Semantic Segmentation on COCO-Stuff-10K (test)
53.46mIoUSegViT
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| SegViTBackbone=BEiTv2-Large, GFLOPs=388.2, Multi-scale inference=true, Crop size=512 x 5122023.06 | 53.46 | — | — | — | |
| SegViT (Shrunk++)Backbone=BEiTv2-Large, GFLOPs=213.3, Multi-scale inference=true, Crop size=512 x 5122023.06 | 50.54 | — | — | — | |
| SegViTBackbone=ViT-Large, GFLOPs=383.9, multi-scale inference=true, Crop Size=512 x 5122022.10 | 50.3 | — | — | — | |
| SegViTBackbone=ViT-Large, GFLOPs=383.9, Multi-scale inference=true, Crop size=512 x 5122023.06 | 50.3 | — | — | — | |
| SenFormerBackbone=Swin-Large, GFLOPs=>400, Multi-scale inference=true, Crop size=512 x 5122023.06 | 50.1 | — | — | — | |
| SenFormerbackbone=Swin-L2021.11 | 49.8 | — | 51.5 | — | |
| SegViT (Shrunk)Backbone=ViT-Large, GFLOPs=224.8, multi-scale inference=true, Crop Size=512 x 5122022.10 | 49.4 | — | — | — | |
| SegViT (Shrunk)Backbone=ViT-Large, GFLOPs=224.8, Multi-scale inference=true, Crop size=512 x 5122023.06 | 49.4 | — | — | — | |
| StructTokenBackbone=ViT-Large, GFLOPs=>400, multi-scale inference=true, Crop Size=512 x 5122022.10 | 49.1 | — | — | — | |
| StructTokenBackbone=ViT-Large, GFLOPs=>400, Multi-scale inference=true, Crop size=512 x 5122023.06 | 49.1 | — | — | — | |
| MCIBIBackbone=ViT-Large, GFLOPs=>380, multi-scale inference=true, Crop Size=512 x 5122022.10 | 44.9 | — | — | — | |
| MCIBIBackbone=ViT-Large, GFLOPs=>380, Multi-scale inference=true, Crop size=512 x 5122023.06 | 44.9 | — | — | — | |
| ProtoSegBackbone=MiT-B4, # Param (M)=64.02022.03 | 43.3 | — | — | — | |
| SegFormerBackbone=MiT-B4, # Param (M)=64.12022.03 | 42.5 | — | — | — | |
| ProtoSegBackbone=Swin-Base, # Param (M)=90.52022.03 | 42.4 | — | — | — | |
| ISNetBackbone=Dilated-ResNeSt-101, GFLOPs=228.3, multi-scale inference=true, Crop Size=512 x 5122022.10 | 42.1 | — | — | — | |
| ISNetBackbone=Dilated-ResNeSt-101, GFLOPs=228.3, Multi-scale inference=true, Crop size=512 x 5122023.06 | 42.1 | — | — | — | |
| SwinBackbone=Swin-Base, # Param (M)=90.62022.03 | 41.5 | — | — | — | |
| RecoNetBackbone=Dilated-ResNet-101, GFLOPs=>200, multi-scale inference=true, Crop Size=512 x 5122022.10 | 41.5 | — | — | — | |
| RecoNetBackbone=Dilated-ResNet-101, GFLOPs=>200, Multi-scale inference=true, Crop size=512 x 5122023.06 | 41.5 | — | — | — | |
| SenFormerbackbone=R1012021.11 | 41 | — | 42.1 | — | |
| GINetBackbone=JPU-ResNet-101, GFLOPs=>200, multi-scale inference=true, Crop Size=512 x 5122022.10 | 40.6 | — | — | — | |
