ResearchBenchmarks3D Object Description on ScanNetFollow80.3CLIP ScoreTri-MARF57.83663.66869.575.332Jan 7, 2026Evaluation ResultsMethodMethodLinksCLIP ScoreViLT R@5 (I2T)ViLT R@5 (T2I)GPT-4 ScoreTri-MARF2026.0180.335.233.74Human Annotation2026.0179.534.833.24.3ScoreAgg2026.0175.632.130.43.5Cap3D2026.0173.229.828.13.23D-LLM2026.0172.529.127.63.1ULIP-22026.0170.627.826.32.9PointCLIP2026.0160.819.317.92.1GPT4Point2026.0158.717.4161.9