Semantic Segmentation on Stanford2D3DS (3-fold cross-validation)
61.57mIoUPanoSAMic
Evaluation results
| Method | Links | ||
|---|---|---|---|
| PanoSAMicConfiguration / Modalities=RGB-D-N, # Params (millions)=1912026.01 | 61.57 | 74.04 | |
| PanoSAMicConfiguration / Modalities=RGB-D, # Params (millions)=1842026.01 | 60.9 | 73.95 | |
| OursModal=RGB-H2023.08 | 60.6 | 70.68 | |
| OursModal=RGB-N-H2023.08 | 60.24 | 70.61 | |
| OursModal=RGB-D-H2023.08 | 59.99 | 70.44 | |
| PanoSAMicConfiguration / Modalities=RGB, # Params (millions)=1782026.01 | 59.62 | 74.11 | |
| OursModal=RGB-D-N2023.08 | 59.43 | 69.03 | |
| SFSS-MMSIConfiguration / Modalities=RGB-D-N, # Params (millions)=1232026.01 | 59.43 | 69.03 | |
| OursModal=RGB-N2023.08 | 58.24 | 68.79 | |
| PanoFormerModal=RGB-D2023.08 | 57.03 | 68.08 | |
| PanoFormerConfiguration / Modalities=RGB-D, # Params (millions)=202026.01 | 57.03 | 68.08 | |
| HoHoNetModal=RGB-D2023.08 | 56.73 | 68.23 | |
| HoHoNetConfiguration / Modalities=RGB-D, # Params (millions)=702026.01 | 56.73 | 68.23 | |
| CBFCResolution=512 × 1024, Input=RGB-D2022.07 | 56.7 | 70.8 | |
| CBFCModal=RGB-D2023.08 | 56.7 | 70.8 | |
| CBFCConfiguration / Modalities=RGB-D2026.01 | 56.7 | 70.8 | |
| HoHoNetResolution=1024 × 2048, Input=RGB-D2022.07 | 56.3 | 68.9 | |
| HoHoNetResolution=1024 × 2048, Input=RGB-D, Backbone Type=ResNet backbone w/ high-resolution 360°2020.11 | 56.3 | 68.9 | |
| OursModal=RGB-D2023.08 | 55.49 | 68.57 | |
| SFSS-MMSIConfiguration / Modalities=RGB-D, # Params (millions)=812026.01 | 55.49 | 68.57 | |
| MultiPanoWiseConfiguration / Modalities=RGB2026.01 | 54.6 | — | |
| 360BEVConfiguration / Modalities=RGB-D, # Params (millions)=27.72026.01 | 54.3 | — | |
| CBFCResolution=256 × 512, Input=RGB-D2022.07 | 53.8 | 66.5 | |
| OursModal=RGB2023.08 | 52.87 | 63.96 | |
| SFSS-MMSIConfiguration / Modalities=RGB, # Params (millions)=402026.01 | 52.87 | 63.96 | |
| Tangent ImageInput resolution (s)=10, Input=RGB-D, Base subdivision level (b)=1, Backbone=FCN-ResNet 1012019.12 | 52.5 | 70.1 | |
| TangentModal=RGB-D2023.08 | 52.5 | 70.1 | |
| TangentConfiguration / Modalities=RGB-D2026.01 | 52.5 | 70.1 | |
| PanoFormerModal=RGB2023.08 | 52.35 | 64.31 | |
| PanoFormerConfiguration / Modalities=RGB, # Params (millions)=202026.01 | 52.35 | 64.31 | |
| CBFCResolution=512 × 1024, Input=RGB2022.07 | 52.2 | 65.6 | |
| CBFCModal=RGB2023.08 | 52.2 | 65.6 | |
| CBFCConfiguration / Modalities=RGB2026.01 | 52.2 | 65.6 | |
| SAM3Configuration / Modalities=Open-vocabulary with RGB, # Params (millions)=8502026.01 | 52.05 | 65.02 | |
| Trans4PASS+Modal=RGB2023.08 | 52.04 | 63.98 | |
| Trans4PASS+Configuration / Modalities=RGB, # Params (millions)=392026.01 | 52.04 | 63.98 | |
| HoHoNetResolution=1024 × 2048, Input=RGB2022.07 | 52 | 65 | |
| HoHoNetResolution=1024 × 2048, Input=RGB, Backbone Type=ResNet backbone w/ high-resolution 360°2020.11 | 52 | 65 | |
| HoHoNetModal=RGB2023.08 | 51.99 | 62.97 | |
| HoHoNetConfiguration / Modalities=RGB, # Params (millions)=702026.01 | 51.99 | 62.97 | |
| TangentImgResolution=2048 × 4096, Input=RGB-D2022.07 | 51.9 | 69.1 | |
| TangentImgResolution=2048 × 4096, Input=RGB-D, Backbone Type=ResNet backbone w/ high-resolution 360°2020.11 | 51.9 | 69.1 | |
| PanoFormer2022.03 | 48.9 | 64.5 | |
| CBFCResolution=64 × 128, Input=RGB-D2022.07 | 47.2 | 61.2 | |
| Tangent ImageInput resolution (s)=10, Input=RGB, Base subdivision level (b)=1, Backbone=FCN-ResNet 1012019.12 | 45.6 | 65.2 | |
| TangentImgResolution=2048 × 4096, Input=RGB2022.07 | 45.6 | 65.2 | |
| TangentModal=RGB2023.08 | 45.6 | 65.2 | |
| TangentConfiguration / Modalities=RGB2026.01 | 45.6 | 65.2 | |
