Unsupervised domain adaptation for semantic segmentation

Abstract

Unsupervised domain adaptation (UDA) has emerged as a vital research area in the field of machine learning and computer vision. It addresses the challenge of adapting models trained on a labeled source domain to perform well on an unlabeled target domain with different distribution characteristics. Unlike traditional supervised learning, UDA aims to bridge the domain gap and transfer knowledge from the source to the target domain without relying on target domain annotations. As one would expect, a typical street scene in the Indian Subcontinent looks very different from a typical street scene in Europe and the USA. Unfortunately, most of the work done in DA and UDA does not show results on any South-East Asian Dataset. Hence, there is a need to both benchmark existing techniques on a dataset from the Asian Continent, and then try to improve the state of the art on UDA in context of an Indian setting, but also generally. This is exactly what this work achieves. There is a further need to develop an analysis toolkit for domain adaptation and semantic segmentation in general which is missing in the current literature, which this work also develops.