Improving the Student-Teacher Approach for Semi-Supervised Semantic Segmentation

Abstract

We build off of the work of (Tarvainen et. al. 2017) and apply the mean-teacher approach to semi-supervised semantic segmentation. We propose sliding-window teacher evaluation and the ensembling of two mean-teachers with different update rates. Only the latter modification improved results over the baseline mean-teacher. See linked one-minute video for more details.

Tarvainen, Antti, and Harri Valpola. “Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results.” Advances in neural information processing systems 30 (2017).