Open Vocabulary Semantic Segmentation on Pascal VOC 20
98.3mIoUESC-Net
Evaluation results
| Method | Links | |
|---|---|---|
| ESC-NetVLM=CLIP ViT-L/14, Training Dataset=COCO-Stuff, Additional Dataset=false2024.11 | 98.3 | |
| FG-CLIP 2Backbone=ViT-So/162025.10 | 97.8 | |
| MROVSegVLM=CLIP ViT-L/14, Training Dataset=COCO-Stuff2024.12 | 97.6 | |
| FG-CLIP 2Backbone=ViT-L/162025.10 | 97.6 | |
| FG-CLIPBackbone=ViT-L/142025.10 | 97.4 | |
| ESC-NetVLM=CLIP ViT-B/16, Training Dataset=COCO-Stuff, Additional Dataset=false2024.11 | 97.3 | |
| EBSegVLM=CLIP ViT-L/14, Additional Backbone=SAM ViT-B, Training Dataset=COCO-Stuff, Additional Dataset=false2024.11 | 97.2 | |
| FG-CLIP 2Backbone=ViT-B/162025.10 | 97.1 | |
| CAT-SegVLM=CLIP ViT-L/14, Training Dataset=COCO-Stuff2024.12 | 97 | |
| SCANVLM=CLIP ViT-B/16, Additional Backbone=Swin-B, Training Dataset=COCO-Stuff, Additional Dataset=false2024.11 | 97 | |
| SCANVLM=CLIP ViT-L/14, Additional Backbone=Swin-B, Training Dataset=COCO-Stuff, Additional Dataset=false2024.11 | 97 | |
| CAT-SegVLM=CLIP ViT-L/14, Training Dataset=COCO-Stuff, Additional Dataset=false2024.11 | 97 | |
| SigLIP 2Backbone=ViT-L/162025.10 | 97 | |
| CLIPBackbone=ViT-L/142025.10 | 96.6 | |
| MAFT+VLM=ConvNeXt-L, Training Dataset=COCO-Stuff, Additional Dataset=false2024.11 | 96.5 | |
| EBSegVLM=CLIP ViT-L/14, Training Dataset=COCO-Stuff2024.12 | 96.4 | |
| MAFTPVLM=CLIP ConvNeXt-L, Training Dataset=COCO-Stuff2024.12 | 96.4 | |
| SEDVLM=CLIP ConvNeXt-L, Training Dataset=COCO-Stuff2024.12 | 96.1 | |
| SEDVLM=ConvNeXt-L, Training Dataset=COCO-Stuff, Additional Dataset=false2024.11 | 96.1 | |
| GBAVLM=CLIP ConvNeXt-L, Training Dataset=COCO Panoptic2024.12 | 95.8 | |
| MAFTP w/ MaskAdapterVLM=CLIP ConvNeXt-L, Training Dataset=COCO-Stuff2024.12 | 95.8 | |
| MAFTPVLM=CLIP ConvNeXt-B, Training Dataset=COCO-Stuff2024.12 | 95.5 | |
| FC-CLIP w/ MaskAdapterVLM=CLIP ConvNeXt-L, Training Dataset=COCO Panoptic2024.12 | 95.5 | |
| FC-CLIPVLM=CLIP ConvNeXt-L, Training Dataset=COCO Panoptic2024.12 | 95.4 | |
| FG-CLIPBackbone=ViT-B/162025.10 | 95.3 | |
| CAT-SegPrompt Tuning=MMRL+VaMP2025.11 | 95.2 | |
| MAFTP w/ MaskAdapterVLM=CLIP ConvNeXt-B, Training Dataset=COCO-Stuff2024.12 | 95.1 | |
| EBSegVLM=CLIP ViT-B/16, Training Dataset=COCO-Stuff2024.12 | 94.6 | |
| CAT-SegVLM=CLIP ViT-B/16, Training Dataset=COCO-Stuff2024.12 | 94.6 | |
| SANVLM=CLIP ViT-L/14, Training Dataset=COCO-Stuff2024.12 | 94.6 | |
| EBSegVLM=CLIP ViT-B/16, Additional Backbone=SAM ViT-B, Training Dataset=COCO-Stuff, Additional Dataset=false2024.11 | 94.6 | |
| CAT-SegVLM=CLIP ViT-B/16, Training Dataset=COCO-Stuff, Additional Dataset=false2024.11 | 94.6 | |
| SANVLM=CLIP ViT-L/14, Training Dataset=COCO-Stuff, Additional Dataset=false2024.11 | 94.6 | |
| CAT-SegPrompt Tuning=-2025.11 | 94.6 | |
| OvSegVLM=CLIP ViT-L/14, Training Dataset=COCO-Stuff2024.12 | 94.5 | |
| OVSegVLM=CLIP ViT-L/14, Additional Backbone=Swin-B, Training Dataset=COCO-Stuff, Additional Dataset=true2024.11 | 94.5 | |
