Semi-supervised Semantic Segmentation Network based on Knowledge Distillation

Abstract

Semantic segmentation of complex traffic scenes is a challenging research topic in the field of computer vision. In order to reduce the dependence of the segmentation model on the pixel-level annotation data of traffic scenes, we propose a semi-supervised semantic segmentation algorithm model based on knowledge distillation. The self-correcting module is used to iteratively optimize the weakly labeled data and generate pseudo-labels. The collaborative learning of multiple students enhances the ability of students to accept potential knowledge online. The proposed method uses the knowledge distillation structure of the teacher-student network to transfer semantic structured information. It solves the problem of insufficient fine label samples in the Cityscapes dataset. The network performance obtained by training with the original label data combined with the pseudo label data can be further improved.