Cross-modal Retrieval on InstVL img 10K (Global)
95.77T2V Recall@1InstAP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| InstAP2026.04 | 95.77 | 94.71 | |
| VideoPrism2026.04 | 88.19 | 89.62 | |
| SigLIP2026.04 | 87.18 | 91.97 | |
| UMT-L (InstVL; g)training_corpus=InstVL global captions2026.04 | 85.7 | 87.03 | |
| OpenCLIP2026.04 | 84.98 | 92.06 | |
| UMT-L2026.04 | 83.95 | 85.41 | |
| ViCLIP2026.04 | 81.47 | 79.33 | |
| UMT-L (InstVL; g+i)training_corpus=all InstVL captions treated as global2026.04 | 80.3 | 81.62 | |
| CLIP4Clip2026.04 | 79.22 | 84.25 | |
| Coca2026.04 | 70.8 | 76.16 | |
| CLIP-ViP2026.04 | 54.94 | 72 | |
| MCQ2026.04 | 31.45 | 34.12 |