Spatial Reasoning on Isle Bricks V2
100Alignment ScoreLLaVA
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LLaVAModel Scale=1.5-13B2026.01 | 100 | 27 | 52 | |
| Emb. TextProtocol=Direct2026.01 | 100 | 64 | 76 | |
| Emb. Tokens (COCO)Protocol=Direct2026.01 | 100 | 73 | 82 | |
| Emb. Tokens (COCO)Protocol=CoT2026.01 | 100 | 82 | 88 | |
| Rotation TextProtocol=Direct2026.01 | 100 | 86 | 91 | |
| Rotation TokensProtocol=Direct2026.01 | 91 | 82 | 85 | |
| Emb. TextProtocol=CoT2026.01 | 82 | 68 | 73 | |
| Emb. Tokens (ViTPose)Protocol=Direct2026.01 | 82 | 64 | 70 | |
| Emb. Tokens (ViTPose)Protocol=CoT2026.01 | 82 | 68 | 73 | |
| Rotation TextProtocol=CoT2026.01 | 82 | 82 | 82 | |
| Rotation TokensProtocol=CoT2026.01 | 82 | 91 | 88 |