| GINetBackbone=JPU-ResNet-101, GFLOPs=>200, Multi-scale inference=true, Crop size=512 x 5122023.06 | 40.6 | — | — | — | |
| OCRBackbone=HRNetV2-W48, # Param (M)=70.32022.03 | 40.5 | — | — | — | |
| OCRNetBackbone=HRNetV2-W48, GFLOPs=167.9, multi-scale inference=true, Crop Size=512 x 5122022.10 | 40.5 | — | — | — | |
| OCRNetBackbone=HRNet V2-W48, GFLOPs=167.9, Multi-scale inference=true, Crop size=512 x 5122023.06 | 40.5 | — | — | — | |
| ACNetBackbone=ResNet-1012022.03 | 40.1 | — | — | — | |
| SenFormerbackbone=R502021.11 | 40 | — | 41.3 | — | |
| SpyGRBackbone=ResNet-1012022.03 | 39.9 | — | — | — | |
| ProtoSegBackbone=HRNetV2-W48, # Param (M)=65.82022.03 | 39.9 | — | — | — | |
| EMANetBackbone=Dilated-ResNet-101, GFLOPs=247.4, multi-scale inference=true, Crop Size=512 x 5122022.10 | 39.9 | — | — | — | |
| SpyGRBackbone=ResNet-101-fpn, GFLOPs=>80, multi-scale inference=true, Crop Size=512 x 5122022.10 | 39.9 | — | — | — | |
| EMANetBackbone=Dilated-ResNet-101, GFLOPs=247.4, Multi-scale inference=true, Crop size=512 x 5122023.06 | 39.9 | — | — | — | |
| SpyGRBackbone=ResNet-101-fpn, GFLOPs=>80, Multi-scale inference=true, Crop size=512 x 5122023.06 | 39.9 | — | — | — | |
| MaskFormerBackbone=ResNet-101, # Param (M)=60.02022.03 | 39.8 | — | — | — | |
| MaskFormerBackbone=ResNet-101-fpn, GFLOPs=81.7, multi-scale inference=true, Crop Size=512 x 5122022.10 | 39.8 | — | — | — | |
| MaskFormerBackbone=ResNet-101-fpn, GFLOPs=81.7, Multi-scale inference=true, Crop size=512 x 5122023.06 | 39.8 | — | — | — | |
| DANetBackbone=ResNet-101, # Param (M)=69.12022.03 | 39.7 | — | — | — | |
| DANetBackbone=Dilated-ResNet-101, GFLOPs=289.3, multi-scale inference=true, Crop Size=512 x 5122022.10 | 39.7 | — | — | — | |
| DANetBackbone=Dilated-ResNet-101, GFLOPs=289.3, Multi-scale inference=true, Crop size=512 x 5122023.06 | 39.7 | — | — | — | |
| SVCNetBackbone=ResNet-1012022.03 | 39.6 | — | — | — | |
| HRNetBackbone=HRNetV2-W48, # Param (M)=65.92022.03 | 38.7 | — | — | — | |
| MaskFormerbackbone=R1012021.11 | 38.1 | — | 39.8 | — | |
| MaskFormerbackbone=R502021.11 | 37.1 | — | 38.9 | — | |
| PerPixelBaseline+backbone=R502021.11 | 34.2 | — | 35.8 | — | |
| ProtoSegBackbone=ResNet-101, # Param (M)=68.52022.03 | 34 | — | — | — | |
| FCNBackbone=ResNet-101, # Param (M)=68.62022.03 | 32.5 | — | — | — | |
| CAAbackbone=R1012021.11 | — | — | 41.2 | — | |
| CAAbackbone=EN-B72021.11 | — | — | 45.4 | — | |
| CAABackbone=EfficientNet-B7-D8, Auxiliary loss=true, Optimizer=SGD, Note=Original SOTA score reported2022.03 | — | — | — | 45.4 | |
