Panoptic Segmentation Network Based on Recursive Layer Aggregation

Most of the existing panoptic segmentation algorithms have the problems of high computational cost and insufficient accuracy. EfficientPS（Efficient Panoptical Segmentation） network provides a solution, but the performance still has room for improvement. Therefore, a panoptic segmentation network based on recursive layer aggregation structure（RLAPS, Recursive Layer Aggregation Panoptic Segmentation） was proposed to improve the panoptic segmentation effect. The structure of the backbone network was changed to the residual network with recursive layer aggregation structure, which can better reuse the features extracted from the shallow network without increasing redundancy and has the better ability to learn the structural information in the image. In addition, the channel diversification module is added after the structure of feature pyramid networks（FPN） in the backbone network, which makes up for the problem that the convolution network focuses on a few main channel features with the deepening of layers, and enhances the ability of the backbone network to extract features. The first part of the semantic segmentation added the branch of the combination of jump connection and global attention module, so that the extracted features were related to the global information. Experiments show that the panoptic segmentation quality of this network is improved by 0.9% compared with EfficientPS. At the same time, the segmentation accuracy of foreground instance target and background filled area is improved by 0.5% and 1.3% respectively.