Unsupervised Domain Adaptation for semantic segmentation has gained immensepopularity since it can transfer knowledge from simulation to real (Sim2Real) In this work, we present a novel two-phase adaptation scheme . In the first step, we exhaustively distill source domain knowledge using supervised loss functions . The second step is video self-training (VST), focusing only on the target data . To construct robust pseudo labels, we exploit the temporal information in the video, which has been rarely explored in the previous image-based methods .We set strong baseline scores on ‘VIPER to CityscapeVPS’ adaptation scenario.We show that our proposals significantly outperform previous image based UDAmethods both on image-level (mIoU) and video-level evaluation metrics. We also show that their proposals significantly out-per-evaluated methods significantly outperformed previous images-based

Author(s) : Inkyu Shin, Kwanyong Park, Sanghyun Woo, In So Kweon

Links : PDF - Abstract

Code :
Coursera

Keywords : video - adaptation - significantly - previous - image -

Leave a Reply

Your email address will not be published. Required fields are marked *