LoveDA: A Remote Sensing Land-Cover Dataset for Domain Adaptation Semantic Segmentation

Junjue Wang, Zhuo Zheng, Ailong Ma, Xiaoyan Lu, Yanfei Zhong

Keywords: Land-cover mapping, Unsupervised domain adaptation, Semantic segmentation

TL;DR: A remote sensing land-cover domain adaptation semantic segmentation dataset is proposed with three considerable challenges in large-scale mapping: multi-scale objects, complex background samples, and inconsistent class distributions.

Abstract: Deep learning approaches have shown promising results in remote sensing high spatial resolution (HSR) land-cover mapping. However, urban and rural scenes can show completely different geographical landscapes, and the inadequate generalizability of these algorithms hinders city-level or national-level mapping. Most of the existing HSR land-cover datasets only focus on improvement of the semantic segmentation in one domain (urban or rural), thereby ignoring the model transferability. In this paper, we introduce the Land-cOVEr Domain Adaptation semantic segmentation (LoveDA) dataset to promote large-scale land-cover mapping. The LoveDA dataset contains 3338 aerial images with 86,516 annotated objects for seven common land-cover categories. Compared to the existing datasets, the LoveDA dataset encompasses two domains (urban and rural), which brings considerable challenges due to the: 1) multi-scale objects; 2) complex background samples; and 3) inconsistent class distributions. The LoveDA dataset is suitable for both land-cover semantic segmentation and unsupervised domain adaptation (UDA) tasks. Accordingly, we benchmarked the LoveDA dataset on nine semantic segmentation methods and eight UDA methods. Some exploratory studies were also carried out to find alternative ways to address these challenges. The code and data will be available at: https://github.com/Junjue-Wang/LoveDA.

URL: https://drive.google.com/drive/folders/1BJa48DE5GiFUTf981zIUno8wF7Z6jV3D?usp=sharing