Text Generation on ScanNet and 3D-FRONT (test)
100ROUGE-1LLaVA-3D-7B (Teacher)
Evaluation Results
| Method | Links | |
|---|---|---|
| LLaVA-3D-7B (Teacher)Params=7.0B, FPS=0.0075, Size (GB)=26.72026.05 | 100 | |
| Our Student (Distilled)Params=2.29B, FPS=0.065, Size (GB)=8.72026.05 | 11.9 | |
| PaliGemma-3BParams=3.0B, FPS=0.055, Size (GB)=11.22026.05 | 6.1 | |
| MobileVLM-2BParams=2.0B, FPS=0.072, Size (GB)=7.52026.05 | 5.2 | |
| LLaVA-v1.5-7B (2D)Params=7.0B, FPS=0.008, Size (GB)=26.72026.05 | 4.5 |