Cross-modal Retrieval on InstVL video (Global)
94.5T2V R@1InstAP
Evaluation results
| Method | Links | ||
|---|---|---|---|
| InstAP2026.04 | 94.5 | 95.5 | |
| UMT-L2026.04 | 88.3 | 85.5 | |
| UMT-L (InstVL; g)training_corpus=InstVL global captions2026.04 | 84.8 | 82.4 | |
| VideoPrism2026.04 | 82.71 | 83.62 | |
| OpenCLIP2026.04 | 82 | 77.15 | |
| UMT-L (InstVL; g+i)training_corpus=all InstVL captions treated as global2026.04 | 79.9 | 77.2 | |
| SigLIP2026.04 | 74.72 | 76.14 | |
| CLIP4Clip2026.04 | 67.5 | 70.5 | |
| ViCLIP2026.04 | 62.89 | 62.69 | |
| MCQ2026.04 | 61.48 | 60.67 | |
| Coca2026.04 | 46.92 | 43.78 | |
| CLIP-ViP2026.04 | 35.59 | 61.07 |