Unsupervised Domain Adaptation by Optical Flow Augmentation in Semantic Segmentation

TL;DR

This paper proposes using dense optical flow augmentation to improve unsupervised domain adaptation in semantic segmentation, reducing the need for real-life image labels by bridging the domain gap.

Contribution

It introduces optical flow augmentation into domain adaptation for semantic segmentation, enhancing performance without additional real-world labels.

Findings

Optical flow augmentation improves domain adaptation performance.

Flow maps help preserve geometric information across domains.

The method reduces reliance on labeled real-world datasets.

Abstract

It is expensive to generate real-life image labels and there is a domain gap between real-life and simulated images, hence a model trained on the latter cannot adapt to the former. Solving this can totally eliminate the need for labeling real-life datasets completely. Class balanced self-training is one of the existing techniques that attempt to reduce the domain gap. Moreover, augmenting RGB with flow maps has improved performance in simple semantic segmentation and geometry is preserved across domains. Hence, by augmenting images with dense optical flow map, domain adaptation in semantic segmentation can be improved.