SE-PSNet: Silhouette-based Enhancement Feature for Panoptic Segmentation Network

Shuo-En Chang, Yi Chen, Yi-Cheng Yang, En-Ting Lin, Pei-Yung Hsiao, Li-Chen Fu

Abstract: Highlights•The proposed additional silhouette loss improved the segmentation quality.•New mask quality score helps to solve the occlusion problem.•The model has less chance of ignoring small objects.•The model guarantees both the performance and the inference speed.