Panoptic Segmentation On Cityscapes Val

评估指标

AP
PQ
PQst
PQth
mIoU

评测结果

各个模型在此基准测试上的表现结果

Paper TitleRepository
OneFormer (ConvNeXt-L, single-scale, 512x1024, Mapillary Vistas-pretrained)48.770.174.164.684.6OneFormer: One Transformer to Rule Universal Image Segmentation
Panoptic-DeepLab (SWideRNet [1, 1, 4.5], Mapillary Vistas, multi-scale)46.869.6--85.3Scaling Wide Residual Networks for Panoptic Segmentation-
OneFormer (ConvNeXt-L, single-scale)46.568.51--83.0OneFormer: One Transformer to Rule Universal Image Segmentation
Axial-DeepLab-XL (Mapillary Vistas, multi-scale) 44.268.5--84.6Axial-DeepLab: Stand-Alone Axial-Attention for Panoptic Segmentation
Panoptic-DeepLab (SWideRNet [1, 1, 4.5], Mapillary Vistas, single-scale)42.868.5--84.6Scaling Wide Residual Networks for Panoptic Segmentation-
OneFormer (ConvNeXt-XL, single-scale)46.768.4--83.6OneFormer: One Transformer to Rule Universal Image Segmentation
kMaX-DeepLab (single-scale)44.068.4--83.5kMaX-DeepLab: k-means Mask Transformer
AFF-Base (single-scale, point-based Mask2Former)46.267.771.562.583.0AutoFocusFormer: Image Segmentation off the Grid
OneFormer (DiNAT-L, single-scale)45.667.6--83.1OneFormer: One Transformer to Rule Universal Image Segmentation
EfficientPS43.567.570.363.282.1EfficientPS: Efficient Panoptic Segmentation
DiNAT-L (Mask2Former)44.567.2--83.4Dilated Neighborhood Attention Transformer
OneFormer (Swin-L, single-scale)45.667.2--83.0OneFormer: One Transformer to Rule Universal Image Segmentation
AFF-Small (single-scale, point-based Mask2Former)44.266.970.861.582.2AutoFocusFormer: Image Segmentation off the Grid
Mask2Former (Swin-L)43.666.6--82.9Masked-attention Mask Transformer for Universal Image Segmentation
EfficientPS (Cityscapes-fine)39.164.967.761.090.3EfficientPS: Efficient Panoptic Segmentation
CMT-DeepLab (MaX-S, single-scale, IN-1K)-64.6--81.4CMT-DeepLab: Clustering Mask Transformers for Panoptic Segmentation
Panoptic-DeepLab (X71)38.564.1--81.5Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation
Mask2Former + Intra-Batch Supervision (ResNet-50)-62.467.354.7-Intra-Batch Supervision for Panoptic Segmentation on High-Resolution Images
COPS (ResNet-50)34.162.167.255.179.3Combinatorial Optimization for Panoptic Segmentation: A Fully Differentiable Approach
AdaptIS (ResNeXt-101)36.362.064.458.779.2AdaptIS: Adaptive Instance Selection Network-
0 of 36 row(s) selected.
Panoptic Segmentation On Cityscapes Val | SOTA | HyperAI超神经