3D Object Captioning on Objaverse-XL (5k sampled)
86.1CLIPScoreTri-MARF
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Tri-MARFSpeed (objects/hour)=12k2026.01 | 86.1 | — | 40.5 | 38.9 | |
| Human AnnotationSpeed (objects/hour)=0.12k2026.01 | 81 | 2.9 | 37 | 35.5 | |
| ScoreAggSpeed (objects/hour)=9k2026.01 | 78.5 | 3.7 | 34.5 | 33 | |
| Cap3DSpeed (objects/hour)=8k2026.01 | 76.4 | 3.5 | 32.1 | 30.5 | |
| 3D-LLMSpeed (objects/hour)=6.5k2026.01 | 75.6 | 3.4 | 31.8 | 30.3 | |
| ULIP-2Speed (objects/hour)=7k2026.01 | 73.8 | 3.2 | 29.7 | 28.2 | |
| PointCLIPSpeed (objects/hour)=5k2026.01 | 63.1 | 2.3 | 19.5 | 18 | |
| GPT4PointSpeed (objects/hour)=4k2026.01 | 60.5 | 2 | 16.3 | 14.8 |