Semantic Segmentation on NYU Depth V2 (test)
61.2mIoUDAT++-S
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| DAT++-SBackbone=DAT++-S, Params(M)=172.8, Sampling steps=3, Training dataset=NYUv2, Scale value s=0.012024.09 | 61.2 | — | — | — | — | — | |
| DAT++-S (Cross-eval)Backbone=DAT++-S, Params(M)=172.8, Sampling steps=3, Training dataset=SUN-RGBD, Scale value s=0.012024.09 | 61.1 | — | — | — | — | — | |
| GeminiFusionBackbone=Swin-L, Params(M)=369.2, Training dataset=NYUv22024.09 | 60.9 | — | — | — | — | — | |
| DAT++-BBackbone=DAT++-B, Params(M)=280.0, Sampling steps=3, Training dataset=NYUv2, Scale value s=0.012024.09 | 60.8 | — | — | — | — | — | |
| DAT++-B (Cross-eval)Backbone=DAT++-B, Params(M)=280.0, Sampling steps=3, Training dataset=SUN-RGBD, Scale value s=0.012024.09 | 60.4 | — | — | — | — | — | |
| DAT++-TBackbone=DAT++-T, Params(M)=73.2, Sampling steps=3, Training dataset=NYUv2, Scale value s=0.012024.09 | 59.9 | — | — | — | — | — | |
| DAT++-T (Cross-eval)Backbone=DAT++-T, Params(M)=73.2, Sampling steps=3, Training dataset=SUN-RGBD, Scale value s=0.012024.09 | 59.7 | — | — | — | — | — | |
| DFormerv2-LBackbone=DFormerv2-Large, Params=95.5M, Input size=480 x 640, Flops=124.1G2025.04 | 58.4 | — | — | — | — | — | |
| SwinMTLBackbone=SwinV2-B2024.03 | 58.14 | — | — | — | — | — | |
| GeminiFusionBackbone=MiT-B5, Params(M)=137.2, Training dataset=NYUv22024.09 | 57.7 | — | — | — | — | — | |
| DFormerv2-BBackbone=DFormerv2-Base, Params=53.9M, Input size=480 x 640, Flops=67.2G2025.04 | 57.7 | — | — | — | — | — | |
| GeminiFusion24Backbone=MiT-B5, Params=137.2M, Input size=480 x 640, Flops=256.1G2025.04 | 57.7 | — | — | — | — | — | |
| DFormer-LBackbone=DFormer-L, Params(M)=39.0, Training dataset=NYUv22024.09 | 57.2 | — | — | — | — | — | |
| DFormer24Backbone=DFormer-Large, Params=39.0M, Input size=480 x 640, Flops=65.7G2025.04 | 57.2 | — | — | — | — | — | |
| CMX (MiT-B5)Backbone=MiT-B5, Params(M)=181.1, Training dataset=NYUv22024.09 | 56.9 | — | — | — | — | — | |
| DFormer-BBackbone=DFormer-B, Params(M)=29.5, Training dataset=NYUv22024.09 | 56.9 | — | — | — | — | — | |
| CMXModalities=RGB-Depth, Backbone=MiT-B52023.03 | 56.9 | — | — | — | — | — | |
| CMNeXtModalities=RGB-Depth, Backbone=MiT-B42023.03 | 56.9 | — | — | — | — | — | |
| CMXBackbone=MiT-B5, Multi-scale test=true2022.03 | 56.9 | 80.1 | — | — | — | — | |
| CMX22Backbone=MiT-B5, Params=181.1M, Input size=480 x 640, Flops=167.8G2025.04 | 56.9 | — | — | — | — | — | |
| CMNext23Backbone=MiT-B4, Params=119.6M, Input size=480 x 640, Flops=131.9G2025.04 | 56.9 | — | — | — | — | — | |
| GeminiFusionBackbone=MiT-B3, Params(M)=75.8, Training dataset=NYUv22024.09 | 56.8 | — | — | — | — | — | |
| OmnivoreModalities=RGB-Depth2023.03 | 56.8 | — | — | — | — | — | |
| CMXBackbone=MiT-B5, Multi-scale test=false2022.03 | 56.8 | 79.9 | — | — | — | — | |
| GeminiFusion24Backbone=MiT-B3, Params=75.8M, Input size=480 x 640, Flops=138.2G2025.04 | 56.8 | — | — | — | — | — | |
| CMX (MiT-B4)Backbone=MiT-B4, Params(M)=139.9, Training dataset=NYUv22024.09 | 56.3 | — | — | — | — | — | |
