Cross-modal Retrieval on InstVL img-zero 10K (Global)
83.33T2V R@1InstAP
Evaluation results
| Method | Links | ||
|---|---|---|---|
| InstAP2026.04 | 83.33 | 82.21 | |
| VideoPrism2026.04 | 73.05 | 75.11 | |
| UMT-L2026.04 | 72.6 | 72.59 | |
| UMT-L (InstVL; g)training_corpus=InstVL global captions2026.04 | 72.5 | 74.18 | |
| OpenCLIP2026.04 | 70.75 | 78.13 | |
| SigLIP2026.04 | 68.64 | 75.66 | |
| UMT-L (InstVL; g+i)training_corpus=all InstVL captions treated as global2026.04 | 68.16 | 69.76 | |
| ViCLIP2026.04 | 58.51 | 58.21 | |
| CLIP4Clip2026.04 | 56.95 | 63.96 | |
| Coca2026.04 | 46.05 | 50.64 | |
| MCQ2026.04 | 34.13 | 38.26 | |
| CLIP-ViP2026.04 | 32.48 | 51.3 |