Semantic Segmentation on VOC 2012
97.7mIoUX-Decoder (DaViT-d5)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| X-Decoder (DaViT-d5)Additions=X-Decoder2026.01 | 97.7 | — | |
| SAN (ViT-L)Additions=Decoupled Head2026.01 | 94.6 | — | |
| Youtu-VL (4B)Additions=None2026.01 | 92.5 | — | |
| CASSAdditions=VFM Graph2026.01 | 87.8 | — | |
| FPModel=Segformer-mit-b2, Optimization Technique=Full Precision, Evaluation Protocol=Fine-tuning2025.03 | 80.93 | — | |
| LoRAModel=Segformer-mit-b2, Optimization Technique=LoRA, Evaluation Protocol=Fine-tuning2025.03 | 79.4 | — | |
| HOT + LoRAModel=Segformer-mit-b2, Optimization Technique=HOT + LoRA, Evaluation Protocol=Fine-tuning2025.03 | 79.1 | — | |
| HOTModel=Segformer-mit-b2, Optimization Technique=HOT, Evaluation Protocol=Fine-tuning2025.03 | 78.86 | — | |
| LUQModel=Segformer-mit-b2, Optimization Technique=LUQ, Evaluation Protocol=Fine-tuning2025.03 | 78.7 | — | |
| LBP-WHTModel=Segformer-mit-b2, Optimization Technique=LBP-WHT, Evaluation Protocol=Fine-tuning2025.03 | 78.4 | — | |
| CLIPSurgeryAdditions=Consistent Attn2026.01 | 77.5 | — | |
| SwinMTL (Swin-B)Additions=MLP Decoder2026.01 | 76.41 | — | |
| SMoGBackbone=ResNet-50, Segmentation Model=FCN, Evaluation Protocol=fine-tune2022.07 | 76.22 | 85.01 | |
| SwAVBackbone=ResNet-50, Segmentation Model=FCN, Evaluation Protocol=fine-tune2022.07 | 74.97 | 84.27 | |
| BYOLBackbone=ResNet-50, Segmentation Model=FCN, Evaluation Protocol=fine-tune2022.07 | 74.76 | 84.37 | |
| CM-GLasso2026.04 | 74.75 | — | |
| SupervisedBackbone=ResNet-50, Segmentation Model=FCN, Evaluation Protocol=fine-tune2022.07 | 73.59 | 83.74 | |
| PrivObNet2026.04 | 71.5 | — | |
| AuxSegNet+2026.04 | 70.9 | — | |
| MoCoV2Backbone=ResNet-50, Segmentation Model=FCN, Evaluation Protocol=fine-tune2022.07 | 70.86 | 80.37 | |
| H8(ViT-H/16)FLOPs (x10^9)=77.5, Evaluation Protocol=linear2025.05 | 70.8 | — | |
| ViT-H/16FLOPs (x10^9)=127.7, Evaluation Protocol=linear2025.05 | 70.7 | — | |
| H2(ViT-H/16)FLOPs (x10^9)=97.0, Evaluation Protocol=linear2025.05 | 70.7 | — | |
| I8(ViT-H/16)FLOPs (x10^9)=77.7, Evaluation Protocol=linear2025.05 | 70.6 | — | |
| I8(ViT-L/16)FLOPs (x10^9)=38.5, Evaluation Protocol=linear2025.05 | 70 | — | |
| H2(ViT-L/16)FLOPs (x10^9)=46.9, Evaluation Protocol=linear2025.05 | 69.3 | — | |
| ViT-L/16FLOPs (x10^9)=61.9, Evaluation Protocol=linear2025.05 | 69.1 | — | |
| H8(ViT-L/16)FLOPs (x10^9)=37.7, Evaluation Protocol=linear2025.05 | 69.1 | — | |
| ResNet18 + RFCAConv(r)Backbone=ResNet18, Stride=162023.04 | 68 | — | |
| ResNet18 + RFCBAMConv(r)Backbone=ResNet18, Stride=162023.04 | 67.7 | — | |
| ResNet18 + CAConv(r)Backbone=ResNet18, Stride=162023.04 | 66.6 | — | |
| EQ-VMamba-SBackbone=EQ-VMamba-S, #Param. (M)=25M2026.03 | 66.57 | — | |