| CMXModalities=RGB-Depth, Backbone=MiT-B42023.03 | 56.3 | — | — | — | — | — | |
| CMXBackbone=MiT-B4, Multi-scale test=true2022.03 | 56.3 | 79.9 | — | — | — | — | |
| CMX22Backbone=MiT-B4, Params=139.9M, Input size=480 x 640, Flops=134.3G2025.04 | 56.3 | — | — | — | — | — | |
| MultiMAEModalities=RGB-Depth2023.03 | 56 | — | — | — | — | — | |
| CMXBackbone=MiT-B4, Multi-scale test=false2022.03 | 56 | 79.6 | — | — | — | — | |
| DFormerv2-SBackbone=DFormerv2-Small, Params=26.7M, Input size=480 x 640, Flops=33.9G2025.04 | 56 | — | — | — | — | — | |
| MultiMAE22Backbone=ViT-Base, Params=95.2M, Input size=640 x 640, Flops=267.9G2025.04 | 56 | — | — | — | — | — | |
| DFormer24Backbone=DFormer-Base, Params=29.5M, Input size=480 x 640, Flops=41.9G2025.04 | 55.6 | — | — | — | — | — | |
| TransD-FusionModalities=RGB-Depth2023.03 | 55.5 | — | — | — | — | — | |
| TaskPrompter2024.03 | 55.3 | — | — | — | — | — | |
| AsymFormer24Backbone=MiT-B0+ConvNeXt-Tiny, Params=33.0M, Input size=480 x 640, Flops=39.4G2025.04 | 55.3 | — | — | — | — | — | |
| CMXBackbone=MiT-B2, Multi-scale test=true2022.03 | 54.4 | 79.9 | — | — | — | — | |
| CMX22Backbone=MiT-B2, Params=66.6M, Input size=480 x 640, Flops=67.6G2025.04 | 54.4 | — | — | — | — | — | |
| TokenFusionBackbone=MiT-B3, Params(M)=45.9, Training dataset=NYUv22024.09 | 54.2 | — | — | — | — | — | |
| TokenFusionModalities=RGB-Depth2023.03 | 54.2 | — | — | — | — | — | |
| TokenFusion22Backbone=MiT-B3, Params=45.9M, Input size=480 x 640, Flops=94.4G2025.04 | 54.2 | — | — | — | — | — | |
| CMXModal=RGB-D2023.12 | 54.1 | — | — | — | — | — | |
| ShareCMPModal=RGB-D, w/o PGA and CPALoss=true2023.12 | 54.1 | — | — | — | — | — | |
| CMXBackbone=MiT-B2, Multi-scale test=false2022.03 | 54.1 | 78.7 | — | — | — | — | |
| Omnivore22Backbone=Swin-Base, Params=95.7M, Input size=480 x 640, Flops=109.3G2025.04 | 54 | — | — | — | — | — | |
| PGDENetModalities=RGB-Depth2023.03 | 53.7 | — | — | — | — | — | |
| PGDENet22Backbone=ResNet-34, Params=100.7M, Input size=480 x 640, Flops=178.8G2025.04 | 53.7 | — | — | — | — | — | |
| DFormer-SBackbone=DFormer-S, Params(M)=18.7, Training dataset=NYUv22024.09 | 53.6 | — | — | — | — | — | |
| DFormer24Backbone=DFormer-Small, Params=18.7M, Input size=480 x 640, Flops=25.6G2025.04 | 53.6 | — | — | — | — | — | |
| FRNet22Backbone=ResNet-34, Params=85.5M, Input size=480 x 640, Flops=115.6G2025.04 | 53.6 | — | — | — | — | — | |
| InvPT2024.03 | 53.56 | — | — | — | — | — | |
| TokenFusion22Backbone=MiT-B2, Params=26.0M, Input size=480 x 640, Flops=55.2G2025.04 | 53.3 | — | — | — | — | — | |
| InverseFormBackbone=ResNet-101, Decoder=DeepLab-V3+2021.04 | 53.1 | 78.1 | — | — | — | — | |
| Omnivore22Backbone=Swin-Small, Params=51.3M, Input size=480 x 640, Flops=59.8G2025.04 | 52.7 | — | — | — | — | — | |
| Stitch FusionModel=Specialist Model, Resolution=512 x 5122026.03 | 52.64 | — | — | — | — | — | |
| CAINet2024.01 | 52.6 | — | — | — | 65.9 | — | |
| CEN20Backbone=ResNet-152, Params=133.9M, Input size=480 x 640, Flops=664.4G2025.04 | 52.5 | — | — | — | — | — | |
| SA-GateModal=RGB-D2023.12 | 52.4 | — | — | — | — | — | |
| SA-Gate2020.07 | 52.4 | 77.9 | — | — | — | — | |