| SEDVLM=CLIP ConvNeXt-B, Training Dataset=COCO-Stuff2024.12 | 94.4 | |
| SEDVLM=ConvNeXt-B, Training Dataset=COCO-Stuff, Additional Dataset=false2024.11 | 94.4 | |
| SigLIP 2Backbone=ViT-B/162025.10 | 94.4 | |
| CAT-SegPrompt Tuning=MMRL2025.11 | 94.3 | |
| SANVLM=CLIP ViT-B/16, Training Dataset=COCO-Stuff2024.12 | 94 | |
| SANVLM=CLIP ViT-B/16, Training Dataset=COCO-Stuff, Additional Dataset=false2024.11 | 94 | |
| SANPrompt Tuning=-2025.11 | 94 | |
| CLIPBackbone=ViT-B/162025.10 | 93.7 | |
| ZegCLIPVLM=CLIP ViT-B/16, Training Dataset=COCO-Stuff-156, Additional Dataset=false2024.11 | 93.6 | |
| OvSegVLM=CLIP ViT-B/16, Training Dataset=COCO-Stuff2024.12 | 92.6 | |
| OVSegVLM=CLIP ViT-B/16, Additional Backbone=ResNet-101c, Training Dataset=COCO-Stuff, Additional Dataset=true2024.11 | 92.6 | |
| OVSegPrompt Tuning=-2025.11 | 92.6 | |
| MAFTVLM=CLIP ViT-L/14, Training Dataset=COCO-Stuff, Additional Dataset=false2024.11 | 92.1 | |
| DeOPVLM=CLIP ViT-B/16, Training Dataset=COCO-Stuff2024.12 | 91.7 | |
| DeOPVLM=CLIP ViT-B/16, Additional Backbone=ResNet-101c, Training Dataset=COCO-Stuff-156, Additional Dataset=false2024.11 | 91.7 | |
| SPARBackbone=DINOv3 - ViT-L-16, Inference=single-pass, Upsampler=AnyUp [45]2026.04 | 91.5 | |
| SPARBackbone=DINOv3 - ViT-L-16, Inference=single-pass, Upsampler=bilinear2026.04 | 91.3 | |
| Pre-trainedBackbone=DINOv3 - ViT-L-16, Inference=sliding-window, Upsampler=bilinear2026.04 | 90.8 | |
| Pre-trainedBackbone=DINOv3 - ViT-L-16, Inference=single-pass, Upsampler=AnyUp [45]2026.04 | 89.9 | |
| LPOSS + SPARBackbone=SigLIP2 - ViT-B-16, Inference=single-pass, Upsampler=bilinear2026.04 | 89.7 | |
| LPOSS + Pre-tr.Backbone=SigLIP2 - ViT-B-16, Inference=single-pass, Upsampler=bilinear2026.04 | 89.6 | |
| ZegFormerVLM=CLIP ViT-B/16, Training Dataset=COCO-Stuff2024.12 | 89.5 | |
| FC-CLIPVLM=ConvNeXt-L, Training Dataset=COCO Panoptic, Additional Dataset=false2024.11 | 89.5 | |
| ZegFormerPrompt Tuning=-2025.11 | 89.5 | |
| Pre-trainedBackbone=DINOv3 - ViT-L-16, Inference=single-pass, Upsampler=bilinear2026.04 | 89.5 | |
| ZSsegVLM=CLIP ViT-B/16, Additional Backbone=ResNet-101, Training Dataset=COCO-Stuff, Additional Dataset=false2024.11 | 88.4 | |
| CASS*Pub. & Year=CVPR’25, Extra data=✗, Extra backbone=DINOv3 (ViT-B/16), Training free=✓, Mask Refinement=true2026.03 | 87.1 | |
| Talk2DINOBackbone=DINOv2 B/142026.04 | 87.1 | |
| PEARLExtra data=✗, Extra backbone=✗, Training free=✓, Mask Refinement=false2026.03 | 86.9 | |
| CASS*Pub. & Year=CVPR’25, Extra data=✗, Extra backbone=DINOv2† (ViT-B/14), Training free=✓, Mask Refinement=true2026.03 | 86.4 | |
| ZegFormerVLM=CLIP ViT-B/16, Additional Backbone=ResNet-101, Training Dataset=COCO-Stuff-156, Additional Dataset=false2024.11 | 86.2 | |
| SPARBackbone=SigLIP2 - ViT-B-16, Inference=single-pass, Upsampler=AnyUp [45]2026.04 | 86.2 | |
| PixelCLIP-ViTBackbone=CLIP B/162026.04 | 85.9 | |
| SFP*Pub. & Year=ICCV’25, Extra data=✗, Extra backbone=✗, Training free=✓, Mask Refinement=true2026.03 | 85.1 | |
