Text-to-Video Retrieval on MSR-VTT 1K (test)
93.61R@1ViPro w/o ϵ
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| ViPro w/o ϵVictim Model=C4V[8]2025.08 | 93.61 | 98.56 | — | — | — | 96.09 | — | — | — | |
| ViPro w/o ϵVictim Model=DRL[7]2025.08 | 87.37 | 97.09 | — | — | — | 92.23 | — | — | — | |
| ViProVictim Model=C4V[8]2025.08 | 75.2 | 88.91 | — | — | — | 82.06 | — | — | — | |
| Co-Attack (SIMT2V)Victim Model=C4V[8]2025.08 | 75.04 | 88.57 | — | — | — | 81.81 | — | — | — | |
| SGAVictim Model=C4V[8]2025.08 | 72.13 | 86.99 | — | — | — | 79.56 | — | — | — | |
| ViPro w/o ϵVictim Model=Sing[6]2025.08 | 66.85 | 82.53 | — | — | — | 74.69 | — | — | — | |
| ViProVictim Model=Singularity, Attack Setting=White-box, Constraint=None (w/o epsilon)2025.08 | 66.85 | 82.53 | — | — | — | 74.69 | 61.98 | 59.59 | 60.79 | |
| ViProVictim Model=DRL[7]2025.08 | 61.24 | 84.62 | — | — | — | 72.93 | — | — | — | |
| Co-Attack (SIMT2V)Victim Model=DRL[7]2025.08 | 61.07 | 83.89 | — | — | — | 72.48 | — | — | — | |
| InternVideo2 6BRetrieval Stage=Dual, Parameters=6B, Zero-shot=true2026.01 | 55.9 | 78.3 | — | — | — | — | — | — | — | |
| VIRTUE v2 7BRetrieval Stage=Dual, Parameters=7B, Zero-shot=true2026.01 | 55.3 | 76.6 | — | — | — | — | — | — | — | |
| SGAVictim Model=DRL[7]2025.08 | 53.53 | 80.37 | — | — | — | 66.95 | — | — | — | |
| VIRTUE v1 7BRetrieval Stage=Dual, Parameters=7B, Zero-shot=true2026.01 | 52.4 | 76 | — | — | — | — | — | — | — | |
| Side4Video E/14Pretrain=LAION-2B, Fine-tuning Protocol=Frozen backbone, Backbone=ViT-E/142023.11 | 52.3 | 75.5 | 84.2 | 1 | 12.8 | — | — | — | — | |
| InternVideo2V-Enc Params=1.0B, Video Data=102M, Resolution=224, Frame/ fps=8, Zero-shot Protocol=true2025.12 | 51.9 | — | — | — | — | — | — | — | — | |
| PEAV-LV-Enc Params=0.5B, Video Data=124M, Resolution=336, Frame/ fps=30*, Zero-shot Protocol=true2025.12 | 51.9 | — | — | — | — | — | — | — | — | |
| Cap4VideoPretrain=CLIP-400M, Fine-tuning Protocol=Full Fine-tuning2023.11 | 51.4 | 75.7 | 83.9 | 1 | 12.4 | — | — | — | — | |
| Side4Video L/14Pretrain=CLIP-400M, Fine-tuning Protocol=Frozen backbone, Backbone=ViT-L/142023.11 | 51.4 | 75.8 | 84.5 | 1 | 12.5 | — | — | — | — | |
| PE-GV-Enc Params=1.9B, Video Data=22M, Resolution=448, Frame/ fps=8, Zero-shot Protocol=true2025.12 | 51.2 | — | — | — | — | — | — | — | — | |
| UniAdapterPretrain=BLIP-129M, Fine-tuning Protocol=Frozen backbone2023.11 | 50.5 | 73.9 | 81.7 | 1 | — | — | — | — | — | |
| PE-LV-Enc Params=0.3B, Video Data=22M, Resolution=336, Frame/ fps=8, Zero-shot Protocol=true2025.12 | 50.3 | — | — | — | — | — | — | — | — | |
| STANPretrain=CLIP-400M, Fine-tuning Protocol=Full Fine-tuning2023.11 | 50 | 75.2 | 84.1 | 1.5 | — | — | — | — | — | |
