Semantic Segmentation on ADE20K 15 (val)
62mIoU (ss)Mask2Former
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Mask2FormerBackbone=BEIT-3, #Params=1.9B, Crop Size=896x896, Extra Data=true, Training Protocol=Individual Training2022.11 | 62 | 62.8 | |
| Mask DINOBackbone=Swin-L, #Params=223M, Crop Size=896x896, Extra Data=true, Training Protocol=Individual Training2022.11 | 59.5 | 60.8 | |
| Mask2FormerBackbone=ViT-Adapter-L, #Params=568M, Crop Size=896x896, Extra Data=true, Training Protocol=Individual Training2022.11 | 59.4 | 60.5 | |
| UPerNetBackbone=SwinV2-G, #Params=>3B, Crop Size=896x896, Extra Data=true, Training Protocol=Individual Training2022.11 | 59.3 | 59.9 | |
| OneFormerBackbone=DINAT-L, #Params=223M, Crop Size=640x640, Extra Data=false, Training Protocol=Joint Training2022.11 | 58.3 | 58.4 | |
| OneFormerBackbone=DINAT-L, #Params=223M, Crop Size=1280x1280, Extra Data=false, Training Protocol=Joint Training2022.11 | 58.2 | 58.7 | |
| OneFormerBackbone=DINAT-L, #Params=223M, Crop Size=896x896, Extra Data=false, Training Protocol=Joint Training2022.11 | 58.1 | 58.6 | |
| UPerNetBackbone=ViT-Adapter-L, #Params=571M, Crop Size=640x640, Extra Data=false, Training Protocol=Individual Training2022.11 | 58 | 58.4 | |
| OneFormerBackbone=Swin-L, #Params=219M, Crop Size=896x896, Extra Data=false, Training Protocol=Joint Training2022.11 | 57.4 | 58.3 | |
| OneFormerBackbone=ConvNeXt-XL, #Params=372M, Crop Size=640x640, Extra Data=false, Training Protocol=Joint Training2022.11 | 57.4 | 58.8 | |
| MSFaPN-Mask2FormerBackbone=SeMask Swin-L†, Crop Size=640x640, Extra Data=false, Training Protocol=Individual Training2022.11 | 57 | 58.2 | |
| OneFormerBackbone=Swin-L, #Params=219M, Crop Size=640x640, Extra Data=false, Training Protocol=Joint Training2022.11 | 57 | 57.7 | |
| OneFormerBackbone=Swin-L, #Params=219M, Crop Size=1280x1280, Extra Data=false, Training Protocol=Joint Training2022.11 | 57 | 57.7 | |
| OneFormerBackbone=ConvNeXt-L, #Params=220M, Crop Size=640x640, Extra Data=false, Training Protocol=Joint Training2022.11 | 56.6 | 57.4 | |
| FaPN-Mask2FormerBackbone=Swin-L, Crop Size=640x640, Extra Data=false, Training Protocol=Individual Training2022.11 | 56.4 | 57.7 | |
| SeMask Mask2FormerBackbone=SeMask Swin-L†, Crop Size=640x640, Extra Data=false, Training Protocol=Individual Training2022.11 | 56.4 | 57.5 | |
| Mask2Former-SemanticBackbone=Swin-L, #Params=216M, Crop Size=640x640, Extra Data=false, Training Protocol=Individual Training2022.11 | 56.1 | 57.3 | |
| kMaX-DeepLabBackbone=ConvNeXt-L†, #Params=232M, Crop Size=1281x1281, Extra Data=false, Training Protocol=Individual Training2022.11 | 55.2 | — | |
| kMaX-DeepLabBackbone=ConvNeXt-L†, #Params=232M, Crop Size=641x641, Extra Data=false, Training Protocol=Individual Training2022.11 | 54.8 | — | |
| Mask2Former-PanopticBackbone=Swin-L, #Params=216M, Crop Size=640x640, Extra Data=false, Training Protocol=Individual Training2022.11 | 54.5 | — | |
| UPerNetBackbone=FD-SwinV2-G, #Params=>3B, Crop Size=896x896, Extra Data=true, Training Protocol=Individual Training2022.11 | — | 61.4 |