UPSNet: A Unified Panoptic Segmentation Network

CVPR 2019 Yuwen XiongRenjie LiaoHengshuang ZhaoRui HuMin BaiErsin YumerRaquel Urtasun

In this paper, we propose a unified panoptic segmentation network (UPSNet) for tackling the newly proposed panoptic segmentation task. On top of a single backbone residual network, we first design a deformable convolution based semantic segmentation head and a Mask R-CNN style instance segmentation head which solve these two subtasks simultaneously... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK USES EXTRA
TRAINING DATA
RESULT LEADERBOARD
Panoptic Segmentation Cityscapes val UPSNet (ResNet-101) PQ 60.5 # 8
PQst 63.0 # 7
PQth 57.0 # 6
mIoU 77.8 # 7
AP 37.8 # 5
Panoptic Segmentation Cityscapes val UPSNet (ResNet-50) PQ 59.3 # 10
PQst 62.7 # 9
PQth 54.6 # 11
mIoU 75.2 # 12
AP 33.3 # 11
Panoptic Segmentation Cityscapes val UPSNet (ResNet-101, multiscale) PQ 61.8 # 5
PQst 64.8 # 4
PQth 57.6 # 4
mIoU 79.2 # 5
AP 39.0 # 3
Panoptic Segmentation COCO test-dev UPSNet (ResNet-101-FPN) PQ 46.6 # 3
PQst 36.7 # 2
PQth 53.2 # 6
Panoptic Segmentation Indian Driving Dataset UPSNet PQ 47.1 # 3
Panoptic Segmentation KITTI Panoptic Segmentation UPSNet PQ 39.9 # 3

Methods used in the Paper