Video Compositional Reasoning on VELOCITI
87.97Accuracy (Ag Rand)SRL-CLIP
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| SRL-CLIPData=VidSitu (VS), Backbone=ViT-L/142024.01 | 87.97 | 55.17 | 63.93 | 60.26 | 55.01 | 55.16 | 48.7 | |
| SRL-CLIPData=VS+KS, Backbone=ViT-L/142024.01 | 87.17 | 52.98 | 64.38 | 58.95 | 55.09 | 54.28 | 49.59 | |
| SRL-CLIPData=Kinetics-SRL (KS), Backbone=ViT-L/142024.01 | 85.8 | 56.27 | 65.75 | 56.77 | 57.52 | 57.82 | 48.54 | |
| CLIPBackbone=ViT-L/142024.01 | 82.47 | 56.07 | 64.84 | 54.8 | 58.48 | 53.06 | 49.43 |