Fully Convolutional Networks for Panoptic Segmentation

In this paper, we present a conceptually simple, strong, and efficient framework for panoptic segmentation, called Panoptic FCN. Our approach aims to represent and predict foreground things and background stuff in a unified fully convolutional pipeline... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Panoptic Segmentation Cityscapes val Panoptic FCN* (ResNet-FPN) PQ 61.4 # 7
PQst 66.6 # 3
PQth 54.8 # 10
Panoptic Segmentation COCO minival Panoptic FCN* (ResNet-50-FPN) PQ 44.3 # 1
SQ 80.7 # 1
RQ 53 # 1
PQth 50 # 1
SQth 83.4 # 1
RQth 59.3 # 1
PQst 35.6 # 1
SQst 76.7 # 1
RQst 43.5 # 1
Panoptic Segmentation COCO test-dev Panoptic FCN*++ (DCN-101-FPN) PQ 47.5 # 9
PQst 38.2 # 3
PQth 53.7 # 12
Panoptic Segmentation Mapillary val Panoptic FCN* (ResNet-FPN) PQ 36.9 # 5
PQth 32.9 # 2
PQst 42.3 # 2

Methods used in the Paper


METHOD TYPE
Max Pooling
Pooling Operations
Convolution
Convolutions
FCN
Semantic Segmentation Models