3D-assisted Dialog on 3D-LLM Held-In Dataset 1.0 (test)
47.4BLEU-1Agentic Collaborative Cognition
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Agentic Collaborative CognitionSupervision=Zero-shot2026.06 | 47.4 | 24 | 28.6 | 47.3 | |
| 3D-LLMSupervision=Fully Supervised2026.06 | 39 | 16.6 | 18.9 | 39.3 | |
| Agent3D-ZeroSupervision=Zero-shot2026.06 | 32.8 | 9.8 | 19.3 | 39.3 | |
| BLIP2-flant5Supervision=Fully Supervised2026.06 | 32.4 | 9.5 | 11 | 29.5 | |
| flamingoSupervision=Fully Supervised2026.06 | 30.6 | 9.1 | 10.4 | 27.9 | |
| flant5Supervision=Fully Supervised2026.06 | 27.4 | 8.7 | 9.5 | 27.5 |