| ResNet-50 FCNModel=ResNet-50, Architecture=FCN2026.05 | 66.5 | — | |
| VMamba-SBackbone=VMamba-S, #Param. (M)=82M2026.03 | 66.14 | — | |
| MSVMamba-TBackbone=MSVMamba-T, #Param. (M)=65M2026.03 | 65.98 | — | |
| ResNet18 + CAMConv(r)Backbone=ResNet18, Stride=162023.04 | 65.5 | — | |
| ResNet18 + RFAConv(r)Backbone=ResNet18, Stride=162023.04 | 65.4 | — | |
| ResNet18Backbone=ResNet18, Stride=162023.04 | 64.6 | — | |
| EQ-VMamba-TBackbone=EQ-VMamba-T, #Param. (M)=18M2026.03 | 64.46 | — | |
| ResNet18 + RFCAConv(r)Backbone=ResNet18, Stride=82023.04 | 63.9 | — | |
| VMamba-TBackbone=VMamba-T, #Param. (M)=62M2026.03 | 63.7 | — | |
| ResNet18 + CBAMConv(r)Backbone=ResNet18, Stride=162023.04 | 63.6 | — | |
| MaskCLIPAdditions=Attn Adaptation2026.01 | 62.9 | — | |
| ResNet18 + CAConv(r)Backbone=ResNet18, Stride=82023.04 | 62.1 | — | |
| ResNet18 + RFCBAMConv(r)Backbone=ResNet18, Stride=82023.04 | 62.1 | — | |
| H8(ViT-H/16)FLOPs (x10^9)=77.5, Evaluation Protocol=k-NN2025.05 | 61.7 | — | |
| WWT-SEGModel=WWT, Task head setting=Proposed task-head2026.05 | 61.3 | — | |
| H2(ViT-H/16)FLOPs (x10^9)=97.0, Evaluation Protocol=k-NN2025.05 | 61.2 | — | |
| I8(ViT-H/16)FLOPs (x10^9)=77.7, Evaluation Protocol=k-NN2025.05 | 61.2 | — | |
| ResNet18 + CAMConv(r)Backbone=ResNet18, Stride=82023.04 | 60.9 | — | |
| ViT-H/16FLOPs (x10^9)=127.7, Evaluation Protocol=k-NN2025.05 | 60.9 | — | |
| ResNet18 + RFAConv(r)Backbone=ResNet18, Stride=82023.04 | 60.8 | — | |
| WWT + seg. head on tokensModel=WWT, Task head setting=heads-on-patch-tokens2026.05 | 59.9 | — | |
| MSVMamba-MBackbone=MSVMamba-M, #Param. (M)=42M2026.03 | 59.78 | — | |
| I8(ViT-L/16)FLOPs (x10^9)=38.5, Evaluation Protocol=k-NN2025.05 | 59.6 | — | |
| H8(ViT-L/16)FLOPs (x10^9)=37.7, Evaluation Protocol=k-NN2025.05 | 59.5 | — | |
| ResNet18 + CBAMConv(r)Backbone=ResNet18, Stride=82023.04 | 59.3 | — | |
| ResNet18Backbone=ResNet18, Stride=82023.04 | 58.9 | — | |
| ConvNeXt-SBackbone=ConvNeXt-S, #Param. (M)=70M2026.03 | 58.83 | — | |
| XCiT-M24Backbone=XCiT-M24, #Param. (M)=112M2026.03 | 58.63 | — | |
| ViT-L/16FLOPs (x10^9)=61.9, Evaluation Protocol=k-NN2025.05 | 58.1 | — | |
| XCiT-S24Backbone=XCiT-S24, #Param. (M)=76M2026.03 | 58.08 | — | |
| H2(ViT-L/16)FLOPs (x10^9)=46.9, Evaluation Protocol=k-NN2025.05 | 58 | — | |
| ConvNeXt-TBackbone=ConvNeXt-T, #Param. (M)=48M2026.03 | 53.99 | — | |
| GroupViT2026.04 | 52.3 | — | |
| Swin-SBackbone=Swin-S, #Param. (M)=69M2026.03 | 38.89 | — | |
| Swin-TBackbone=Swin-T, #Param. (M)=48M2026.03 | 35.73 | — | |
| Vim-SBackbone=Vim-S, #Param. (M)=46M2026.03 | 33.16 | — | |
| Vim-TBackbone=Vim-T, #Param. (M)=13M2026.03 | 32.2 | — | |
| DeiT-BBackbone=DeiT-B, #Param. (M)=121M2026.03 | 24.85 | — | |
| DeiT-SBackbone=DeiT-S, #Param. (M)=52M2026.03 | 24.17 | — |