Video-to-Text Retrieval on MSVD (test)
83.1R@1InternVideo2 6B
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| InternVideo2 6Bretrieval-stage=dual, zero-shot=true, parameters=6B2026.01 | 83.1 | 94.2 | — | — | — | — | |
| VIRTUE v2 7Bretrieval-stage=dual, zero-shot=true, parameters=7B2026.01 | 79 | 90.1 | — | — | — | — | |
| LanguageBindretrieval-stage=single, zero-shot=true, architecture=CLIP-based2026.01 | 75.1 | 90 | — | — | — | — | |
| VIRTUE-Embed 7Bretrieval-stage=single, zero-shot=true, architecture=MLLM-based, parameters=7B2026.01 | 73.4 | 87 | — | — | — | — | |
| Side4VideoBackbone=L/14, Pretrain=CLIP-400M, Protocol=Frozen backbone2023.11 | 71.7 | 94.4 | 97.9 | 1 | 2.3 | — | |
| RAPType=Adapter, Frozen visual encoder=true2024.05 | 69.6 | 91.9 | 95.7 | — | 2.9 | — | |
| Long-CLIP L/14retrieval-stage=single, zero-shot=true, architecture=CLIP-based, backbone=L/142026.01 | 69.3 | 86 | — | — | — | — | |
| PAU2023.09 | 68.9 | 93.1 | 97.1 | 1 | 2.4 | — | |
| CenterCLIPBackbone=ViT-B/16, Clustering Strategy=k-medoids++, Clustering Configuration=B6-4, 1602022.05 | 68.4 | 90.1 | 95 | 1 | 3 | — | |
| X-Pool2022.03 | 66.4 | 90 | 94.2 | 1 | 3.3 | — | |
| XPool2023.09 | 66.4 | 90 | 94.2 | 1 | 3.3 | — | |
| Side4VideoBackbone=E/14, Pretrain=LAION-2B, Protocol=Frozen backbone2023.11 | 65.9 | 90.6 | 95.4 | 1 | 2.5 | — | |
| RAPType=Adapter, Frozen visual encoder=false2024.05 | 65.2 | 88.7 | 93.1 | — | 4.2 | — | |
| LORAType=Adapter2024.05 | 64.3 | 87.3 | 92.5 | — | 4.1 | — | |
| CM AdapterPretrain=CLIP-400M, Protocol=Frozen backbone2023.11 | 63.6 | 90 | 94.7 | — | 3 | — | |
| CenterCLIPBackbone=ViT-B/32, Clustering Strategy=spectral, Clustering Configuration=B6-3, 492022.05 | 63.5 | 86.4 | 92.6 | 1 | 3.8 | — | |
| CenterCLIP2023.09 | 63.5 | 86.4 | 92.6 | 1 | 3.8 | — | |
| CLIP4Clipvariant=dagger symbol marked in paper2024.01 | 63.3 | 85.4 | 92.6 | — | — | 80.1 | |
| NeighborRetr2025.03 | 63.3 | 89.6 | 95.4 | 1 | 3.3 | — | |
| CenterCLIPBackbone=ViT-B/32, Clustering Strategy=spectral, Clustering Configuration=B6-4, 492022.05 | 62.7 | 88.1 | 92.8 | 1 | 4.1 | — | |
| CLIP4clip (meanP)Backbone=ViT-B/162022.05 | 62.7 | 83.9 | 89.4 | 1 | 6.1 | — | |
| Side4VideoBackbone=B/16, Pretrain=CLIP-400M, Protocol=Frozen backbone2023.11 | 62.3 | 85.3 | 89.8 | 1 | 7.7 | — | |
| DRL2024.01 | 62.3 | 86.3 | 92.2 | — | — | 80.3 | |
| CLIP4Clip-seqTransfTrainD=W+M, E2E=false2021.04 | 62 | 87.3 | 92.6 | 1 | 4.3 | — | |
| CLIP4ClipAggregation Strategy=seqTransf2022.03 | 62 | 87.3 | 92.6 | 1 | 4.3 | — | |
| CLIP4clip (seqTransf)2022.05 | 62 | 87.3 | 92.6 | 1 | 4.3 | — | |
| EERCF2024.01 | 62 | 87.1 | 94.3 | — | — | 81.1 | |
| CLIP4Clip-seqTransfpooling=seqTransf2023.09 | 62 | 87.3 | 92.6 | 1 | 4.3 | — | |
| CLIP4Clip2025.03 | 62 | 87.3 | 92.6 | 1 | 4.5 | — | |
| SSFType=Adapter2024.05 | 61.9 | 87 | 90.7 | — | 4.5 | — | |
| Diffusion2025.03 | 61.9 | 88.3 | 92.9 | 1 | 4.5 | — | |
