-G/14+Zero-shot=true2024.02 | 69 | 87.8 | 93.1 | 85.7 | — | |
| EVA-02-CLIPBackbone=ViT-E/14+, Zero-shot=true2024.12 | 68.8 | — | 92.8 | — | — | |
| VLM2VecBackbone=Mistral-7B, Zero-shot=true2024.12 | 68.2 | — | 93.2 | — | — | |
| OpenCLIPBackbone=ViT-BigG/14, Zero-shot=true2024.12 | 67.3 | — | 92.6 | — | — | |
| OpenCLIPBackbone=ViT-G/14, Zero-shot=true2024.12 | 66.3 | — | 91.8 | — | — | |
| BLIPBackbone=ViT-L, Zero-shot=true2024.12 | 63.5 | — | 92.5 | — | — | |
| SuperCLIPPretrain=L-12.8B, Zero-shot=true2025.12 | 63.1 | — | — | — | — | |
| BLIP2Backbone=ViT-L, Zero-shot=true2024.12 | 63 | — | 93.1 | — | — | |
| CLIPPretrain=L-12.8B, Zero-shot=true2025.12 | 62.5 | — | — | — | — | |
| E5-VBackbone=LLAVA-Next-8B, Zero-shot=true2024.12 | 62 | — | 89.7 | — | — | |
| EVA-01-CLIP-g/14Zero-shot=true2024.02 | 61.7 | 83.2 | 89.9 | 80.9 | — | |
| VLM2VecTrain Data=1M, MLLM=Qwen2-VL 7B2025.09 | 60.6 | 82.7 | 89.6 | — | — | |
| CLIPBackbone=ViT-L, Zero-shot=true2024.12 | 58.1 | — | 87.8 | — | — | |
| E5-VBackbone=LLAVA-1.5-7B, Zero-shot=true2024.12 | 57.4 | — | 88.4 | — | — | |
| FreeRet_{r20}Train Data=-, MLLM=Qwen2.5-VL 3B2025.09 | 55.3 | 79.6 | 84.1 | — | — | |
| FreeRet_{r20}Train Data=-, MLLM=Qwen2.5-VL 7B2025.09 | 53.8 | 81.3 | 88 | — | — | |
| SuperCLIPPretrain=L-512M, Zero-shot=true2025.12 | 52.2 | — | — | — | — | |
| CLIPPretrain=L-512M, Zero-shot=true2025.12 | 49.6 | — | — | — | — | |
| SuperCLIPPretrain=B-512M, Zero-shot=true2025.12 | 47.8 | — | — | — | — | |
| CLIPPretrain=B-512M, Zero-shot=true2025.12 | 46.7 | — | — | — | — | |
| FreeRet-embedTrain Data=-, MLLM=Qwen2.5-VL 3B2025.09 | 42 | 67.4 | 77.2 | — | — | |
| E5VTrain Data=-, MLLM=Qwen2.5-VL 7B2025.09 | 35.7 | 60.6 | 71.6 | — | — | |
| LLaVA-1.5-7BZero-shot=true2024.12 | 35.6 | — | 70.5 | — | — | |
| E5VTrain Data=-, MLLM=Qwen2.5-VL 3B2025.09 | 35.6 | 59.7 | 70.5 | — | — | |
| PTM-STPairs=500, Ratio=4.4‰2026.03 | 6.9 | 20.1 | 30 | — | — | |
| PTM-STNumber of distilled pairs=10002026.03 | 6.8 | 20.9 | 30.8 | — | — | |
| PTM-STPairs=200, Ratio=1.7‰2026.03 | 5.7 | 18.2 | 27.8 | — | — | |
| MTT-VLPairs=500, Ratio=4.4‰2026.03 | 5 | 17.2 | 26 | — | — | |
| LoRSNumber of distilled pairs=10002026.03 | 4.2 | 15.2 | 24.1 | — | — | |
| PTM-STPairs=100, Ratio=0.8‰2026.03 | 4.1 | 13.4 | 22 | — | — | |
| LoRSPairs=500, Ratio=4.4‰2026.03 | 4.1 | 12.8 | 20.2 | — | — | |
| EDGENumber of distilled pairs=10002026.03 | 3.9 | 13 | 21 | — | — | |
| HERDPairs=500, Ratio=4.4‰2026.03 | 3.8 | 7.8 | 13.7 | — | — | |
| K-CENTPairs=500, Ratio=4.4‰2026.03 | 3.7 | 8.7 | 13.8 | — | — | |
| LoRSPairs=200, Ratio=1.7‰2026.03 | 3.6 | 12.1 | 20.8 | — | — | |
| MTT-VLPairs=200, Ratio=1.7‰2026.03 | 3.3 | 11.9 | 19.4 | — | — | |
| RANDPairs=500, Ratio=4.4‰2026.03 | 3.2 | 7.5 | 12.5 | — | — | |
| LoRSPairs=100, Ratio=0.8‰2026.03 | 3 | 11 | 18.5 | — | — | |
| EDGEPairs=500, Ratio=4.4‰2026.03 | 2.9 | 9.5 | 15.7 | — | — | |
| MTT-VLPairs=100, Ratio=0.8‰2026.03 | 2.5 | 10 | 15.7 | — | — | |
| FORGETPairs=500, Ratio=4.4‰2026.03 | 2.3 | 8.2 | 13 | — | — | |
| K-CENTPairs=100, Ratio=0.8‰2026.03 | 1.4 | 3.7 | 5.5 | — | — | |
| K-CENTPairs=200, Ratio=1.7‰2026.03 | 1.2 | 3.8 | 7.5 | — | — | |
| FORGETPairs=200, Ratio=1.7‰2026.03 | 1.1 | 3.5 | 7 | — | — | |
| RANDPairs=200, Ratio=1.7‰2026.03 | 1 | 4 | 7.2 | — | — | |
| HERDPairs=200, Ratio=1.7‰2026.03 | 1 | 3.6 | 7.7 | — | — | |
| RANDPairs=100, Ratio=0.8‰2026.03 | 0.8 | 3 | 5 | — | — | |
| HERDPairs=100, Ratio=0.8‰2026.03 | 0.8 | 2.1 | 4.9 | — | — | |
| FORGETPairs=100, Ratio=0.8‰2026.03 | 0.7 | 2.6 | 4.8 | — | — | |
| SAFT-LAdversarial Attack=AutoAttack, Perturbation Budget (epsilon)=1/2552026.02 | — | — | — | — | 43.32 | |
| TGA-ZSRAdversarial Attack=AutoAttack, Perturbation Budget (epsilon)=1/2552026.02 | — | — | — | — | 41.64 |