| PEAV-SV-Enc Params=0.3B, Video Data=124M, Resolution=336, Frame/ fps=30*, Zero-shot Protocol=true2025.12 | 49.3 | — | — | — | — | — | — | — | — | |
| PEAV-LV-Enc Params=0.5B, Video Data=124M, Resolution=336, Frame/ fps=16, Zero-shot Protocol=true2025.12 | 49 | — | — | — | — | — | — | — | — | |
| LanguageBindV-Enc Params=0.3B, Video Data=10M, Resolution=224, Frame/ fps=16, Zero-shot Protocol=true2025.12 | 48.6 | — | — | — | — | — | — | — | — | |
| PEAV-BV-Enc Params=0.4B, Video Data=124M, Resolution=336, Frame/ fps=16, Zero-shot Protocol=true2025.12 | 48.6 | — | — | — | — | — | — | — | — | |
| ViProVictim Model=Sing[6]2025.08 | 47.81 | 73.66 | — | — | — | 60.74 | — | — | — | |
| ViProVictim Model=Singularity, Attack Setting=White-box2025.08 | 47.81 | 73.66 | — | — | — | 60.74 | 42.94 | 50.72 | 46.83 | |
| PEAV-BV-Enc Params=0.4B, Video Data=124M, Resolution=336, Frame/ fps=30*, Zero-shot Protocol=true2025.12 | 47.7 | — | — | — | — | — | — | — | — | |
| Side4Video B/16Pretrain=CLIP-400M, Fine-tuning Protocol=Frozen backbone, Backbone=ViT-B/162023.11 | 47.2 | 73.8 | 83.7 | 2 | 13.1 | — | — | — | — | |
| VIRTUE-Embed 7BRetrieval Stage=Single, Architecture Type=MLLM-based, Parameters=7B, Zero-shot=true2026.01 | 46.8 | 70.8 | — | — | — | — | — | — | — | |
| PEAV-SV-Enc Params=0.3B, Video Data=124M, Resolution=336, Frame/ fps=16, Zero-shot Protocol=true2025.12 | 46.7 | — | — | — | — | — | — | — | — | |
| InternVL-GFrames=8, Zero-shot=true2023.12 | 46.3 | 70.5 | 79.6 | — | — | 63.4 | — | — | — | |
| CLIP2VideoPretrain=CLIP-400M, Fine-tuning Protocol=Full Fine-tuning2023.11 | 45.6 | 72.6 | 81.7 | 2 | 14.6 | — | — | — | — | |
| CM AdapterPretrain=CLIP-400M, Fine-tuning Protocol=Frozen backbone2023.11 | 45.4 | 73.3 | 82.3 | — | 12.8 | — | — | — | — | |
| LanguageBindFrames=8, Zero-shot=true, Temporal attention layers=true2023.12 | 44.8 | 70 | 78.7 | — | — | 62.8 | — | — | — | |
| InternVL-CFrames=8, Zero-shot=true2023.12 | 44.7 | 68.2 | 78.4 | — | — | 61.5 | — | — | — | |
| LamRA 7BRetrieval Stage=Dual, Parameters=7B, Zero-shot=true2026.01 | 44.7 | 68.6 | — | — | — | — | — | — | — | |
| InternVLV-Enc Params=5.5B, Resolution=224, Frame/ fps=8, Zero-shot Protocol=true2025.12 | 44.7 | — | — | — | — | — | — | — | — | |
| Side4Video B/32Pretrain=CLIP-400M, Fine-tuning Protocol=Frozen backbone, Backbone=ViT-B/322023.11 | 44.2 | 71.1 | 81 | 2 | 15.1 | — | — | — | — | |
| CaRe 7BRetrieval Stage=Single, Architecture Type=MLLM-based, Parameters=7B, Zero-shot=true2026.01 | 43.9 | 67 | — | — | — | — | — | — | — | |
| VLM2Vec 7BRetrieval Stage=Single, Architecture Type=MLLM-based, Parameters=7B, Zero-shot=true2026.01 | 43.5 | — | — | — | — | — | — | — | — | |