| X-Clip2025.03 | 60.9 | 87.8 | — | — | 4.7 | — | |
| Baseline2023.09 | 60.1 | 82.8 | 87.3 | 1 | 7.5 | — | |
| CLIP-straightTrainD=W, E2E=false2021.04 | 59.9 | 85.2 | 90.7 | 1 | — | — | |
| Straight-CLIP2022.03 | 59.9 | 85.2 | 90.7 | 1 | — | — | |
| CLIPEvaluation Protocol=zero-shot2022.05 | 59.9 | 85.2 | 90.7 | 1 | — | — | |
| VPTType=Prompt2024.05 | 59.5 | 82.9 | 88.8 | — | 5.9 | — | |
| X-CLIPvariant=dagger symbol marked in paper2024.01 | 59 | 83.1 | 86.5 | — | — | 76.2 | |
| CLIP2Video2023.09 | 58.7 | 85.6 | 91.6 | 1 | 4.3 | — | |
| Side4VideoBackbone=B/32, Pretrain=CLIP-400M, Protocol=Frozen backbone2023.11 | 58.1 | 83.6 | 90.1 | 1 | 7.7 | — | |
| CenterCLIPBackbone=ViT-B/32, Clustering Strategy=k-medoids++, Clustering Configuration=B6-3, 492022.05 | 57.9 | 83.6 | 90.5 | 1 | 5.2 | — | |
| CLIP4Clip (Frozen ViT)Pretrain=CLIP-400M, Protocol=Frozen backbone2023.11 | 57 | 79.2 | 84.9 | 1 | 12.1 | — | |
| CLIP4Clip-meanPTrainD=W+M, E2E=false2021.04 | 56.6 | 79.7 | 84.3 | 1 | 7.6 | — | |
| CLIP4ClipAggregation Strategy=meanP2022.03 | 56.6 | 79.7 | 84.3 | 1 | 7.6 | — | |
| CLIP4clip (meanP)2022.05 | 56.6 | 79.7 | 84.3 | 1 | 7.6 | — | |
| CLIP4ClipPretrain=CLIP-400M, Protocol=Full Fine-tuning2023.11 | 56.6 | 79.7 | 84.3 | 1 | 7.6 | — | |
| CLIP4Clip-meanPpooling=meanP2023.09 | 56.6 | 79.7 | 84.3 | 1 | 7.6 | — | |
| CLIP4ClipType=Fine-tune, Frozen visual encoder=false2024.05 | 56.6 | 79.7 | 84.3 | — | 7.6 | — | |
| CLIP4ClipType=Fine-tune, Frozen visual encoder=true2024.05 | 56.3 | 82.6 | 89.8 | — | 4.8 | — | |
| TS2-Netvariant=dagger symbol marked in paper2024.01 | 55.6 | 77.2 | 83 | — | — | 71.9 | |
| CLIP4Clip-tightTransfTrainD=W+M, E2E=true2021.04 | 54.3 | 85.3 | 91 | 1 | 6 | — | |
| EMCL-Net2025.03 | 54.3 | 81.3 | 88.1 | 1 | 5.6 | — | |
| CenterCLIPBackbone=ViT-B/32, Clustering Strategy=k-medoids++, Clustering Configuration=B6-4, 492022.05 | 54.2 | 78.4 | 84.9 | 1 | 7.6 | — | |
| CoOpType=Prompt2024.05 | 53.8 | 78.3 | 82.9 | — | 12.4 | — | |
| ST-AdapterType=Adapter2024.05 | 53.6 | 80.5 | 87.3 | — | 5.8 | — | |
| HAT-VTRZero-shot=true2026.02 | 53.28 | 79.1 | 86.27 | — | — | — | |
| CLIP4Clip-seqLSTMTrainD=W+M, E2E=false2021.04 | 52.5 | 74 | 78.1 | 1 | 14.7 | — | |
| CLIP4clip (meanP)Backbone=ViT-B/322022.05 | 51 | 76.3 | 82.2 | 1 | 9.1 | — | |
| TCRZero-shot=true2026.02 | 47.31 | 75.67 | 85.97 | — | — | — | |
| EATAZero-shot=true2026.02 | 44.78 | 74.93 | 84.48 | — | — | — | |
| SARZero-shot=true2026.02 | 44.48 | 73.88 | 83.43 | — | — | — | |
| TentZero-shot=true2026.02 | 44.33 | 74.18 | 83.88 | — | — | — | |
| CLIPZero-shot=true2026.02 | 44.03 | 73.73 | 82.54 | — | — | — | |
| READZero-shot=true2026.02 | 43.58 | 73.28 | 82.24 | — | — | — | |
| TT-CE+TrainD=W, E2E=false2021.04 | 27.1 | 55.3 | 67.1 | 4 | — | — | |
| TeachText-CE+2022.03 | 27.1 | 55.3 | 67.1 | 4 | — | — | |
| TT-CE+2022.05 | 27.1 | 55.3 | 67.1 | 1 | — | — | |
| TeachText2024.01 | 27.1 | 55.3 | 67.1 | — | — | 49.8 |