Combinatorial Optimization for Panoptic Segmentation: A Fully Differentiable Approach

NeurIPS 2021  ·  Ahmed Abbas, Paul Swoboda ·

We propose a fully differentiable architecture for simultaneous semantic and instance segmentation (a.k.a. panoptic segmentation) consisting of a convolutional neural network and an asymmetric multiway cut problem solver. The latter solves a combinatorial optimization problem that elegantly incorporates semantic and boundary predictions to produce a panoptic labeling. Our formulation allows to directly maximize a smooth surrogate of the panoptic quality metric by backpropagating the gradient through the optimization problem. Experimental evaluation shows improvement by backpropagating through the optimization problem w.r.t. comparable approaches on Cityscapes and COCO datasets. Overall, our approach shows the utility of using combinatorial optimization in tandem with deep learning in a challenging large scale real-world problem and showcases benefits and insights into training such an architecture.

PDF Abstract NeurIPS 2021 PDF NeurIPS 2021 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Panoptic Segmentation Cityscapes test COPS (ResNet-50) PQ 60 # 8
Panoptic Segmentation Cityscapes val COPS (ResNet-50) PQ 62.1 # 19
PQst 67.2 # 8
PQth 55.1 # 14
mIoU 79.3 # 19
AP 34.1 # 24
Panoptic Segmentation COCO test-dev COPS (ResNet-50) PQ 38.5 # 35
PQst 34.8 # 25
PQth 41.0 # 33

Methods


No methods listed for this paper. Add relevant methods here