| CAABackbone=Swin-Large + JPU, Auxiliary loss=false, Optimizer=SGD2022.03 | — | 44.22 | — | 45.31 | |
| CAABackbone=ConvNeXt-Large + JPU, Auxiliary loss=false, Optimizer=SGD2022.03 | — | 46.49 | — | 47.23 | |
| CAA + CARBackbone=Swin-Large + JPU, Auxiliary loss=false, Optimizer=SGD2022.03 | — | 45.48 | — | 46.99 | |
| CAA + CARBackbone=ConvNeXt-Large + JPU, Auxiliary loss=false, Optimizer=SGD2022.03 | — | 46.7 | — | 47.77 | |
| CAA + CARBackbone=ConvNeXt-Large + JPU, Auxiliary loss=false, Optimizer=Adam2022.03 | — | 48.2 | — | 48.83 | |
| CAA + CARBackbone=ConvNeXt-Large + JPU, Auxiliary loss=true, Optimizer=Adam2022.03 | — | 49.03 | — | 50.01 | |
| CAREncoder=ConvNext-L2022.12 | — | 49.03 | 50.01 | — | |
| DANetbackbone=R1012021.11 | — | — | 39.7 | — | |
| EMANetbackbone=R502021.11 | — | — | 37.6 | — | |
| EMANetbackbone=R1012021.11 | — | — | 39.9 | — | |
| InternImage-HMulti-scale testing=true, Crop size=512x512, Iterations=40k, Pre-training=COCO-Stuff-164K2022.11 | — | — | 59.6 | — | |
| MaskFormerbackbone=R502021.07 | — | 37.1 | 38.9 | — | |
| MaskFormerbackbone=R1012021.07 | — | 38.1 | 39.8 | — | |
| MaskFormerbackbone=R101c2021.07 | — | 38 | 39.3 | — | |
| OCRNetbackbone=R101c2021.07 | — | — | 39.5 | — | |
| OCRNetbackbone=R1012021.11 | — | — | 39.5 | — | |
| OCRNetbackbone=HRNet2021.11 | — | — | 40.5 | — | |
| PerPixelBaselinebackbone=R502021.07 | — | 32.4 | 34.4 | — | |
| PerPixelBaseline+backbone=R502021.07 | — | 34.2 | 35.8 | — | |
| RSSeg-ViTEncoder=ViT-L, Pre-train weights=ViT-AugReg [26]2022.12 | — | 50.42 | 51.99 | — | |
| RSSeg-ViTEncoder=BEIT-L2022.12 | — | 51.91 | 52.63 | — | |
| SegmenterEncoder=ViT-L, Pre-train weights=ViT-AugReg [26], Reported in=[80]2022.12 | — | 45.5 | 47.1 | — | |
| Segmenter+RankSegEncoder=ViT-L, Pre-train weights=ViT-AugReg [26]2022.12 | — | 46.6 | 47.9 | — | |
| SegViTEncoder=ViT-L, Pre-train weights=ViT-AugReg [26]2022.12 | — | — | 50.3 | — | |
| SeMaskEncoder=Swin-L2022.12 | — | 47.47 | 48.54 | — | |
| SenFormerEncoder=Swin-L2022.12 | — | 49.8 | 51.5 | — | |
| StructToken-SSEEncoder=ViT-L, Pre-train weights=ViT-AugReg [26]2022.12 | — | — | 49.1 | — | |
| UperNetBackbone=Swin-Large, Auxiliary loss=false, Optimizer=SGD2022.03 | — | 44.25 | — | 46.1 | |
| UperNetEncoder=BEIT-L, Reported in=[80]2022.12 | — | 49.7 | 49.9 | — | |
| UperNet + CARBackbone=Swin-Large, Auxiliary loss=false, Optimizer=SGD2022.03 | — | 44.88 | — | 46.64 | |
| UperNet+RankSegEncoder=BEIT-L2022.12 | — | 49.9 | 50.3 | — | |
| ViT-AdapterMulti-scale testing=true2022.11 | — | — | 54.2 | — | |
| ViT-Adapter-UperNetEncoder=BEIT-L2022.12 | — | 51 | 51.4 | — |