Multi-view video understanding on Ego-Exo4D Demonstrator Proficiency
44.2AccuracyPAVE-7B
Evaluation Results
| Method | Links | |
|---|---|---|
| PAVE-7BEvaluation protocol=Task-specific, FLOPs (TB)=98.70, Total Params=8.2B, Trainable Params=170.5M2025.03 | 44.2 | |
| TimeSFormer (Ego+Exo)*Evaluation protocol=Task-specific2025.03 | 43.7 | |
| PAVE-0.5BEvaluation protocol=Task-specific, FLOPs (TB)=8.15, Total Params=0.9B, Trainable Params=41.4M2025.03 | 32.4 | |
| LLaVA-OV-7B-FTEvaluation protocol=Task-specific fine-tuning, FLOPs (TB)=98.53, Total Params=8.2B, Trainable Params=161.5M2025.03 | 29.8 | |
| LLaVA-OV-0.5B-FTEvaluation protocol=Task-specific fine-tuning, FLOPs (TB)=8.01, Total Params=0.9B, Trainable Params=35.2M2025.03 | 28.2 | |
| LLaVA-OV-0.5BEvaluation protocol=Zero-shot, FLOPs (TB)=8.01, Total Params=0.9B2025.03 | 23.6 | |
| LLaVA-OV-7BEvaluation protocol=Zero-shot, FLOPs (TB)=98.53, Total Params=8.2B2025.03 | 23.6 |