Consistency Regularization for Domain Adaptation

TL;DR

This paper introduces a consistency regularization technique for unsupervised domain adaptation in semantic segmentation, enhancing model performance by modeling inter-pixel relationships, and demonstrates its effectiveness on benchmark datasets.

Contribution

It proposes a novel consistency regularization term for semi-supervised UDA, improving upon existing methods like DAFormer for semantic segmentation tasks.

Findings

Improved mIoU on GTA5 to Cityscapes by 0.8

Enhanced mIoU on SYNTHIA to Cityscapes by 1.2

Effective modeling of inter-pixel relationships boosts adaptation performance

Abstract

Collection of real world annotations for training semantic segmentation models is an expensive process. Unsupervised domain adaptation (UDA) tries to solve this problem by studying how more accessible data such as synthetic data can be used to train and adapt models to real world images without requiring their annotations. Recent UDA methods applies self-learning by training on pixel-wise classification loss using a student and teacher network. In this paper, we propose the addition of a consistency regularization term to semi-supervised UDA by modelling the inter-pixel relationship between elements in networks' output. We demonstrate the effectiveness of the proposed consistency regularization term by applying it to the state-of-the-art DAFormer framework and improving mIoU19 performance on the GTA5 to Cityscapes benchmark by 0.8 and mIou16 performance on the SYNTHIA to Cityscapes benchmark by 1.2.