Semantic Segmentation on Pascal VOC (test)
91.9mIoUFreeSeg
Evaluation results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| FreeSegProtocol=Zero-shot2023.09 | 91.9 | — | — | — | — | — | — | — | — | — | 78.6 | 84.7 | — | — | |
| DeOPBackbone=R101c, Training dataset=COCO-Stuff-156, Decoupled network=true, Number of Passes=12023.04 | 91.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ZegFormer + MAFTProtocol=Zero-shot, Method Component=Mask-aware Fine-tuning2023.09 | 91.5 | — | — | — | — | — | — | — | — | — | 85.7 | 88.4 | — | — | |
| FreeSeg + MAFTProtocol=Zero-shot, Method Component=Mask-aware Fine-tuning2023.09 | 91.4 | — | — | — | — | — | — | — | — | — | 81.8 | 86.3 | — | — | |
| ZegFormerProtocol=Zero-shot2023.09 | 90.1 | — | — | — | — | — | — | — | — | — | 70.6 | 79.2 | — | — | |
| SimBaseBackbone=R101c, Training dataset=COCO-Stuff-156, Decoupled network=true, Number of Passes=N'2023.04 | 88.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ZSSeg + MAFTProtocol=Zero-shot, Method Component=Mask-aware Fine-tuning2023.09 | 88.4 | — | — | — | — | — | — | — | — | — | 66.2 | 75.7 | — | — | |
| EMANetBackbone=ResNet-1522019.07 | 88.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DUpsamplingBackbone=Xception-65, Multi-scale testing=true, Left-right flipping=true, Output stride=162019.03 | 88.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MSCIBackbone=ResNet-1522019.07 | 88 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ExfuseBackbone=ResNeXt-1312019.07 | 87.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeepLabv3+Backbone=Xception-65, Output stride=82019.03 | 87.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeepLabv3+Additional Training Dataset Used=JFT-300M [22]2019.12 | 87.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeeplabV3+Backbone=Xception-712019.07 | 87.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EMANetBackbone=ResNet-1012019.07 | 87.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DIS2019.03 | 86.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DISBackbone=ResNet-1012019.07 | 86.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CASIA_IVA_SDN2019.03 | 86.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ZSSegProtocol=Zero-shot2023.09 | 86.6 | — | — | — | — | — | — | — | — | — | 59.7 | 70.7 | — | — | |
| SDNBackbone=DenseNet-1612019.07 | 86.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IDW-CNN2019.03 | 86.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| IDW-CNNBackbone=ResNet-1012019.07 | 86.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DFN2019.03 | 86.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DFNBackbone=ResNet101, COCO Pre-training=true2019.09 | 86.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DFNCOCO Pre-training=true, Backbone=ResNet1012019.09 | 86.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DFNBackbone=ResNet-1012019.07 | 86.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ExfuseBackbone=ResNet-1012019.07 | 86.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SANetBackbone=ResNet101, COCO Pre-training=true2019.09 | 86.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SANetCOCO Pre-training=true, Backbone=ResNet1012019.09 | 86.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Mask2FormerBackbone=Swin-L, Supervision=M2024.03 | 86.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EncNet2019.03 | 85.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EncNetBackbone=ResNet101, COCO Pre-training=true2019.09 | 85.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EncNetCOCO Pre-training=true, Backbone=ResNet1012019.09 | 85.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| HamNetPre-training=w/o COCO, Backbone=ResNet-101, Output stride=82021.09 | 85.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EncNetBackbone=ResNet-1012019.07 | 85.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ZeroSegTraining Dataset=ImageNet-1K [61], Backbone=ViT-B/16, Additional Labels=-, VFM=CLIP, SAM2024.09 | 85.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeepLabv32019.03 | 85.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Deeplabv3Additional Training Dataset Used=JFT-300M [22]2019.12 | 85.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeeplabV3Backbone=ResNet101, COCO Pre-training=true2019.09 | 85.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeeplabV3COCO Pre-training=true, Backbone=ResNet1012019.09 | 85.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeeplabV3Backbone=ResNet-1012019.07 | 85.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PSANetBackbone=ResNet-1012019.07 | 85.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Auto-DeepLab-LAdditional Training Dataset Used=JFT-300M [22]2019.12 | 85.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PSPNet2019.03 | 85.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PSPNetBackbone=ResNet101, COCO Pre-training=true2019.09 | 85.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PSPNetCOCO Pre-training=true, Backbone=ResNet1012019.09 | 85.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PSPNetBackbone=ResNet-1012019.07 | 85.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DUpsamplingBackbone=Xception-65, COCO pre-training=false, Multi-scale testing=true, Left-right flipping=true2019.03 | 85.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| JoDiffusionBackbone=Swin-S, Data Size=40k2025.12 | 85.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Wide ResNetBackbone=WideResNet-382019.07 | 84.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DMNet*Pre-training=w/o COCO, Backbone=ResNet-101, Output stride=82021.09 | 84.