Video-to-Text Retrieval on MSRVTT 1kA (Noise Robustness)
41.6Recall@1 (20% Noise)HAT-VTR
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| HAT-VTRBackbone=Xpool, Adaptation Method=Ours2026.02 | 41.6 | 37.6 | 32.3 | 29 | 35.1 | |
| EATABackbone=Xpool, Adaptation Method=EATA2026.02 | 39.9 | 34.9 | 29.3 | 25.9 | 32.5 | |
| TentBackbone=Xpool, Adaptation Method=Tent2026.02 | 39 | 34.1 | 29.2 | 23.7 | 31.5 | |
| SARBackbone=Xpool, Adaptation Method=SAR2026.02 | 38.9 | 34.2 | 29.3 | 23.8 | 31.6 | |
| TCRBackbone=Xpool, Adaptation Method=TCR2026.02 | 38.6 | 34.1 | 29.3 | 24.3 | 31.6 | |
| READBackbone=Xpool, Adaptation Method=READ2026.02 | 38.5 | 33.7 | 28.5 | 23.4 | 31 | |
| XpoolBackbone=Xpool, Adaptation Method=None2026.02 | 38.4 | 33.6 | 28.8 | 23.8 | 31.2 | |
| HAT-VTRBackbone=CLIP4Clip, Adaptation Method=Ours2026.02 | 37.4 | 34.4 | 29.3 | 27.6 | 32.2 | |
| CLIP4ClipBackbone=CLIP4Clip, Adaptation Method=None2026.02 | 36.7 | 31.1 | 27.2 | 21 | 29 | |
| EATABackbone=CLIP4Clip, Adaptation Method=EATA2026.02 | 36.6 | 31.6 | 27.7 | 23.7 | 29.9 | |
| SARBackbone=CLIP4Clip, Adaptation Method=SAR2026.02 | 36.4 | 31.1 | 27.3 | 21.4 | 29.1 | |
| TentBackbone=CLIP4Clip, Adaptation Method=Tent2026.02 | 36.3 | 31.1 | 27.2 | 21.5 | 29 | |
| READBackbone=CLIP4Clip, Adaptation Method=READ2026.02 | 36.1 | 30.7 | 26.9 | 21 | 28.7 | |
| TCRBackbone=CLIP4Clip, Adaptation Method=TCR2026.02 | 36.1 | 30.5 | 26.5 | 21.7 | 28.7 |