NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:230
Title:Category Anchor-Guided Unsupervised Domain Adaptation for Semantic Segmentation

The paper presents a novel approach for unsupervised domain adaptation, which employs category-wise feature alignment and self-supervised training with pseudo-features. An active target sample selection strategy is proposed leveraging distance category anchors for pseudo labeling. Overall the approach is clearly presented, convincing, and well supported by empirical evaluation. The reviewers and AC have examined the authors feedback, which satisfactorily addresses the points raised in the reviews. We strongly recommend that the authors incorporate this feedback in the revised paper.