Open-Vocabulary Segmentation on LoveDA
50mIoUOracle
Evaluation Results
| Method | Links | |
|---|---|---|
| OracleEvaluation Protocol=Fully supervised2026.02 | 50 | |
| CONCEPTBANKPub. & Year=-, Backbone (Size)=SAM3 (PE-L+/14), Optimization Protocol=Parameter-free, Evaluation Setting=with background2026.02 | 49.4 | |
| ProC-SAM3Backbone=SAM 32026.06 | 46.1 | |
| SegEarth-OV3*Backbone=SAM 32026.06 | 41.6 | |
| ActiveSAM*Backbone=SAM32026.06 | 41.1 | |
| ConInferBackbone=CLIP+DINOv32026.06 | 39.3 | |
| SkySense-OPub. & Year=CVPR’25, Backbone (Size)=CLIP (ViT-L/14@512px), Optimization Protocol=Training-based, Evaluation Setting=with background2026.02 | 38.3 | |
| OursEvaluation Protocol=Zero-shot2026.02 | 38.2 | |
| ReAttnCLIPBackbone=CLIP2026.06 | 37 | |
| SegEarth-OV*Pub. & Year=CVPR’25, Backbone (Size)=CLIP, Optimization Protocol=Parameter-free, Evaluation Setting=with background2026.02 | 36.9 | |
| CorrCLIP*Pub. & Year=ICCV’25, Backbone (Size)=CLIP + DINO (ViT-B/8) + SAM2 (Hiera-L), Optimization Protocol=Parameter-free, Evaluation Setting=with background2026.02 | 36.9 | |
| SegEarth-OVEvaluation Protocol=Trained on remote sensing data2026.02 | 36.9 | |
| SegEarth-OVBackbone=CLIP2026.06 | 36.9 | |
| CorrCLIPBackbone=CLIP+DINO+SAM22026.06 | 36.9 | |
| SAM3*Pub. & Year=ICLR’26, Backbone (Size)=SAM3 (PE-L+/14), Optimization Protocol=Parameter-free, Evaluation Setting=with background2026.02 | 35.6 | |
| RSCLIP*Backbone=CLIP2026.06 | 35.5 | |
| ProxyCLIPPub. & Year=ECCV’24, Backbone (Size)=CLIP + DINOv2 (ViT-B/14), Optimization Protocol=Parameter-free, Evaluation Setting=with background2026.02 | 34.3 | |
| ProxyCLIPBackbone=CLIP+DINOv22026.06 | 34.3 | |
| RSKT-SegPub. & Year=AAAI’26, Backbone (Size)=CLIP (ViT-L/14@336px) + RemoteCLIP (ViT-B/16) + DINO (ViT-B/32), Optimization Protocol=Training-based, Evaluation Setting=with background2026.02 | 33.2 | |
| ClearCLIPEvaluation Protocol=Zero-shot2026.02 | 32.4 | |
| ClearCLIPBackbone=CLIP2026.06 | 32.4 | |
| GEMPub. & Year=CVPR’24, Backbone (Size)=CLIP, Optimization Protocol=Parameter-free, Evaluation Setting=with background2026.02 | 31.6 | |
| GEMEvaluation Protocol=Zero-shot2026.02 | 31.6 | |
| VIP*Backbone=DINOv3 + dino.txt2026.06 | 31.2 | |
| SCLIP*Pub. & Year=ECCV’24, Backbone (Size)=CLIP, Optimization Protocol=Parameter-free, Evaluation Setting=with background2026.02 | 30.4 | |
| SCLIPEvaluation Protocol=Zero-shot2026.02 | 30.4 | |
| SCLIPBackbone=CLIP2026.06 | 30.4 | |
| Cat-SegPub. & Year=CVPR’24, Backbone (Size)=CLIP (ViT-L/14@336px), Optimization Protocol=Training-based, Evaluation Setting=with background2026.02 | 28.6 | |
| MaskCLIPPub. & Year=ECCV’22, Backbone (Size)=CLIP, Optimization Protocol=Parameter-free, Evaluation Setting=with background2026.02 | 27.8 | |
| MaskCLIPEvaluation Protocol=Zero-shot2026.02 | 27.8 | |
| SANPub. & Year=CVPR’23, Backbone (Size)=CLIP (ViT-L/14@336px), Optimization Protocol=Training-based, Evaluation Setting=with background2026.02 | 25.3 | |
| CLIPPub. & Year=ICML’21, Backbone (Size)=CLIP, Optimization Protocol=Parameter-free, Evaluation Setting=with background2026.02 | 12.4 | |
| CLIPEvaluation Protocol=Zero-shot2026.02 | 12.4 |