Semantic Segmentation on ADE20K (test)
56.23mIoUDNL
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| DNLBackbone=ResNet-1012020.06 | 56.23 | — | — | — | |
| DNLBackbone=HRNetV2-W482020.06 | 55.98 | — | — | — | |
| ACNetBackbone=ResNet-1012020.06 | 55.84 | — | — | — | |
| EncNetBackbone=ResNet-1012020.06 | 55.67 | — | — | — | |
| NLBackbone=HRNetV2-W482020.06 | 55.6 | — | — | — | |
| NLBackbone=ResNet-1012020.06 | 55.58 | — | — | — | |
| PSANetBackbone=ResNet-1012020.06 | 55.46 | — | — | — | |
| JoDiffusionBackbone=Swin-S, Data Size=40k2025.12 | 52.2 | — | — | — | |
| FreeMaskBackbone=Swin-S, Data Size=40k2025.12 | 52.1 | — | — | — | |
| Raw DataBackbone=Swin-S, Data Size=20k2025.12 | 51.6 | — | — | — | |
| SegGenBackbone=ResNet50, Data Size=1M2025.12 | 49.9 | — | — | — | |
| JoDiffusionBackbone=ResNet50, Data Size=40k2025.12 | 48.4 | — | — | — | |
| FreeMaskBackbone=ResNet50, Data Size=40k2025.12 | 48.2 | — | — | — | |
| Raw DataBackbone=ResNet50, Data Size=20k2025.12 | 47.2 | — | — | — | |
| DeepMAD-29M*Backbone=DeepMAD-29M*, Params=26.5 M, FLOPs=55.5 G, Framework=UperNet, Pre-training=ImageNet-1K2023.03 | 46.9 | — | — | — | |
| ConvNeXt-TBackbone=ConvNeXt-T, Params=27.8 M, FLOPs=93.2 G, Framework=UperNet, Pre-training=ImageNet-1K2023.03 | 46.7 | — | — | — | |
| Agent-Swin-TType=SA, GMACS=24.2, iPhone13 Latency (ms)=26.8, RTX 4090 Latency (s)=0.26, Input Resolution=512x512, Head=Semantic FPN2024.11 | 46.7 | — | — | — | |
| Swin-TBackbone=Swin-T, Params=27.5 M, FLOPs=95.8 G, Framework=UperNet, Pre-training=ImageNet-1K2023.03 | 45.8 | — | — | — | |
| DeepMAD-R50Backbone=DeepMAD-R50, Params=24.2 M, FLOPs=85.2 G, Framework=UperNet, Pre-training=ImageNet-1K2023.03 | 45.6 | — | — | — | |
| ResNet-101Backbone=ResNet-101, Params=42.5 M, FLOPs=164.3 G, Framework=UperNet, Pre-training=ImageNet-1K2023.03 | 44.8 | — | — | — | |
| FLatten-Swin-TType=LA, GMACS=24.2, iPhone13 Latency (ms)=28.6, RTX 4090 Latency (s)=0.28, Input Resolution=512x512, Head=Semantic FPN2024.11 | 44.8 | — | — | — | |
| DeepLabV3 + BANELoss=L_CE + L_PA (Ours), Sampling=Boundary-aware (Ours)2024.04 | 44.12 | — | — | — | |
| DeepLabV3 + [31]Loss=L_CE + L_ems + L_ccs, Sampling=Random2024.04 | 43.86 | — | — | — | |
| OCRNet + BANELoss=L_CE + L_PA (Ours), Sampling=Boundary-aware (Ours)2024.04 | 43.84 | — | — | — | |
| EfficientFormer-L3Type=CONV, GMACS=21, iPhone13 Latency (ms)=10.3, RTX 4090 Latency (s)=0.2, Input Resolution=512x512, Head=Semantic FPN2024.11 | 43.5 | — | — | — | |
| CARE-S2Type=LA+CONV, GMACS=10.1, iPhone13 Latency (ms)=7.3, RTX 4090 Latency (s)=0.14, Input Resolution=512x512, Head=Semantic FPN2024.11 | 43.5 | — | — | — | |
| HRNet + BANELoss=L_CE + L_PA (Ours), Sampling=Boundary-aware (Ours)2024.04 | 43.42 | — | — | — | |
