Open-Vocabulary Semantic Segmentation on ORAD-3D
28.89mIoUReSiReg Full
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ReSiReg FullBackbone=ViT-L, Base Model=RADIOv4 (SigLIP 2 Head), Param=475M, Method Variant=Full2026.06 | 28.89 | 9.78 | |
| ReSiReg LiteBackbone=ViT-L, Base Model=RADIOv4 (SigLIP 2 Head), Param=475M, Method Variant=Lite2026.06 | 28.8 | 9.69 | |
| ReSiReg LiteBackbone=ViT-S, Base Model=EUPE (Our Head), Param=25M, Method Variant=Lite2026.06 | 25.93 | 6.08 | |
| ReSiReg FullBackbone=ViT-B, Base Model=RadSeg, Param=113M, Method Variant=Full2026.06 | 25 | 2.83 | |
| ReSiReg FullBackbone=ViT-S, Base Model=EUPE (Our Head), Param=25M, Method Variant=Full2026.06 | 24.45 | 4.6 | |
| ReSiReg LiteBackbone=ViT-B, Base Model=RadSeg, Param=113M, Method Variant=Lite2026.06 | 24.21 | 2.04 | |
| Self-CalibrationBackbone=ViT-B, Base Model=RADIOv3 (CLIP Head), Param=113M, Method Variant=Calib2026.06 | 23.3 | 6.11 | |
| Self-CalibrationBackbone=ViT-B, Base Model=RADIOv3 (SigLIP 2 Head), Param=113M, Method Variant=Calib2026.06 | 22.93 | 6.13 | |
| ReSiReg LiteBackbone=ViT-B, Base Model=RADIOv3 (CLIP Head), Param=113M, Method Variant=Lite2026.06 | 22.55 | 5.36 | |
| RadSegBackbone=ViT-B, Param=113M, Method Variant=BB2026.06 | 22.17 | — | |
| ReSiReg FullBackbone=ViT-B, Base Model=RADIOv3 (CLIP Head), Param=113M, Method Variant=Full2026.06 | 22.05 | 4.68 | |
| ReSiReg FullBackbone=ViT-B, Base Model=RADIOv3 (SigLIP 2 Head), Param=113M, Method Variant=Full2026.06 | 21.32 | 4.52 | |
| EUPE (Our Head)Backbone=ViT-S, Param=25M, Method Variant=BB2026.06 | 19.85 | — | |
| RADIOv4 (SigLIP 2 Head)Backbone=ViT-L, Param=475M, Method Variant=BB2026.06 | 19.11 | — | |
| ReSiReg LiteBackbone=ViT-B, Base Model=RADIOv3 (SigLIP 2 Head), Param=113M, Method Variant=Lite2026.06 | 18.31 | 1.51 | |
| RADIOv3 (CLIP Head)Backbone=ViT-B, Param=113M, Method Variant=BB2026.06 | 17.19 | — | |
| RADIOv3 (SigLIP 2 Head)Backbone=ViT-B, Param=113M, Method Variant=BB2026.06 | 16.8 | — | |
| Self-CalibrationBackbone=ViT-B, Base Model=RadSeg, Param=113M, Method Variant=Calib2026.06 | 16.02 | -6.15 | |
| TIPSv2Backbone=ViT-B, Param=86M, Method Variant=BB2026.06 | 15.3 | — | |
| Self-CalibrationBackbone=ViT-L, Base Model=dino.txt*, Param=330M, Method Variant=Calib2026.06 | 14.01 | 1.64 | |
| TIPSv2Backbone=ViT-L, Param=304M, Method Variant=BB2026.06 | 13.67 | — | |
| OTAS*Backbone=ViT-B, Param=103M, Method Variant=BB, Input Resolution=Lower2026.06 | 12.53 | — | |
| ReSiReg FullBackbone=ViT-L, Base Model=dino.txt*, Param=330M, Method Variant=Full2026.06 | 12.52 | 0.15 | |
| ReSiReg LiteBackbone=ViT-L, Base Model=dino.txt*, Param=330M, Method Variant=Lite2026.06 | 12.44 | 0.07 | |
| dino.txt*Backbone=ViT-L, Param=330M, Method Variant=BB, Input Resolution=Lower2026.06 | 12.37 | — | |
| ReSiReg FullBackbone=ViT-B, Base Model=CLIP*, Param=86M, Method Variant=Full2026.06 | 10.78 | 2.61 | |
| Self-CalibrationBackbone=ViT-B, Base Model=CLIP*, Param=86M, Method Variant=Calib2026.06 | 10.56 | 2.39 | |
| ReSiReg LiteBackbone=ViT-B, Base Model=CLIP*, Param=86M, Method Variant=Lite2026.06 | 9.03 | 0.86 | |
| CLIP*Backbone=ViT-B, Param=86M, Method Variant=BB, Input Resolution=Lower2026.06 | 8.17 | — | |
| Self-CalibrationBackbone=ViT-L, Base Model=CLIP*, Param=428M, Method Variant=Calib2026.06 | 7.8 | 4.17 | |
| ReSiReg FullBackbone=ViT-L, Base Model=CLIP*, Param=428M, Method Variant=Full2026.06 | 7.41 | 3.78 | |
| Self-CalibrationBackbone=ViT-S, Base Model=EUPE (Our Head), Param=25M, Method Variant=Calib2026.06 | 4.87 | -14.98 | |
| Self-CalibrationBackbone=ViT-L, Base Model=RADIOv4 (SigLIP 2 Head), Param=475M, Method Variant=Calib2026.06 | 4.79 | -14.32 | |
| ReSiReg LiteBackbone=ViT-L, Base Model=CLIP*, Param=428M, Method Variant=Lite2026.06 | 4.39 | 0.76 | |
| CLIP*Backbone=ViT-L, Param=428M, Method Variant=BB, Input Resolution=Lower2026.06 | 3.63 | — |