3D Object Captioning on Objaverse-LVIS 1k sampled
88.7CLIPScoreTri-MARF
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Tri-MARFSpeed (objects/hour)=12k2026.01 | 88.7 | — | 45.2 | 43.8 | |
| Human AnnotationSpeed (objects/hour)=0.12k2026.01 | 82.4 | 2.3 | 40 | 38.5 | |
| ScoreAggSpeed (objects/hour)=9k2026.01 | 80.1 | 3.9 | 37.8 | 36 | |
| Cap3DSpeed (objects/hour)=8k2026.01 | 78.6 | 3.3 | 35.2 | 33.4 | |
| 3D-LLMSpeed (objects/hour)=6.5k2026.01 | 77.4 | 3.2 | 34.9 | 33.3 | |
| ULIP-2Speed (objects/hour)=7k2026.01 | 75.2 | 3 | 33.1 | 31.5 | |
| PointCLIPSpeed (objects/hour)=5k2026.01 | 65.3 | 2 | 22.4 | 20.8 | |
| Metadata2026.01 | 65.2 | 1.5 | 20.1 | 18.7 | |
| GPT4PointSpeed (objects/hour)=4k2026.01 | 62.9 | 1.8 | 18.7 | 17.1 |