| OCRNet + [31]Loss=L_CE + L_ems + L_ccs, Sampling=Random2024.04 | 43.28 | — | — | — | |
| HRNet + [31]Loss=L_CE + L_ems + L_ccs, Sampling=Random2024.04 | 43.27 | — | — | — | |
| MobileViGv2-MType=GNN+CONV, GMACS=8.5, iPhone13 Latency (ms)=5.5, RTX 4090 Latency (s)=0.14, Input Resolution=512x512, Head=Semantic FPN2024.11 | 42.9 | — | — | — | |
| DeepLabV3Loss=L_CE, Sampling=None2024.04 | 42.85 | — | — | — | |
| ResNet-50Backbone=ResNet-50, Params=23.5 M, FLOPs=86.3 G, Framework=UperNet, Pre-training=ImageNet-1K2023.03 | 42.8 | — | — | — | |
| Panoptic-DeepLab w/ SWideRNet-(1, 1, 4)MS=true2020.11 | 42.09 | — | 59.14 | 76.19 | |
| HRNetLoss=L_CE, Sampling=None2024.04 | 41.86 | — | — | — | |
| OCRNetLoss=L_CE, Sampling=None2024.04 | 41.51 | — | — | — | |
| Swin-TType=SA, GMACS=24.2, iPhone13 Latency (ms)=28.2, RTX 4090 Latency (s)=0.28, Input Resolution=512x512, Head=Semantic FPN2024.11 | 41.5 | — | — | — | |
| CARE-S1Type=LA+CONV, GMACS=5.4, iPhone13 Latency (ms)=5.5, RTX 4090 Latency (s)=0.11, Input Resolution=512x512, Head=Semantic FPN2024.11 | 41 | — | — | — | |
| Panoptic-DeepLab w/ SWideRNet-(1, 1.5, 3)MS=true2020.11 | 40.47 | — | 57.84 | 75.21 | |
| PoolFormer-S24Type=CONV, GMACS=17.7, iPhone13 Latency (ms)=9.8, RTX 4090 Latency (s)=0.17, Input Resolution=512x512, Head=Semantic FPN2024.11 | 40.3 | — | — | — | |
| Agent-PVT-TType=SA, GMACS=10.7, iPhone13 Latency (ms)=17.6, RTX 4090 Latency (s)=0.12, Input Resolution=512x512, Head=Semantic FPN2024.11 | 40.2 | — | — | — | |
| PyConvSegNet-152single model=true, training epochs=120, training sets=training+validation2020.06 | 39.13 | — | 56.52 | 73.91 | |
| EfficientFormer-L1Type=CONV, GMACS=6.9, iPhone13 Latency (ms)=4.8, RTX 4090 Latency (s)=0.09, Input Resolution=512x512, Head=Semantic FPN2024.11 | 38.9 | — | — | — | |
| ResNet101Type=CONV, GMACS=40.8, iPhone13 Latency (ms)=11.5, RTX 4090 Latency (s)=0.22, Input Resolution=512x512, Head=Semantic FPN2024.11 | 38.8 | — | — | — | |
| CARE-S0Type=LA+CONV, GMACS=3.8, iPhone13 Latency (ms)=4.1, RTX 4090 Latency (s)=0.07, Input Resolution=512x512, Head=Semantic FPN2024.11 | 38.5 | — | — | — | |
| FastViT-SA12Type=SA+CONV, GMACS=8, iPhone13 Latency (ms)=6.9, RTX 4090 Latency (s)=0.14, Input Resolution=512x512, Head=Semantic FPN2024.11 | 38 | — | — | — | |
| PyConvSegNet-152single model=true2020.06 | 37.75 | — | 55.68 | 73.61 | |
| PoolFormer-S12Type=CONV, GMACS=9.5, iPhone13 Latency (ms)=5.7, RTX 4090 Latency (s)=0.09, Input Resolution=512x512, Head=Semantic FPN2024.11 | 37.2 | — | — | — | |
| FLatten-PVT-TType=LA, GMACS=10.4, iPhone13 Latency (ms)=18.6, RTX 4090 Latency (s)=0.14, Input Resolution=512x512, Head=Semantic FPN2024.11 | 37.2 | — | — | — | |
