Cross-modal Retrieval on InstVL img-zero 1K (Global)
88.7T2V Recall@1InstAP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| InstAP2026.04 | 88.7 | 88.3 | |
| VideoPrism2026.04 | 85.7 | 85.8 | |
| UMT-L (InstVL; g)training_corpus=InstVL global captions2026.04 | 85.3 | 86.4 | |
| SigLIP2026.04 | 83.9 | 86.5 | |
| UMT-L2026.04 | 83.9 | 83.7 | |
| OpenCLIP2026.04 | 83.4 | 86.9 | |
| UMT-L (InstVL; g+i)training_corpus=all InstVL captions treated as global2026.04 | 82.4 | 84.3 | |
| CLIP4Clip2026.04 | 78.2 | 81.7 | |
| ViCLIP2026.04 | 77.8 | 77.6 | |
| Coca2026.04 | 67.4 | 70.5 | |
| MCQ2026.04 | 58.9 | 62.7 | |
| CLIP-ViP2026.04 | 55.6 | 73.2 |