| SA-GateBackbone=ResNet-1012021.04 | 52.4 | 77.9 | — | — | — | — | |
| SA-GateModalities=RGB-Depth2023.03 | 52.4 | — | — | — | — | — | |
| SA-Gate2022.03 | 52.4 | 77.9 | — | — | — | — | |
| SA-Gate2024.01 | 52.4 | — | — | — | — | — | |
| SA-Gate20Backbone=ResNet-101, Params=110.9M, Input size=480 x 640, Flops=193.7G2025.04 | 52.4 | — | — | — | — | — | |
| NANetModal=RGB-D2023.12 | 52.3 | — | — | — | — | — | |
| NANetModalities=RGB-Depth2023.03 | 52.3 | — | — | — | — | — | |
| NANet2022.03 | 52.3 | 77.9 | — | — | — | — | |
| Gemini FusionModel=Specialist Model, Resolution=512 x 5122026.03 | 52.18 | — | — | — | — | — | |
| DFormer24Backbone=DFormer-Tiny, Params=6.0M, Input size=480 x 640, Flops=11.7G2025.04 | 51.8 | — | — | — | — | — | |
| CEN20Backbone=ResNet-101, Params=118.2M, Input size=480 x 640, Flops=618.7G2025.04 | 51.7 | — | — | — | — | — | |
| ECGFNet2024.01 | 51.5 | — | — | — | 65.2 | — | |
| ShapeConvModal=RGB-D2023.12 | 51.3 | — | — | — | — | — | |
| ShapeConvModalities=RGB-Depth2023.03 | 51.3 | — | — | — | — | — | |
| ShapeConv2022.03 | 51.3 | 76.4 | — | — | — | — | |
| ShapeConv21Backbone=ResNext-101, Params=86.8M, Input size=480 x 640, Flops=124.6G2025.04 | 51.3 | — | — | — | — | — | |
| SGNetModal=RGB-D2023.12 | 51.1 | — | — | — | — | — | |
| SGNetModalities=RGB-Depth2023.03 | 51.1 | — | — | — | — | — | |
| CENModality=RGB+D, Backbone=ResNet-101, MAdds (G)=618.32022.03 | 51.1 | — | — | — | — | — | |
| SGNet2022.03 | 51.1 | 76.8 | — | — | — | — | |
| SGNet20Backbone=ResNet-101, Params=64.7M, Input size=480 x 640, Flops=108.5G2025.04 | 51.1 | — | — | — | — | — | |
| DynMM-bModality=RGB+D, Backbone=ResNet-50, MAdds (G)=52.22022.03 | 51 | — | — | — | — | — | |
| EMSANet22Backbone=ResNet-34, Params=46.9M, Input size=480 x 640, Flops=45.4G2025.04 | 51 | — | — | — | — | — | |
| Malleable 2.5DBackbone=ResNet-1012021.04 | 50.9 | 76.9 | — | — | — | — | |
| ESANetModality=RGB+D, Backbone=ResNet-50, MAdds (G)=56.92022.03 | 50.5 | — | — | — | — | — | |
| Pattern-Affinitive Propagation (Ours-ResNet50)Data Modality=RGB, Backbone=ResNet-502019.06 | 50.4 | 76.2 | — | — | 62.5 | — | |
| PAP2020.07 | 50.4 | 76.2 | — | — | — | — | |
| PAP-NetBackbone=ResNet-502021.04 | 50.4 | 76.2 | — | — | — | — | |
| SA-GateModality=RGB+D, Backbone=ResNet-50, MAdds (G)=147.62022.03 | 50.4 | — | — | — | — | — | |
| PAD-NetData Modality=RGB2019.06 | 50.2 | 75.2 | — | — | 62.3 | — | |
| PADNet2020.07 | 50.2 | 75.2 | — | — | — | — | |
| PAD-NetBackbone=ResNet-502021.04 | 50.2 | 75.2 | — | — | — | — | |
| RDFNet2024.01 | 50.1 | — | — | — | 62.8 | — | |
| DynMM-aModality=RGB+D, Backbone=ResNet-50, MAdds (G)=43.42022.03 | 49.9 | — | — | — | — | — | |
| Omnivore22Backbone=Swin-Tiny, Params=29.1M, Input size=480 x 640, Flops=32.7G2025.04 | 49.7 | — | — | — | — | — | |
| UMTInitialization=ImageNet Pre-train2021.06 | 49.14 | — | — | — | — | 0.66 | |
| RDF-101Modal=RGB-D2023.12 | 49.1 | — | — | — | — | — | |
| RDF-101Backbone=ResNet-1012020.07 | 49.1 | 75.6 | — | — | — | — | |
| RDF-1012022.03 | 49.1 | 75.6 | — | — | — | — | |
| RTFNet2024.01 | 49.1 | — | — | — | 64.8 | — |