Panoptic Part Segmentation on Cityscapes Panoptic Parts (val)
63.3PartPQ (All Parts)JPPF
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| JPPFBackbone=EfficientNet-L2, Scale=Multi-Scale2023.11 | 63.3 | — | — | — | 52.2 | 67.2 | — | — | — | — | 68.2 | |
| PPF++Backbone=ConvNext, Scale=Multi-Scale2023.11 | 63.1 | — | — | — | 46.4 | 69.1 | — | — | — | — | 66.5 | |
| PPF++Backbone=Swin, Scale=Multi-Scale2023.11 | 62.3 | — | — | — | 46 | 68.2 | — | — | — | — | 65.7 | |
| Panoptic-PartFormerBackbone=Swin-base2022.04 | 61.9 | 66.6 | 65.1 | 67.2 | 45.6 | 68 | — | — | — | — | — | |
| PPFBackbone=Swin, Scale=Multi-Scale2023.11 | 61.9 | — | — | — | 45.6 | 68 | — | — | — | — | 65.3 | |
| JPPFScale=Multi-Scale, Run time [ms]=2498, Model size [M]=44.192022.12 | 61.8 | — | — | — | 50.8 | 65.7 | 81.8 | 41.3 | 78.5 | 99.5 | — | |
| JPPFBackbone=EfficientNet-B5, Scale=Multi-Scale2023.11 | 61.8 | — | — | — | 50.8 | 65.7 | — | — | — | — | 68.2 | |
| HRNet-OCR & PolyTransform + BSANetPanoptic seg. method=HRNet-OCR & PolyTransform, Part seg. method=BSANet, Backbone (Panoptic)=HRNetv2-W48, Backbone (Part)=ResNet1012022.04 | 61.4 | 66.2 | 64.2 | 67 | 45.8 | 67 | — | — | — | — | — | |
| JPPFBackbone=EfficientNet-L2, Scale=Single-Scale2023.11 | 61.4 | — | — | — | 49.5 | 65.7 | — | — | — | — | 67.3 | |
| EfficientPS + BSANetPanoptic seg. method=EfficientPS, Part seg. method=BSANet, Backbone (Panoptic)=Efficient Net, Backbone (Part)=ResNet1012022.04 | 60.2 | 65 | 64.2 | 65.2 | 46.1 | 65.2 | — | — | — | — | — | |
| BaselineScale=Multi-Scale2022.12 | 60.2 | — | — | — | 46.1 | 65.2 | 80.3 | 39.7 | 76 | — | — | |
| BaselineBackbone=EfficientNet / ResNet101, Scale=Multi-Scale2023.11 | 60.2 | — | — | — | 46.1 | 65.2 | — | — | — | — | — | |
| JPPFScale=Single-Scale, Run time [ms]=397, Model size [M]=44.192022.12 | 59.6 | — | — | — | 47.7 | 63.8 | 80.5 | 37.9 | 77 | 99.33 | — | |
| JPPFBackbone=EfficientNet-B5, Scale=Single-Scale2023.11 | 59.6 | — | — | — | 47.7 | 63.8 | — | — | — | — | 66.1 | |
| PPF++Backbone=ResNet50, Scale=Multi-Scale2023.11 | 59.2 | — | — | — | 42.5 | 65.1 | — | — | — | — | 62.1 | |
| BaselineScale=Single-Scale, Status=Reproduced baseline, Run time [ms]=871, Model size [M]=68.82022.12 | 57.7 | — | — | — | 44.2 | 62.5 | 79.7 | 36.6 | 74.5 | 98.84 | — | |
| Baseline*Backbone=EfficientNet-B5, Scale=Single-Scale2023.11 | 57.7 | — | — | — | 44.2 | 62.5 | — | — | — | — | — | |
| Panoptic-PartFormerBackbone=ResNet502022.04 | 57.4 | 61.6 | 60 | 62.2 | 43.9 | 62.2 | — | — | — | — | — | |
| PPFBackbone=ResNet50, Scale=Multi-Scale2023.11 | 57.4 | — | — | — | 43.9 | 62.2 | — | — | — | — | 60.5 | |
| DeepLabv3+ & Mask R-CNN + DeepLabv3+Panoptic seg. method=DeepLabv3+ & Mask R-CNN, Part seg. method=DeepLabv3+, Backbone (Panoptic)=ResNet50, Backbone (Part)=Xception-652022.04 | 56.9 | 61 | 58.7 | 61.9 | 43 | 61.9 | — | — | — | — | — | |
| UPSNet + DeepLabv3+Panoptic seg. method=UPSNet, Part seg. method=DeepLabv3+, Backbone (Panoptic)=ResNet50, Backbone (Part)=ResNet502022.04 | 55.1 | 59.1 | 57.3 | 59.7 | 42.3 | 59.7 | — | — | — | — | — |