Task-level mistake detection on EgoPER
83.3Quesadilla O-AccZeProM (Q-L)
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ZeProM (Q-L)Supervision level=Zero-shot, Model variant=Q-L2026.06 | 83.3 | 52.6 | 93.6 | 80.6 | 65.2 | 46.6 | 52.6 | 10.5 | 66 | 27 | 72.1 | 43.5 | |
| ZeProM (Q-M)Supervision level=Zero-shot, Model variant=Q-M2026.06 | 79.2 | 48.1 | 96.8 | 46.9 | 66.1 | 33.3 | 47.4 | 8.33 | 44.7 | 17.5 | 66.8 | 30.8 | |
| EgoPEDSupervision level=Fully supervised2026.06 | 77.1 | 66.1 | 82.5 | 72.2 | 63.2 | 54 | 47.4 | 39.1 | 55.3 | 39.4 | 65.1 | 54.2 | |
| GTG2VidSupervision level=Fully supervised2026.06 | 64.6 | 50.8 | 95.2 | 70.6 | 76.1 | 47.8 | 52.6 | 30.3 | 87.2 | 42.7 | 75.1 | 48.4 |