Self-other distinction on Robot evaluation setting (Held-out)
99.5AccuracyOurs
Evaluation results
| Method | Links | |
|---|---|---|
| Ours2026.06 | 99.5 | |
| Direct fusion2026.06 | 95.78 | |
| GPT-4o2026.06 | 61.67 | |
| Gemini 1.5 Pro Preview2026.06 | 59.44 | |
| InstructBLIPBackbone=Vicuna 7B2026.06 | 51.28 | |
| LLaVA v1.6Backbone=Mistral 7B2026.06 | 48.78 |