| DSGSBackbone=CLIP B/162026.04 | 85 | |
| SCLIP*Pub. & Year=ECCV’24, Extra data=✗, Extra backbone=✗, Training free=✓, Mask Refinement=true2026.03 | 83.5 | |
| OVS-DINOBackbone=DINOv2 B/142026.04 | 83.5 | |
| TCLPub. & Year=CVPR’23, Extra data=CC3M+CC12M, Extra backbone=✗, Training free=✗, Mask Refinement=true2026.03 | 83.2 | |
| ProxyCLIP*Pub. & Year=ECCV’24, Extra data=✗, Extra backbone=DINOv2† (ViT-B/14), Training free=✓, Mask Refinement=true2026.03 | 83.1 | |
| NACLIP*Pub. & Year=WACV’25, Extra data=✗, Extra backbone=✗, Training free=✓, Mask Refinement=true2026.03 | 83 | |
| LaVGPub. & Year=ECCV’24, Extra data=✗, Extra backbone=DINOv1 (ViT-B/8), Training free=✓, Mask Refinement=true2026.03 | 82.5 | |
| Pre-trainedBackbone=SigLIP2 - ViT-B-16, Inference=single-pass, Upsampler=AnyUp [45]2026.04 | 82.3 | |
| OVDiffParadigm=Build prototypes per class, Extra Backbones=DINO & SD, Background Prompt=No2026.01 | 81.7 | |
| GroupViTPub. & Year=CVPR’22, Extra data=CC12M+RedCaps, Extra backbone=✗, Training free=✗, Mask Refinement=true2026.03 | 81.5 | |
| CLIP-DINOiserPub. & Year=ECCV’24, Extra data=ImageNet-1K, Extra backbone=DINOv1 (ViT-B/16), Training free=✗, Mask Refinement=true2026.03 | 81.5 | |
| SPARBackbone=SigLIP2 - ViT-B-16, Inference=single-pass, Upsampler=bilinear2026.04 | 81.5 | |
| CLIP-DINOiserBackbone=CLIP B/162026.04 | 80.9 | |
| CLIP-DINOiserParadigm=Use Frozen CLIP, Background Prompt=No, Protocol=Rerun (*)2026.01 | 80.8 | |
| INSIGHT_thresParadigm=Use Frozen CLIP with Interpretability, Background Prompt=No2026.01 | 80.7 | |
| GroupViTParadigm=Text/image alignment training with captions, Background Prompt=No2026.01 | 79.7 | |
| CLIP-DIYParadigm=Use Frozen CLIP, Extra Backbones=DINO, Background Prompt=No2026.01 | 79.7 | |
| GroupViTBackbone=CLIP S/162026.04 | 79.7 | |
| FreeDAPub. & Year=CVPR’24, Extra data=COCO Captions, Extra backbone=DINOv2 (ViT-B/14), Training free=✓, Mask Refinement=true2026.03 | 79.5 | |
| TCLParadigm=Text/image alignment training with captions, Background Prompt=No2026.01 | 77.5 | |
| Pre-trainedBackbone=SigLIP2 - ViT-B-16, Inference=sliding-window, Upsampler=bilinear2026.04 | 77.5 | |
| TCLBackbone=CLIP B/162026.04 | 77.5 | |
| INSIGHTParadigm=Use Frozen CLIP with Interpretability, Background Prompt=No2026.01 | 77 | |
| OpenSegVLM=ALIGN, Training Dataset=COCO Panoptic+Loc. Narr.2024.12 | 72.2 | |
| MaskCLIPPub. & Year=ECCV’22, Extra data=✗, Extra backbone=✗, Training free=✓, Mask Refinement=true2026.03 | 72.1 | |
| MaskCLIPParadigm=Use Frozen CLIP, Extra Backbones=ref[9], Background Prompt=No2026.01 | 71.9 | |
| Pre-trainedBackbone=SigLIP2 - ViT-B-16, Inference=single-pass, Upsampler=bilinear2026.04 | 71.3 | |
| NaFlexBackbone=SigLIP2 - ViT-B-16, Inference=single-pass, Upsampler=bilinear2026.04 | 66.6 | |
| CoDeBackbone=CLIP B/162026.04 | 64.9 | |
| ReCoPub. & Year=NeurIPS’22, Extra data=ImageNet-1K, Extra backbone=✗, Training free=✓, Mask Refinement=true2026.03 | 62.4 |