Semantic Segmentation on SUN RGB-D (test)
54.6mIoUGeminiFusion
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| GeminiFusionBackbone=Swin-Large-22k, Additional Strategies=false2024.06 | 54.6 | — | — | — | — | — | — | — | |
| GeminiFusionBackbone=MiT-B5, Additional Strategies=false2024.06 | 53.3 | — | — | — | — | — | — | — | |
| DPLNetScale=MS, Backbone=MiT-B5, Params (M)=7.152023.12 | 52.8 | — | — | — | — | — | — | — | |
| DPLNetBackbone=MiT-B5, Additional Strategies=false2024.06 | 52.8 | — | — | — | — | — | — | — | |
| GeminiFusionBackbone=MiT-B3, Additional Strategies=false2024.06 | 52.7 | — | — | — | — | — | — | — | |
| DFormer-LScale=MS, Backbone=DFormer-L, Params (M)=392023.12 | 52.5 | — | — | — | — | — | — | — | |
| DFormerBackbone=DFormer-L, Additional Strategies=true2024.06 | 52.5 | — | — | — | — | — | — | — | |
| CMX-B5Scale=MS, Backbone=MiT-B5, Params (M)=181.12023.12 | 52.4 | — | — | — | — | — | — | — | |
| CMX-B5multi-scale=true2023.09 | 52.4 | — | — | 83.8 | — | — | — | — | |
| CMXBackbone=MiT-B5, Additional Strategies=false2024.06 | 52.4 | — | — | — | — | — | — | — | |
| CMX-B4Scale=MS, Backbone=MiT-B4, Params (M)=139.92023.12 | 52.1 | — | — | — | — | — | — | — | |
| DPLNetScale=SS, Backbone=MiT-B5, Params (M)=7.152023.12 | 52.1 | — | — | — | — | — | — | — | |
| CMX-B4multi-scale=true2023.09 | 52.1 | — | — | 83.5 | — | — | — | — | |
| CMXBackbone=2x MiT-B4, test-time augmentation=true2026.01 | 52.1 | — | — | — | — | — | — | — | |
| CMXNeXtScale=MS, Backbone=MiT-B4, Params (M)=119.62023.12 | 51.9 | — | — | — | — | — | — | — | |
| CMNeXtBackbone=2x MiT-B42026.01 | 51.9 | — | — | — | — | — | — | — | |
| FRNet2023.09 | 51.8 | — | — | 87.4 | — | — | — | — | |
| TokenFusion†Backbone=MiT-B5, Additional Strategies=false, Reproduced=true2024.06 | 51.8 | — | — | — | — | — | — | — | |
| DFormer-BScale=MS, Backbone=DFormer-B, Params (M)=29.52023.12 | 51.2 | — | — | — | — | — | — | — | |
| DFormerBackbone=DFormer-B, Extra Training Data=true2026.01 | 51.2 | — | — | — | — | — | — | — | |
| Ours-PSPNetModality=RGB-D, Backbone Network=ResNet1522021.12 | 51.1 | 83.5 | 63.2 | — | — | — | — | — | |
| MultiMAEBackbone=ViT-B, Extra Training Data=true2026.01 | 51.1 | — | — | — | — | — | — | — | |
| PGDENetScale=SS, Backbone=ResNet-34, Params (M)=100.72023.12 | 51 | — | — | — | — | — | — | — | |
| TokenFusion-B3Scale=SS, Backbone=MiT-B3, Params (M)=45.92023.12 | 51 | — | — | — | — | — | — | — | |
| PGDENet2023.09 | 51 | — | — | 87.7 | — | — | — | — | |
| DVEFormerBackbone=SwinV2-T-128, Extra Training Data=true, Evaluation Protocol=Linear probing2026.01 | 50.99 | — | — | — | — | — | — | — | |
| PanopticNDT (EMSANet-R34-NBt1D)Protocol=Fine-tuning, LR=0.002, Selection=Best in run2023.09 | 50.91 | — | — | — | — | — | — | — | |
| EMSANetBackbone=EMSANet-R34-NBt1D, Additional Strategies=false2024.06 | 50.9 | — | — | — | — | — | — | — | |
| PanopticNDT (EMSANet-R34-NBt1D)Protocol=Fine-tuning, LR=0.002, Selection=Best on SUNRGB-D2023.09 | 50.86 | — | — | — | — | — | — | — | |
| Ours-RefineNetModality=RGB-D, Backbone Network=ResNet152, Evaluation Scale=multi-scale2021.12 | 50.8 | 83.2 | 62.5 | — | — | — | — | — | |
| SCNModality=RGB-D, Backbone Network=ResNet1522021.12 | 50.7 | — | — | — | — | — | — | — | |
