Semi-supervised tooth instance segmentation

Abstract

Tooth segmentation is getting popular with the development of 3D computer vision technology. Current tooth segmentation models rely on large annotations of data which requires great effort from experts and increases the computation cost for training. In this report we proposed to implement mean teacher, a semi-supervised learning framework to train the tooth instance segmentation model. Our experiment results shows that our network can achieve comparable performance with fully supervised network but requires far less data annotation and computation cost.