| CLIP4ClipPretrain=CLIP-400M, Fine-tuning Protocol=Full Fine-tuning2023.11 | 43.1 | 70.4 | 80.8 | 2 | 16.2 | — | — | — | — | |
| SigLIP2-g-optV-Enc Params=1.1B, Resolution=384, Frame/ fps=8, Zero-shot Protocol=true2025.12 | 43.1 | — | — | — | — | — | — | — | — | |
| UMT-LFrames=8, Zero-shot=true, Temporal attention layers=true2023.12 | 42.6 | 64.4 | 73.1 | — | — | 58 | — | — | — | |
| LanguageBindRetrieval Stage=Single, Architecture Type=CLIP-based, Zero-shot=true2026.01 | 42.1 | 65.9 | — | — | — | — | — | — | — | |
| SigLIP2-L/16V-Enc Params=0.3B, Resolution=384, Frame/ fps=8, Zero-shot Protocol=true2025.12 | 41.5 | — | — | — | — | — | — | — | — | |
| Long-CLIP L/14Retrieval Stage=Single, Architecture Type=CLIP-based, Backbone=L/14, Zero-shot=true2026.01 | 40.9 | 65.5 | — | — | — | — | — | — | — | |
| UMT-LV-Enc Params=0.3B, Video Data=25M, Resolution=224, Frame/ fps=8, Zero-shot Protocol=true2025.12 | 40.7 | — | — | — | — | — | — | — | — | |
| ImageBindV-Enc Params=0.6B, Video Data=3M, Resolution=224, Frame/ fps=16, Zero-shot Protocol=true2025.12 | 40.6 | — | — | — | — | — | — | — | — | |
| CLIP4Clip (ViT)Pretrain=CLIP-400M, Fine-tuning Protocol=Frozen backbone2023.11 | 40 | 67.8 | 78.4 | 2 | 17.7 | — | — | — | — | |
| InternVL-GFrames=1, Zero-shot=true2023.12 | 39.1 | 61.7 | 70.7 | — | — | 55.7 | — | — | — | |
| FlorenceFrames=8, Zero-shot=true2023.12 | 37.6 | 63.8 | 72.6 | — | — | — | — | — | — | |
| InternVL-CFrames=1, Zero-shot=true2023.12 | 37.5 | 60.9 | 70.9 | — | — | 54.6 | — | — | — | |
| CLIP-PromptPretrain=CLIP-400M, Fine-tuning Protocol=Frozen backbone2023.11 | 36.7 | 64.6 | 76.8 | 2 | — | — | — | — | — | |
| OpenAI CLIP-LFrames=8, Zero-shot=true2023.12 | 30.7 | 54.4 | 64 | — | — | 48.1 | — | — | — | |
| OpenAI CLIP-LFrames=1, Zero-shot=true2023.12 | 29 | 50.5 | 59.2 | — | — | 45.7 | — | — | — | |
| SGAVictim Model=Sing[6]2025.08 | 19.3 | 46.77 | — | — | — | 33.04 | — | — | — | |
| SGAVictim Model=Singularity, Attack Setting=White-box2025.08 | 19.3 | 46.77 | — | — | — | 33.04 | 14.43 | 23.83 | 19.13 | |
| Co-Attack (EMBMul)Victim Model=Sing[6]2025.08 | 9.51 | 25.12 | — | — | — | 17.32 | — | — | — | |
| Co-AttackVictim Model=Singularity, Attack Setting=White-box, Attack Objective=EMBMul2025.08 | 9.51 | 25.12 | — | — | — | 17.32 | 4.64 | 2.18 | 3.41 | |
| No AttackVictim Model=Sing[6]2025.08 | 4.87 | 22.94 | — | — | — | 13.91 | — | — | — | |
| No AttackVictim Model=Singularity, Attack Setting=White-box2025.08 | 4.87 | 22.94 | — | — | — | 13.91 | — | — | — | |
| No AttackVictim Model=DRL[7]2025.08 | 4.29 | 15.76 | — | — | — | 10.03 | — | — | — | |
| No AttackVictim Model=C4V[8]2025.08 | 3.76 | 13.42 | — | — | — | 8.59 | — | — | — |