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

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

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

This is an initial benchmark for Unsupervised Domain Adaptation.

Getting Started

Requirements:

• pytorch >= 1.7.0
• python >=3.6
• pandas >= 1.1.5

Prepare LoveDA Dataset

ln -s </path/to/LoveDA> ./LoveDA

Evaluate CBST Model on the predict set

1. Download the pre-trained weights

2. Move weight file to log directory

mkdir -vp ./log/
mv ./CBST_2Urban.pth ./log/CBST_2Urban.pth

3. Evaluate on Urban test set

bash ./scripts/predict_cbst.sh

Submit your test results on LoveDA Unsupervised Domain Adaptation Challenge and you will get your Test score.

Train CBST Model

From Rural to Urban

bash ./scripts/train_cbst.sh

Eval CBST Model on Urban val set

bash ./scripts/eval_cbst.sh