Scaling Wide Residual Networks for Panoptic Segmentation

Benchmark Model Rank Results
panoptic-segmentation-on-cityscapes-testPanoptic-DeepLab (SWideRNet [1, 1, 4.5], Mapillary, multi-scale)PQ: 67.8
panoptic-segmentation-on-cityscapes-valPanoptic-DeepLab (SWideRNet [1, 1, 4.5], Mapillary Vistas, multi-scale)PQ: 69.6mIoU: 85.3AP: 46.8
panoptic-segmentation-on-cityscapes-valPanoptic-DeepLab (SWideRNet [1, 1, 4.5], Mapillary Vistas, single-scale)PQ: 68.5mIoU: 84.6AP: 42.8
panoptic-segmentation-on-coco-test-devPanoptic-DeepLab (SWideRNet-[1, 1, 4], multi-scale)PQ: 46.5PQst: 38.2PQth: 52.0
panoptic-segmentation-on-mapillary-valPanoptic-DeepLab (SWideRNet-(1, 1, 4.5), multi-scale)PQ: 44.8mIoU: 60.0PQst: 51.9PQth: 39.3