ADeLA: Automatic Dense Labeling with Attention for Viewpoint Adaptation in Semantic Segmentation

TL;DR

This paper introduces ADeLA, an unsupervised domain adaptation approach for semantic segmentation that leverages attention-based view transformation and functional label hallucination to handle viewpoint-induced content shifts without relying on traditional domain alignment.

Contribution

It proposes a novel viewpoint adaptation method that does not require aligning image statistics and uses attention mechanisms for generalization, outperforming existing state-of-the-art techniques.

Findings

Outperforms state-of-the-art domain adaptation methods.

Effectively handles viewpoint changes without image alignment.

Utilizes attention for generalization in semantic image hallucination.

Abstract

We describe an unsupervised domain adaptation method for image content shift caused by viewpoint changes for a semantic segmentation task. Most existing methods perform domain alignment in a shared space and assume that the mapping from the aligned space to the output is transferable. However, the novel content induced by viewpoint changes may nullify such a space for effective alignments, thus resulting in negative adaptation. Our method works without aligning any statistics of the images between the two domains. Instead, it utilizes a view transformation network trained only on color images to hallucinate the semantic images for the target. Despite the lack of supervision, the view transformation network can still generalize to semantic images thanks to the inductive bias introduced by the attention mechanism. Furthermore, to resolve ambiguities in converting the semantic images to semantic labels, we treat the view transformation network as a functional representation of an unknown mapping implied by the color images and propose functional label hallucination to generate pseudo-labels in the target domain. Our method surpasses baselines built on state-of-the-art correspondence estimation and view synthesis methods. Moreover, it outperforms the state-of-the-art unsupervised domain adaptation methods that utilize self-training and adversarial domain alignment. Our code and dataset will be made publicly available.