List your company
ResearchFormerly WizWand

Semantic Segmentation on SUN RGB-D

53mIoUTokenFusion
28.0434.524147.48May 20, 2017Nov 17, 2018May 16, 2020Nov 14, 2021May 14, 2023Nov 10, 2024May 11, 2026
Updated 5mo ago

Evaluation results

MethodLinks
TokenFusionInputs=RGB+D, Variant=S2022.045384.764.1
GT DepthArchitecture=GeminiFusion2026.0452.782.8—
SigmaBackbone=VMamba-S, Params=69.8, Input Size=480 x 6402024.0452.4——
CMNeXtBackbone=MiT-B4, Params=119.6, Input Size=530 x 7302024.0451.9——
FRNetBackbone=ResNet-34, Params=85.5, Input Size=530 x 7302024.0451.8——
FRNetBackbone=ResNet-34, Params=85.5M2026.0551.8——
TokenFusionInputs=RGB+D, Variant=Ti2022.0451.48463.3
GT DepthArchitecture=DFormer2026.0451.283.4—
DFormerBackbone=DFormer-Base, Params=29.5M2026.0551.2——
CENBackbone=ResNet-152, Params=133.9, Input Size=530 x 7302024.0451.1——
MultiMAEBackbone=ViT-B, Params=95.2, Input Size=640 x 6402024.0451.1——
CENInputs=RGB+D2022.0451.183.563.2
CENBackbone=ResNet-152, Params=133.9M2026.0551.1——
MultiMAEBackbone=ViT-B, Params=95.2M2026.0551.1——
PGDENetBackbone=ResNet-34, Params=100.7, Input Size=530 x 7302024.0451——
TokenFusionBackbone=MiT-B3, Params=45.9, Input Size=530 x 7302024.0451——
PGDENetBackbone=ResNet-34, Params=100.7M2026.0551——
TokenFusionBackbone=MiT-B3, Params=45.9M2026.0551——
RelFlexformer (w/ DFormer)Backbone=DFormer-Base, Params=29.5M2026.0551——
TokenFusionBackbone=MiT-B2, Params=26.0, Input Size=530 x 7302024.0450.3——
TokenFusionBackbone=MiT-B2, Params=26.0M2026.0550.3——
CENBackbone=ResNet-101, Params=118.2, Input Size=530 x 7302024.0450.2——
CENBackbone=ResNet-101, Params=118.2M2026.0550.2——
SigmaBackbone=VMamba-T, Params=48.3, Input Size=480 x 6402024.0450——
DFormerBackbone=DFormer-Small, Params=18.7M2026.0550——
CMXBackbone=MiT-B2, Params=66.6M2026.0549.7——
PDCNetBackbone=ResNet-101, Input Size=480 x 4802024.0449.6——
SA-GateBackbone=ResNet-101, Params=110.9, Input Size=530 x 7302024.0449.4——
SA-GateBackbone=ResNet-101, Params=110.9M2026.0549.4——
AsymFormerBackbone=MiT-B0+ConvNeXt-Tiny, Params=33.0M2026.0549.1——
ConcatInputs=RGB+D, Variant=S2022.044983.562
DFormerBackbone=DFormer-Tiny, Params=6.0M2026.0548.8——
SGNetBackbone=ResNet-101, Params=64.7, Input Size=530 x 7302024.0448.6——
ShapeConvBackbone=ResNext-101, Params=86.8, Input Size=530 x 7302024.0448.6——
SGNetBackbone=ResNet-101, Params=64.7M2026.0548.6——
ShapeConvBackbone=ResNet-101, Params=86.8M2026.0548.6——
Performer (w/ DFormer)Backbone=DFormer-Base, Params=29.5M2026.0548.5——
EMSANetBackbone=ResNet-34, Params=46.9, Input Size=530 x 7302024.0448.4——
EMSANetBackbone=ResNet-34, Params=46.9M2026.0548.4——
ESANetBackbone=ResNet-34, Params=31.2, Input Size=480 x 6402024.0448.2——
ESANetBackbone=ResNet-34, Params=31.2M2026.0548.2——
ACNetBackbone=ResNet-50, Params=116.6, Input Size=530 x 7302024.0448.1——