| FSFNetBackbone=ResNet-1012021.05 | 50.6 | 81.8 | — | — | — | — | — | — | |
| PSDBackbone=ResNet50, Additional Strategies=false2024.06 | 50.6 | — | — | — | — | — | — | — | |
| FSFNetBackbone=ResNet-101, Additional Strategies=false2024.06 | 50.6 | — | — | — | — | — | — | — | |
| PAP2021.05 | 50.5 | 83.8 | — | — | — | — | — | — | |
| CMNeXtBackbone=MiT-B4, Additional Strategies=false2024.06 | 50.4 | — | — | — | — | — | — | — | |
| MSFNet2023.09 | 50.3 | — | — | — | — | — | — | — | |
| Ours-RefineNetModality=RGB-D, Backbone Network=ResNet101, Evaluation Scale=multi-scale2021.12 | 50.2 | 82.8 | 61.9 | — | — | — | — | — | |
| GeminiFusionBackbone=Swin-Tiny-1k, Additional Strategies=false2024.06 | 50.2 | — | — | — | — | — | — | — | |
| DVEFormerBackbone=SwinV2-T-128, Extra Training Data=true, Resolution=H/4 x W/4, Evaluation Protocol=Linear probing2026.01 | 50.11 | — | — | — | — | — | — | — | |
| Ours-RefineNetModality=RGB-D, Backbone Network=ResNet152, Evaluation Scale=single-scale2021.12 | 50 | 82.3 | 61.7 | — | — | — | — | — | |
| DFormer-S2023.09 | 50 | — | — | — | — | — | — | — | |
| DFormerBackbone=DFormer-S, Extra Training Data=true2026.01 | 50 | — | — | — | — | — | — | — | |
| SceneNetLabeled Examples=Full(5.3k), Backbone=Resnet-502018.11 | 49.8 | — | — | — | — | — | — | — | |
| CMX-B2multi-scale=true2023.09 | 49.7 | — | — | 82.8 | — | — | — | — | |
| CMXBackbone=2x MiT-B2, test-time augmentation=true2026.01 | 49.7 | — | — | — | — | — | — | — | |
| Ours-RefineNetModality=RGB-D, Backbone Network=ResNet101, Evaluation Scale=single-scale2021.12 | 49.6 | 82 | 60.9 | — | — | — | — | — | |
| DCANetmulti-scale testing=true2022.10 | 49.6 | — | — | 82.6 | — | — | — | — | |
| SA-GateBackbone=2x ResNet-50, Test-time augmentation=true, FPS=11.92020.11 | 49.4 | — | — | — | — | — | — | — | |
| SA-Gate2021.05 | 49.4 | 82.5 | — | — | — | — | — | — | |
| SA-Gatemulti-scale testing=true2022.10 | 49.4 | — | — | 82.5 | — | — | — | — | |
| SA-GateScale=MS, Backbone=ResNet-101, Params (M)=110.92023.12 | 49.4 | — | — | — | — | — | — | — | |
| SA-Gate2023.09 | 49.4 | — | — | 82.5 | — | — | — | — | |
| PanopticNDT (EMSANet-R34-NBt1D)Protocol=Application network, LR=0.0005, Selection=Best in run2023.09 | 49.31 | — | — | — | — | — | — | — | |
| EMSANetBackbone=2x ResNet34-NBt1D, Extra Training Data=true2026.01 | 49.31 | — | — | — | — | — | — | — | |
| PanopticNDT (EMSANet-R34-NBt1D)Protocol=Application network, LR=0.0005, Selection=Best on SUNRGB-D2023.09 | 49.3 | — | — | — | — | — | — | — | |
| AsymFormer2023.09 | 49.1 | — | — | 81.9 | — | — | — | — | |
| EMSAFormerBackbone=SwinV2-T-128, Augmentation Strategy=Multi-Aug, Decoder Configuration=Sem(SegFormer)2023.06 | 48.82 | — | — | — | — | — | — | — | |
| MIPANetModel=ResNet50, Backbone=2 x R502023.11 | 48.8 | — | — | 82.3 | — | — | — | — | |
| EMSAFormerBackbone=SwinV2-T-128, Augmentation Strategy=Multi-Aug2023.06 | 48.67 | — | — | — | — | — | — | — | |
| EMSAFormerBackbone=SwinV2-T-1282026.01 | 48.61 | — | — | — | — | — | — | — | |
| SGNet*Backbone=ResNet101, MS=true, SI=Depth, param (M)=64.7, ASPP=true2020.04 | 48.6 | 82 | 60.7 | — | — | — | — | — | |
| ShapeConvmulti-scale testing=true2022.10 | 48.6 | — | — | 82.2 | — | — | — | — | |