| ResNet50Type=CONV, GMACS=21.4, iPhone13 Latency (ms)=7.6, RTX 4090 Latency (s)=0.14, Input Resolution=512x512, Head=Semantic FPN2024.11 | 36.7 | — | — | — | |
| ResNet18Type=CONV, GMACS=9.5, iPhone13 Latency (ms)=3.2, RTX 4090 Latency (s)=0.04, Input Resolution=512x512, Head=Semantic FPN2024.11 | 32.9 | — | — | — | |
| 360+MCG-ICT-CAS_SPmodels=12016.11 | — | 54.68 | — | — | |
| 360+MCG-ICT-CAS_SPmodels=-2016.11 | — | 55.57 | — | — | |
| ACNetBackbone=ResNet-1012021.03 | — | — | 38.5 | — | |
| ACNet-101Backbone=ResNet-1012019.11 | — | — | 55.84 | — | |
| CASIA_IVAmodels=12016.11 | — | 54.33 | — | — | |
| CASIA_IVA_JD2018.03 | — | — | 55.47 | — | |
| CASIA_IVA_JDSingle Model=false2019.03 | — | — | 55.47 | — | |
| CASIA_IVA_JDAward=1st in place 20172019.11 | — | — | 55.47 | — | |
| DilatedNetsingle model=true2020.06 | — | — | 45.67 | — | |
| DLab.v3+Backbone=ResNeSt-101, #param.=66M, FLOPS=1051G, FPS=11.92021.03 | — | — | 55.1 | — | |
| DNLBackbone=ResNet-101, #param.=69M, FLOPS=1249G, FPS=14.82021.03 | — | — | 56.2 | — | |
| EncNetBackbone=ResNet-1012019.03 | — | — | 55.67 | — | |
| EncNetsingle model=true2020.06 | — | — | 55.67 | — | |
| EncNet-101Model Type=single model, Backbone=ResNet-1012018.03 | — | — | 55.67 | — | |
| EncNet-101Backbone=ResNet-1012019.11 | — | — | 55.67 | — | |
| EncNet+JPUBackbone=ResNet-1012019.03 | — | — | 55.84 | — | |
| FCNsingle model=true2020.06 | — | — | 44.8 | — | |
| Model A & Model Cmodels=22016.11 | — | 56.74 | — | — | |
| Model A2models=12016.11 | — | 56.41 | — | — | |
| NTU-SPmodels=22016.11 | — | 53.57 | — | — | |
| OCRNetBackbone=ResNet-101, #param.=56M, FLOPS=923G, FPS=19.32021.03 | — | — | 56 | — | |
| PSANet-101Backbone=ResNet-1012019.11 | — | — | 55.46 | — | |
| PSPNetBackbone=ResNet-2692019.03 | — | — | 55.38 | — | |
| PSPNet (ensemble-model)MS=true2020.11 | — | — | 57.21 | — | |
| PSPNet (single-model)MS=true2020.11 | — | — | 55.38 | — | |
| PSPNet-269Model Type=single model, Backbone=ResNet-2692018.03 | — | — | 55.38 | — | |
| PSPNet-269single model=true2020.06 | — | — | 55.38 | — | |
| PSPNet269Backbone=ResNet-269, Award=1st in place 20162019.11 | — | — | 55.38 | — | |
| SegModelmodels=12016.11 | — | 53.23 | — | — | |
| SegModelmodels=52016.11 | — | 54.65 | — | — | |
| SegNetsingle model=true2020.06 | — | — | 40.79 | — | |
| SenseCUSceneParsingmodels=12016.11 | — | 55.38 | — | — | |
| SenseCUSceneParsingmodels=-2016.11 | — | 57.21 | — | — | |
| SETRBackbone=T-Large, #param.=308M2021.03 | — | — | 61.7 | — | |
| UperNetBackbone=Swin-L, Pre-trained on ImageNet-22K=true, #param.=234M, FLOPS=3230G, FPS=6.22021.03 | — | — | 62.8 | — | |
| WinterIsComing2018.03 | — | — | 55.44 | — | |
| WinterIsComingSingle Model=false2019.03 | — | — | 55.44 | — |