Cross-modal Retrieval on InstVL img-zero 1K (Instance)
41.94T2V R@1InstAP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| InstAP2026.04 | 41.94 | 42.53 | |
| UMT-L (InstVL; g+i)training_corpus=all InstVL captions treated as global2026.04 | 34.68 | 34.99 | |
| UMT-L2026.04 | 29.34 | 30.17 | |
| SigLIP2026.04 | 28.25 | 35.56 | |
| OpenCLIP2026.04 | 26.73 | 36.19 | |
| UMT-L (InstVL; g)training_corpus=InstVL global captions2026.04 | 25.97 | 31.97 | |
| VideoPrism2026.04 | 21.32 | 27.39 | |
| ViCLIP2026.04 | 18.25 | 20.93 | |
| CLIP4Clip2026.04 | 17.82 | 25.1 | |
| MCQ2026.04 | 17.08 | 19.61 | |
| CLIP-ViP2026.04 | 13.81 | 22.96 | |
| Coca2026.04 | 7.08 | 13.19 |