Failure Detection and Reasoning on Sparrow
73.3Detection AccuracyARMOR
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ARMOR2026.02 | 73.3 | 50.3 | 51.4 | |
| Claude-3.7shots=32026.02 | 65 | 40.7 | 45.8 | |
| SFT-S+D2026.02 | 62 | 24.5 | 25.7 | |
| SFT-D2026.02 | 52.3 | 27.8 | 27.4 | |
| Claude-3.7shots=02026.02 | 51.7 | 21.3 | 13.8 | |
| Cosmos-Reasoning2026.02 | 51 | 34.6 | 31.1 | |
| LLaVA-NeXT2026.02 | 50 | 0.8 | 7.8 | |
| Qwen2.5-VLshots=02026.02 | 45.3 | 24 | 13.7 |