Panoptic Part Segmentation on Cityscapes PPS (val)
64.8PartPQTAPPS
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| TAPPSBackbone=Swin-B, Pre-training=ImageNet + COCO2024.06 | 64.8 | 68 | 53 | 69 | 68 | |
| Panoptic-PartFormer++Backbone=ConvNeXt-B, Pre-training=ImageNet + COCO2024.06 | 63.1 | 68.2 | 46.4 | 69.1 | — | |
| SegFormer-B5 + CondInst + BPRBackbone=MiT-B5 + ResNet-50, Pre-training=ImageNet + COCO (instance seg pre-trained)2024.06 | 62.5 | — | 48.6 | 67.5 | — | |
| Panoptic-PartFormer++Backbone=Swin-B, Pre-training=ImageNet + COCO2024.06 | 62.3 | 68 | 46 | 68.2 | — | |
| Panoptic-PartFormerBackbone=Swin-B, Pre-training=ImageNet + COCO2024.06 | 62 | 66.6 | 45.6 | 67.8 | 59 | |
| Panoptic-PartFormerBackbone=Swin-base, Param(M)=100.32, GFlops=408.522022.04 | 61.9 | 66.6 | — | — | — | |
| HRNet(OCR) + PolyTransform + BSANetParam(M)=>181, GFlops=>11542022.04 | 61.4 | 66.2 | — | — | — | |
| TAPPSBackbone=ResNet-50, Pre-training=ImageNet + COCO2024.06 | 61.3 | 64.4 | 48.9 | 65.7 | 66.9 | |
| JPPFBackbone=EfficientNet-B5, Pre-training=ImageNet2024.06 | 59.6 | — | 47.7 | 63.8 | — | |
| TAPPSBackbone=ResNet-50, Pre-training=ImageNet2024.06 | 59.3 | 62.4 | 48.7 | 63.1 | 66.8 | |
| Panoptic-PartFormer++Backbone=ResNet-50, Pre-training=ImageNet + COCO2024.06 | 59.2 | 63.6 | 42.5 | 65.1 | — | |
| Panoptic-PartFormer++Backbone=ResNet-50, Pre-training=ImageNet2024.06 | 57.5 | 61.6 | — | — | — | |
| Panoptic-PartFormerBackbone=ResNet-50, Pre-training=ImageNet + COCO2024.06 | 57.5 | 61.6 | 43.9 | 62.4 | 60.1 | |
| Panoptic-PartFormerBackbone=ResNet50, Param(M)=37.35, GFlops=185.842022.04 | 57.4 | 61.6 | — | — | — | |
| UPSNet + DeepLabv3+Backbone=ResNet50, Param(M)=>87, GFlops=>8902022.04 | 55.1 | 59.1 | — | — | — | |
| Panoptic-PartFormerBackbone=ResNet-50, Pre-training=ImageNet2024.06 | 54.5 | 57.8 | — | — | — |