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

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 mainly promote the research of learning semantic representation, thereby ignoring the model transferability. In this paper, we introduce the Land-cOVEr Domain Adaptive semantic segmentation (LoveDA) dataset to advance semantic and transferable learning. The LoveDA dataset contains 5987 HSR images with 166768 annotated objects from three different cities. 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 eleven semantic segmentation methods and eight UDA methods. Some exploratory studies including multi-scale architectures and strategies, additional background supervision, and pseudo-label analysis were also carried out to address these challenges. The code and data are available at https://github.com/Junjue-Wang/LoveDA.

Corporate Governance Data Storage Tort Liability for Artificial Intelligence and Emotion Recognition

“…We augment the high-resolution raw images with random cropping. iSAID [49] and loveDA [44] focus on semantic understanding in aerial images, with 23262 and 2520 training images for 15 and 6 semantic categories respectively.…”

Prefab-Net: Segmentation of prefabricated building from satellite imagery with deep neural network

“…Remote sensing images have significant differences compared to other categories of images due to the diversity of sensors and are also more sensitive to temporal and spatial effects [28]. Existing remote sensing datasets such as LoveDA [29], ISPRS dataset, and iSAID [30] usually suffer from regional limitations, narrow coverage of categories, and small data volume [31]. It is an intuitive idea to address the shortcomings of training data by using large datasets like ImageNet for pre-training [32].…”

SinoLC-1: the first 1-meter resolution national-scale land-cover map of China created with the deep learning framework and open-access data

“…Based on the context, the quality and resolutions of the published land-cover products have been through the trends of coarse to fine (Cao & Huang, 2022). Nevertheless, due to the low orbit of the VHR image-captured platforms, the corresponding VHR land-cover products generally have a smaller coverage that is insufficient to cover entire China (Wang et al, 2021). Furthermore, even if the national-scale VHR imagery can be obtained by combining different image sources, the immense data volumes, laborious annotations, and onerous processes are still the main obstacles for the national-scale VHR land-cover mapping.…”

“…The current VHR land-cover datasets are generally regional scale (typically covering a few cities/provinces and smaller than a national scale) because of the limitation of the coverage and temporal resolutions of VHR imagery. For example, Wang et al (2021) utilized imagery from airborne cameras and Google Earth to create a 0.3-meter-resolution regional-scale dataset, covering 536.15 km 2 areas (including Nanjing, Changzhou, and Wuhan in China). Huang et al (2020) proposed a 2.1-meter-resolution regional-scale land-cover dataset, called Hi-ULCM, covering 42 major cities in China.…”