3D Object Captioning on ABO 6.4k objects
82.3CLIPScoreTri-MARF
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Tri-MARFSpeed (objects/hour)=12k2026.01 | 82.3 | — | 37.1 | 35.6 | |
| Human AnnotationSpeed (objects/hour)=0.12k2026.01 | 78.9 | 2.9 | 33.8 | 32.5 | |
| ScoreAggSpeed (objects/hour)=9k2026.01 | 76.2 | 4.2 | 31.2 | 30 | |
| Cap3DSpeed (objects/hour)=8k2026.01 | 74.8 | 3.5 | 28.9 | 27.3 | |
| 3D-LLMSpeed (objects/hour)=6.5k2026.01 | 73 | 3.3 | 28.4 | 26.9 | |
| ULIP-2Speed (objects/hour)=7k2026.01 | 71.4 | 3.1 | 26.5 | 25 | |
| Metadata2026.01 | 61.5 | 2.1 | 16.3 | 15 | |
| PointCLIPSpeed (objects/hour)=5k2026.01 | 60.7 | 2.2 | 17.2 | 15.7 | |
| GPT4PointSpeed (objects/hour)=4k2026.01 | 58.2 | 1.9 | 14.6 | 13.1 |