Proficiency estimation on EgoExo4D Ego+Exos view
48.2Top-1 AccuracyProfVLM (AGP)
Evaluation Results
| Method | Links | |
|---|---|---|
| ProfVLM (AGP)Pretrain=K600 + SmolLM2, Frames=8, Train Params=5.3M, Epochs=62025.09 | 48.2 | |
| ProfVLM (AGP)Params=5.3M, Frames=8, Epochs=6, Paradigm=Generative2026.05 | 48.2 | |
| ProfVLM (AGP)Pretrain=K600+SmolLM2, Params=5.3M, Frames=8, Epochs=6, Type=Generative2026.06 | 48.2 | |
| SkillFormer+PATSParams=27M, Frames=24–32, Epochs=4, Paradigm=Discriminative (Classification)2026.05 | 48 | |
| SkillFormer+PATSPretrain=K600, Params=14–27M, Frames=24–32, Epochs=4, Type=Discriminative2026.06 | 48 | |
| SkillMoVPretrain=K600, Params=36.9M, Frames=16, Epochs=20, Type=Discriminative2026.06 | 47.6 | |
| SkillFormerPretrain=K600, Frames=16/24/32, Train Params=14/20/27M, Epochs=42025.09 | 47.5 | |
| SkillFormerParams=27M, Frames=16–32, Epochs=4, Paradigm=Discriminative (Classification)2026.05 | 47.5 | |
| SkillFormerPretrain=K600, Params=14–27M, Frames=16–32, Epochs=4, Type=Discriminative2026.06 | 47.5 | |
| ProfVLM (MLP)Pretrain=K600 + SmolLM2, Frames=8, Train Params=2.2M, Epochs=62025.09 | 42.9 | |
| EgoPulseFormerPretrain=EgoPPG-DB, Frames=16, Train Params=121M, Epochs=152025.09 | 42.4 | |
| EgoPulseFormerParams=121M, Frames=16, Epochs=15, Paradigm=Discriminative (Classification)2026.05 | 42.4 | |
| EgoPulseFormerPretrain=EgoPPG-DB, Params=121M, Frames=16, Epochs=15, Type=Discriminative2026.06 | 42.4 | |
| TimeSFormerPretrain=-, Frames=16, Train Params=121M, Epochs=152025.09 | 40.8 | |
| Ego-Exo4D Baselines (TimeSformer)Params=121M, Frames=16, Epochs=15, Paradigm=Discriminative (Classification)2026.05 | 40.8 | |
| TimeSformerPretrain=–, Params=121M, Frames=16, Epochs=15, Type=Discriminative2026.06 | 40.8 | |
| TimeSFormerPretrain=HowTo100M, Frames=16, Train Params=121M, Epochs=152025.09 | 39.7 | |
| TimeSformerPretrain=HowTo100M, Params=121M, Frames=16, Epochs=15, Type=Discriminative2026.06 | 39.7 | |
| TimeSFormerPretrain=EgoVLP, Frames=16, Train Params=121M, Epochs=152025.09 | 39.5 | |
| TimeSformerPretrain=EgoVLP, Params=121M, Frames=16, Epochs=15, Type=Discriminative2026.06 | 39.5 | |
| TimeSFormerPretrain=K400, Frames=16, Train Params=121M, Epochs=152025.09 | 38.6 | |
| TimeSformerPretrain=K400, Params=121M, Frames=16, Epochs=15, Type=Discriminative2026.06 | 38.6 | |
| TimeSFormerPretrain=EgoVLPv2, Frames=16, Train Params=121M, Epochs=152025.09 | 37.8 | |
| TimeSformerPretrain=EgoVLPv2, Params=121M, Frames=16, Epochs=15, Type=Discriminative2026.06 | 37.8 | |
| SmolVLM2Pretrain=-, Train Params=256M, Evaluation Protocol=0-shot2025.09 | 32.2 | |
| Majority-classPretrain=-2025.09 | 31.1 | |
| SmolVLMPretrain=-, Train Params=256M, Evaluation Protocol=0-shot2025.09 | 31.1 | |
| Majority2026.05 | 31.1 | |
| Majority2026.06 | 31.1 | |
| PandaGPTPretrain=ImageBind + Vicuna, Train Params=52M, Evaluation Protocol=0-shot2025.09 | 25.8 | |
| RandomPretrain=-2025.09 | 24.9 | |
| Random2026.05 | 24.9 | |
| Random2026.06 | 24.9 |