Cross-modal Retrieval on InstVL img-zero 10K
28.25T2V R@1InstAP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| InstAP2026.04 | 28.25 | 31.87 | |
| UMT-L (InstVL; g+i)training_corpus=all InstVL captions treated as global2026.04 | 21.13 | 22.82 | |
| OpenCLIP2026.04 | 17.28 | 25.57 | |
| SigLIP2026.04 | 16.98 | 25.19 | |
| VideoPrism2026.04 | 13.85 | 20.04 | |
| UMT-L (InstVL; g)training_corpus=InstVL global captions2026.04 | 13.33 | 19.21 | |
| UMT-L2026.04 | 11.09 | 16.38 | |
| ViCLIP2026.04 | 9.57 | 11.21 | |
| CLIP4Clip2026.04 | 9.11 | 16.3 | |
| MCQ2026.04 | 7.04 | 8.55 | |
| CLIP-ViP2026.04 | 6.6 | 12.11 | |
| Coca2026.04 | 4.12 | 7.26 |