| SGNetScale=MS, Backbone=ResNet-101, Params (M)=64.72023.12 | 48.6 | — | — | — | — | — | — | — | |
| ShapeConvModel=ResNet101, Backbone=R1012023.11 | 48.6 | — | — | 82.2 | — | — | — | — | |
| EMSANetModel=ResNet34, Backbone=2 x R342023.11 | 48.5 | — | — | — | — | — | — | — | |
| EMSANetBackbone=2x ResNet34-NBt1D2023.06 | 48.39 | — | — | — | — | — | — | — | |
| EMSANetBackbone=2x ResNet34-NBt1D2026.01 | 48.39 | — | — | — | — | — | — | — | |
| ESANet-R50Backbone=2x ResNet-50, FPS=22.62020.11 | 48.31 | — | — | — | — | — | — | — | |
| IEMNetModel=ResNet34, Backbone=Res34NBt1D2023.11 | 48.3 | — | — | 81.9 | — | — | — | — | |
| ESANetModel=ResNet50, Backbone=2 x R502023.11 | 48.3 | — | — | — | — | — | — | — | |
| CANetModel=ResNet101, Backbone=R1012023.11 | 48.3 | — | — | 82 | — | — | — | — | |
| CGBNetBackbone=ResNet101, MS=false2020.04 | 48.2 | 82.3 | 61.3 | — | — | — | — | — | |
| 2.5D ConvBackbone=ResNet-101, Inference complexity=∇ expected to be slower due to complex backbone, FPS=N/A2020.11 | 48.2 | — | — | — | — | — | — | — | |
| CGBNetModel=ResNet101, Backbone=R1012023.11 | 48.2 | — | — | 82.3 | — | — | — | — | |
| ESANet-R34-NBt1DBackbone=2x ResNet-34 NBt1D, FPS=29.72020.11 | 48.17 | — | — | — | — | — | — | — | |
| CFN (RefineNet-152)Backbone=RefineNet-1522018.06 | 48.1 | — | — | — | — | — | — | — | |
| CFNBackbone=RefineNet-1522019.05 | 48.1 | — | — | — | — | — | — | — | |
| ACNetBackbone=ResNet-502019.05 | 48.1 | — | — | — | — | — | — | — | |
| ACNetBackbone=2xResNet50, MS=false, SI=HHA, param (M)=272.22020.04 | 48.1 | — | — | — | — | — | — | — | |
| CFNetBackbone=2xResNet152, MS=true, SI=HHA2020.04 | 48.1 | — | — | — | — | — | — | — | |
| ACNetBackbone=3x ResNet-50, FPS=16.52020.11 | 48.1 | — | — | — | — | — | — | — | |
| CFN2021.05 | 48.1 | — | — | — | — | — | — | — | |
| ACNet2021.05 | 48.1 | — | — | — | — | — | — | — | |
| CFNModality=RGB-D, Backbone Network=ResNet1522021.12 | 48.1 | — | — | — | — | — | — | — | |
| ACNetmulti-scale testing=false2022.10 | 48.1 | — | — | — | — | — | — | — | |
| CRFmulti-scale testing=true2022.10 | 48.1 | — | — | — | — | — | — | — | |
| DCANetmulti-scale testing=false2022.10 | 48.1 | — | — | 82.2 | — | — | — | — | |
| ACNetBackbone=ResNet50, SI=RGB-D2023.08 | 48.1 | — | 60.3 | — | — | — | — | — | |
| ACNetScale=SS, Backbone=ResNet-50, Params (M)=116.62023.12 | 48.1 | — | — | — | — | — | — | — | |
| ACNetModel=ResNet50, Backbone=3 x R502023.11 | 48.1 | — | — | — | — | — | — | — | |
| ACNet2023.09 | 48.1 | — | — | — | — | — | — | — | |
| ESANet (pre. SceneNet)Backbone=2x ResNet-34 NBt1D, Pre-training=SceneNet, FPS=29.72020.11 | 48.04 | — | — | — | — | — | — | — | |
| ESANet2023.09 | 48 | — | — | — | — | — | — | — | |
| EMSANetBackbone=2x ResNet1012023.06 | 47.99 | — | — | — | — | — | — | — | |
| SSMAModality=RGB-D, Backbone Network=ResNet101, re-implemented=true2021.12 | 47.9 | 81.6 | 60.4 | — | — | — | — | — | |
| RedNetBackbone=ResNet-502018.06 | 47.8 | 81.3 | 60.3 | — | — | — | — | — | |
| RedNet2021.05 | 47.8 | 81.3 | — | — | — | — | — | — | |
| SGACNetBackbone=R34-NBt1D, SI=RGB-D2023.08 | 47.8 | — | 60.8 | 81.2 | — | — | — | — | |
| RedNetModel=ResNet50, Backbone=2 x R502023.11 | 47.8 | — | — | 81.3 | — | — | — | — |