w/o fusionInputs=RGB, Variant=S2022.0448.182.961.3
ACNetBackbone=ResNet-50, Params=116.6M2026.0548.1——
ConcatInputs=RGB+D, Variant=Ti2022.0447.982.861.4
GeomPromptArchitecture=DFormer2026.0447.881.6—
RDFNetInputs=RGB+D2022.0447.781.560.1
DA2Architecture=GeminiFusion2026.0447.781.4—
DA2 [Hypersim]Architecture=DFormer2026.0447.581.8—
RefineNetInputs=RGB2022.044781.157.7
w/o fusionInputs=RGB, Variant=Ti2022.044782.360.6
Metric3Dv2Architecture=DFormer2026.0446.681.8—
Metric3Dv2Architecture=GeminiFusion2026.0446.680.6—
GeomPromptArchitecture=GeminiFusion2026.0446.480.3—
RefineNet-Res152Backbone=ResNet-1522017.0545.980.6—
RefineNet-Res101Backbone=ResNet-1012017.0545.780.4—
SSMAInputs=RGB+D2022.0445.78158.1
OursMode=RGB-D, Backbone=FasterNet-M2026.0345.56——
loop2 (test-aug)Backbone=ResNet-50, Number of loops=2, Depth Usage=predicted depth, test-time augmentation=true2017.0545.180.3—
DA2 [Hypersim]Architecture=GeminiFusion2026.0444.579.6—
loop2Backbone=ResNet-50, Number of loops=2, Depth Usage=predicted depth2017.0544.379.9—
CI-NetMode=RGB, Backbone=ResNet1012026.0344.3——
EMSAFormerMode=RGB-D, Backbone=Swin v22026.0344.13——
loop1 w/ pred-depthBackbone=ResNet-50, Number of loops=1, Depth Usage=predicted depth2017.054479.8—
DA2Architecture=DFormer2026.044480.8—
loop1 w/ gt-depthBackbone=ResNet-50, Number of loops=1, Depth Usage=ground-truth depth2017.0543.979.8—
SSMAMode=RGB-D, Backbone=ResNet502026.0343.9——
RGB-onlyArchitecture=GeminiFusion2026.0443.478.8—
loop1 w/o depthBackbone=ResNet-50, Number of loops=1, Depth Usage=none2017.0543.279.3—
DIANetMode=RGB, Backbone=ResNet502026.0343.1——
X-Decoder (L)Training Data (SEG)=true, Training Data (DET)=false, Training Data (ITP)=true, Evaluation Protocol=Zero-shot, Backbone=Large2023.0343——
w/ pred-depthBackbone=ResNet-50, Depth Usage=predicted depth2017.0542.378.9—
Context2017.0542.378.4—
w/ gt-depthBackbone=ResNet-50, Depth Usage=ground-truth depth2017.0542.278.7—
D-CNNMode=RGB-D, Backbone=VGG-Net2026.0342——
OpenSeeD (L)Training Data (SEG)=true, Training Data (DET)=true, Training Data (ITP)=false, Evaluation Protocol=Zero-shot, Backbone=Large2023.0341.9——
RGB-onlyArchitecture=DFormer2026.0441.778.3—
baselineBackbone=ResNet-502017.0540.277.6—
3DGNNMode=RGB-D, Backbone=VGG-Net2026.0340.2——
OpenSeeD (T)Training Data (SEG)=true, Training Data (DET)=true, Training Data (ITP)=false, Evaluation Protocol=Zero-shot, Backbone=Tiny2023.0339——
SSMAMode=RGB, Backbone=ResNet502026.0338.4——
FuseNetInputs=RGB+D2022.0437.376.348.3
X-Decoder (T)Training Data (SEG)=true, Training Data (DET)=false, Training Data (ITP)=true, Evaluation Protocol=Zero-shot, Backbone=Tiny2023.0334.5——
MSeg (B)Training Data (SEG)=true, Training Data (DET)=false, Training Data (ITP)=false, Evaluation Protocol=Zero-shot, Backbone=Base2023.0329.6——
FCN-32sInputs=RGB2022.042968.441.1