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| OCR*Pre-training=w/o COCO, Backbone=ResNet-101, Output stride=82021.09 | 84.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RefineNetBackbone=ResNet152, COCO Pre-training=true2019.09 | 84.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RefineNetCOCO Pre-training=true, Backbone=ResNet1522019.09 | 84.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CFNetBackbone=ResNet-1012021.05 | 84.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpyGRBackbone=ResNet-1012021.05 | 84.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| APCNet*Pre-training=w/o COCO, Backbone=ResNet-101, Output stride=82021.09 | 84.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CFNet*Pre-training=w/o COCO, Backbone=ResNet-101, Output stride=82021.09 | 84.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SpyGR*Pre-training=w/o COCO, Backbone=ResNet-101, Output stride=82021.09 | 84.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RefineNetBackbone=ResNet-1522019.07 | 84.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FreeMaskBackbone=Swin-S, Data Size=40k2025.12 | 84.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EANetBackbone=ResNet-1012021.05 | 84 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ZeroSegTraining Dataset=ImageNet-1K [61], Backbone=ViT-B/16, Additional Labels=-, VFM=CLIP, DINO2024.09 | 83.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Raw DataBackbone=Swin-S, Data Size=11.5k2025.12 | 83.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GCNBackbone=ResNet-1522019.07 | 83.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RefineNet-152Input Resolution=512x512, FLOPs=283B2018.10 | 83.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SANetBackbone=ResNet-1012021.05 | 83.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SANet*Pre-training=w/o COCO, Backbone=ResNet-101, Output stride=82021.09 | 83.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EncNetCOCO pre-training=false2019.03 | 82.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EncNetBackbone=ResNet-1012021.05 | 82.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EncNetPre-training=w/o COCO, Backbone=ResNet-101, Output stride=82021.09 | 82.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RefineNet-LW-152Input Resolution=512x512, FLOPs=71B2018.10 | 82.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DFNCOCO pre-training=false2019.03 | 82.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DFNBackbone=ResNet-1012021.05 | 82.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DFN*Pre-training=w/o COCO, Backbone=ResNet-101, Output stride=82021.09 | 82.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| STRICTProtocol=Zero-shot2023.09 | 82.7 | — | — | — | — | — | — | — | — | — | 35.6 | 49.7 | — | — | |
| PSPNetCOCO pre-training=false2019.03 | 82.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PSPNetBackbone=ResNet-1012021.05 | 82.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DANetBackbone=ResNet-1012021.05 | 82.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PSPNetPre-training=w/o COCO, Backbone=ResNet-101, Output stride=82021.09 | 82.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DANet*Pre-training=w/o COCO, Backbone=ResNet-101, Output stride=82021.09 | 82.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ResNet-38COCO pre-training=false, Backbone=ResNet-382019.03 | 82.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RefineNet-101Input Resolution=512x512, FLOPs=263B2018.10 | 82.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DHRBackbone=Swin-L, Supervision=I2024.03 | 82.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RefineNet-LW-101Input Resolution=512x512, FLOPs=52B2018.10 | 82 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EfficientDetNote=A modified version of EfficientDet-D4, Params=17M, FLOPs=18B2019.11 | 81.74 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FMA-WSSSBackbone=Swin-L, Supervision=I+C+S2024.03 | 81.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RefineNet-LW-50Input Resolution=512x512, FLOPs=33B2018.10 | 81.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeepLabv3+Backbone=ResNet-101, Supervision=M2024.03 | 81 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MobileOne-S4Feature backbone=MobileOne-S4, Squeeze-Excite=false2022.06 | 80.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeepLabV3+Backbone=Xception, FLOPs=177B2019.11 | 80.02 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DHRBackbone=ResNet-101, Supervision=I2024.03 | 79.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeepLab-v2-ResNet-101-CRFInput Resolution=512x5122018.10 | 79.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DeeplabAdditional Training Dataset Used=JFT-300M [22]2019.12 | 79.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Deeplab-v2Backbone=ResNet101, COCO Pre-training=true2019.09 | 79.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Deeplab-v2COCO Pre-training=true, Backbone=ResNet1012019.09 | 79.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GroupViTTraining Dataset=CC12M [56], YFCC15M [57], Backbone=ViT-S/16, Additional Labels=-, VFM=-2024.09 | 79.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Core-tuning (MoCo-v2 pre-trained)Pre-training=MoCo-v2, Fine-tuning=Core-tuning, Backbone=ResNet-50, Architecture=DeepLab-V32021.02 | 79.62 | — | — | — | 88.76 | 90.75 | — | — | — | — | — | — | — | — | |
| WASPnet-CRF2019.12 | 79.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Core-tuning (InfoMin pre-trained)Pre-training=InfoMin, Fine-tuning=Core-tuning, Backbone=ResNet-50, Architecture=DeepLab-V32021.02 | 79.48 | — | — | — | 88.92 | 90.65 | — | — | — | — | — | — | — | — |