Self-other distinction on Robot + Human (Held-out)
99.69AccuracyOurs
Evaluation Results
| Method | Links | |
|---|---|---|
| Ours2026.06 | 99.69 | |
| LLaVA v1.6Backbone=Mistral 7B2026.06 | 77.02 | |
| Direct fusion2026.06 | 75.09 | |
| Gemini 1.5 Pro Preview2026.06 | 62.91 | |
| InstructBLIPBackbone=Vicuna 7B2026.06 | 46.9 | |
| GPT-4o2026.06 | 43.05 |