| TangentImgResolution=2048 × 4096, Input=RGB, Backbone Type=ResNet backbone w/ high-resolution 360°2020.11 | 45.6 | 65.2 | |
| Tangent ImageInput resolution (s)=7, Input=RGB-D, Base subdivision level (b)=0, Backbone=UNet-style2019.12 | 44.9 | 59.1 | |
| Tangent ImageInput resolution (s)=10, Input=RGB, Base subdivision level (b)=0, Backbone=FCN-ResNet 1012019.12 | 44.3 | 61 | |
| HEAL-SWINparameters=1.5M, resolution_n_side=642023.07 | 44.3 | 61.9 | |
| SWSCNNuse of surface normals=true, breaking equivariance (BE)=false2020.06 | 43.4 | 57.5 | |
| SWSCNNuse of surface normals=true, breaking equivariance (BE)=true2020.06 | 43.4 | 58.7 | |
| SWSCNNResolution=64 × 128, Input=RGB-D2022.07 | 43.4 | 58.7 | |
| Zhang et al.Input resolution (s)=5, Input=RGB-D, Base subdivision level (b)=02019.12 | 43.3 | 58.6 | |
| HexRUNet2020.06 | 43.3 | 58.6 | |
| HexRUNetResolution=64 × 128, Input=RGB-D2022.07 | 43.3 | 58.6 | |
| HoHoNetResolution=256 × 512, Input=RGB-D2022.07 | 43.3 | 53.9 | |
| HoHoNet2022.03 | 43.3 | 53.9 | |
| HexRUNet2023.07 | 43.3 | 58.6 | |
| HexRUNetResolution=64 x 128, Input=RGB-D, Backbone Type=Simple backbone w/ low-resolution 360°2020.11 | 43.3 | 58.6 | |
| HoHoNetResolution=256 x 512, Input=RGB-D, Backbone Type=Simple backbone w/ low-resolution 360°2020.11 | 43.3 | 53.9 | |
| Tangent ImageInput resolution (s)=10, Input=RGB, Base subdivision level (b)=2, Backbone=FCN-ResNet 1012019.12 | 42.7 | 61.5 | |
| OOOPSConfiguration / Modalities=Open-vocabulary with RGB, # Params (millions)=8.72026.01 | 42.6 | — | |
| SWSCNNuse of surface normals=false, breaking equivariance (BE)=false2020.06 | 41.9 | 55.6 | |
| Spin-SphCNN2023.07 | 41.9 | 55.6 | |
| TangentImgResolution=256 × 512, Input=RGB-D2022.07 | 41.8 | 54.9 | |
| TangentImg2022.03 | 41.8 | 54.9 | |
| TangentImgResolution=256 x 512, Input=RGB-D, Backbone Type=Simple backbone w/ low-resolution 360°2020.11 | 41.8 | 54.9 | |
| HoHoNetResolution=64 × 128, Input=RGB-D2022.07 | 40.8 | 52.1 | |
| HoHoNetResolution=64 x 128, Input=RGB-D, Backbone Type=Simple backbone w/ low-resolution 360°2020.11 | 40.8 | 52.1 | |
| SphCNN2020.06 | 40.2 | 52.8 | |
| SphCNN2023.07 | 40.2 | 52.8 | |
| OpenSeeDConfiguration / Modalities=Open-vocabulary with RGB, # Params (millions)=65.42026.01 | 40 | — | |
| CAT-segConfiguration / Modalities=Open-vocabulary with RGB, # Params (millions)=59.52026.01 | 39.6 | — | |
| Cohen et al.Input resolution (s)=5, Input=RGB-D, Base subdivision level (b)=02019.12 | 39.4 | 55.9 | |
| Gauge CNN2020.06 | 39.4 | 55.9 | |
| Gauge NetResolution=64 × 128, Input=RGB-D2022.07 | 39.4 | 55.9 | |
| Gauge CNN2023.07 | 39.4 | 55.9 | |
| Gauge NetResolution=64 x 128, Input=RGB-D, Backbone Type=Simple backbone w/ low-resolution 360°2020.11 | 39.4 | 55.9 | |
| UGSCNN2023.07 | 38.8 | 54.7 | |
| Jiang et al.Input resolution (s)=5, Input=RGB-D, Base subdivision level (b)=52019.12 | 38.3 | 54.7 | |
| Tangent ImageInput resolution (s)=5, Input=RGB-D, Base subdivision level (b)=0, Backbone=UNet-style2019.12 | 38.3 | 50.9 | |
| UGSCNN2020.06 | 38.3 | 54.7 | |
| UGSCNNResolution=64 × 128, Input=RGB-D2022.07 | 38.3 | 54.7 | |
| UGSCNNResolution=64 x 128, Input=RGB-D, Backbone Type=Simple backbone w/ low-resolution 360°2020.11 | 38.3 | 54.7 | |
| TangentImgResolution=64 × 128, Input=RGB-D2022.07 | 37.5 | 50.2 | |
| TangentImgResolution=64 x 128, Input=RGB-D, Backbone Type=Simple backbone w/ low-resolution 360°2020.11 | 37.5 | 50.2 | |
| Tateno et al.Input resolution (s)=7, Input=RGB, Base subdivision level (b)=ERP (equirectangular)2019.12 | 34.6 | — | |
| Lee et al.Input resolution (s)=7, Input=RGB, Base subdivision level (b)=72019.12 | — | 51.4 |