Localization on RefCOCO+
67AccuracyVanilla
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VanillaVisual Token Budget=576, FLOPs (T)=4.292026.02 | 67 | 100 | |
| SwiftVLMVisual Token Budget=192, FLOPs (T)=1.752026.02 | 58.5 | 87.3 | |
| SwiftVLMVisual Token Budget=128, FLOPs (T)=1.312026.02 | 46.6 | 69.6 | |
| FEATHERVisual Token Budget=192, FLOPs (T)=1.822026.02 | 45.5 | 67.9 | |
| FEATHERVisual Token Budget=128, FLOPs (T)=1.442026.02 | 34.3 | 51.2 | |
| FastVVisual Token Budget=192, FLOPs (T)=1.712026.02 | 25.8 | 38.5 | |
| PDropVisual Token Budget=192, FLOPs (T)=1.722026.02 | 18.2 | 27.2 | |
| FastVVisual Token Budget=128, FLOPs (T)=1.292026.02 | 11.1 | 16.6 | |
| SparseVLMVisual Token Budget=192, FLOPs (T)=1.722026.02 | 7.5 | 11.2 | |
| VisionZipVisual Token Budget=192, FLOPs (T)=1.712026.02 | 5.7 | 8.5 | |
| SparseVLMVisual Token Budget=128, FLOPs (T)=1.32026.02 | 3.9 | 5.8 | |
| VisionZipVisual Token Budget=128, FLOPs (T)=1.292026.02 | 3.6 | 5.4 | |
| PDropVisual Token Budget=128, FLOPs (T)=1.282026.02 | 2.3 | 3.4 |