Curriculum Domain Adaptation for Semantic Segmentation of Urban Scenes

TL;DR

This paper introduces a curriculum domain adaptation method for semantic segmentation of urban scenes, effectively reducing the domain gap between synthetic and real images by learning global and local label distributions first.

Contribution

It proposes a novel curriculum-style learning approach that minimizes domain mismatch in urban scene segmentation by inferring and regularizing global and local label distributions.

Findings

Outperforms baseline methods on two datasets

Effective across different backbone networks

Shows significant improvement in segmentation accuracy

Abstract

During the last half decade, convolutional neural networks (CNNs) have triumphed over semantic segmentation, which is one of the core tasks in many applications such as autonomous driving. However, to train CNNs requires a considerable amount of data, which is difficult to collect and laborious to annotate. Recent advances in computer graphics make it possible to train CNNs on photo-realistic synthetic imagery with computer-generated annotations. Despite this, the domain mismatch between the real images and the synthetic data cripples the models' performance. Hence, we propose a curriculum-style learning approach to minimize the domain gap in urban scenery semantic segmentation. The curriculum domain adaptation solves easy tasks first to infer necessary properties about the target domain; in particular, the first task is to learn global label distributions over images and local distributions over landmark superpixels. These are easy to estimate because images of urban scenes have strong idiosyncrasies (e.g., the size and spatial relations of buildings, streets, cars, etc.). We then train a segmentation network while regularizing its predictions in the target domain to follow those inferred properties. In experiments, our method outperforms the baselines on two datasets and two backbone networks. We also report extensive ablation studies about our approach.