Cross-modal Retrieval on InstVL img 1K (Global)
99.2T2V R@1InstAP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| InstAP2026.04 | 99.2 | 99.1 | |
| VideoPrism2026.04 | 97.4 | 97.6 | |
| UMT-L (InstVL; g)training_corpus=InstVL global captions2026.04 | 96.2 | 97.1 | |
| SigLIP2026.04 | 95.7 | 98.2 | |
| ViCLIP2026.04 | 95.1 | 93.5 | |
| UMT-L2026.04 | 94.7 | 95.3 | |
| OpenCLIP2026.04 | 94.4 | 98.1 | |
| CLIP4Clip2026.04 | 93.4 | 96 | |
| UMT-L (InstVL; g+i)training_corpus=all InstVL captions treated as global2026.04 | 93.2 | 94.3 | |
| Coca2026.04 | 86.2 | 91.5 | |
| CLIP-ViP2026.04 | 78.4 | 89.2 | |
| MCQ2026.04 | 58.2 | 60.1 |