A Fully Convolutional Tri-branch Network (FCTN) for Domain Adaptation

TL;DR

This paper introduces a novel fully convolutional tri-branch network for domain adaptation in urban scene segmentation, which iteratively refines pseudo labels to improve cross-domain performance, achieving state-of-the-art results.

Contribution

The paper presents a new tri-branch network architecture that progressively learns target-specific features through pseudo-labeling and re-training for domain adaptation.

Findings

Achieves state-of-the-art performance on GTA to Cityscapes adaptation.

Outperforms previous methods significantly.

Demonstrates effective pseudo-labeling and re-training strategy.

Abstract

A domain adaptation method for urban scene segmentation is proposed in this work. We develop a fully convolutional tri-branch network, where two branches assign pseudo labels to images in the unlabeled target domain while the third branch is trained with supervision based on images in the pseudo-labeled target domain. The re-labeling and re-training processes alternate. With this design, the tri-branch network learns target-specific discriminative representations progressively and, as a result, the cross-domain capability of the segmenter improves. We evaluate the proposed network on large-scale domain adaptation experiments using both synthetic (GTA) and real (Cityscapes) images. It is shown that our solution achieves the state-of-the-art performance and it outperforms previous methods by a significant margin.