Home

Awesome

Introduction

The Pytorch implementation for: "LightCDNet: Lightweight Change Detection Network Based on VHR Images" (IEEE GRSL' 2023)

You can get a PDF version of our paper here: Google Drive

DSFM

LightCDNet

Abstract

LEVIR-CD

MethodCrop SizeLr schd#Param (M)MACs (G)PrecisionRecallF1-ScoreIoU
LightCDNet-small256x256400000.351.6591.3689.8190.5782.77
LightCDNet-base256x256400001.323.2292.1290.4391.2783.94
LightCDNet-large256x256400002.825.9492.4390.4591.4384.21

Notice

The code has been integrated into Open-CD, welcome to use it!

Environment installation

This project is implemented based on Open-CD, please refer to the installation method of Open-CD 0.x version.

Usage

(eg: LightCDNet-small)

Train:

python tools/train.py configs/lightcdnet/lightcdnet_s_256x256_40k_levircd.py --work-dir ./exp/lightcdnet_s_levir_workdir --gpu-id 0 --seed 602

Eval:

python tools/test.py configs/lightcdnet/lightcdnet_s_256x256_40k_levircd.py ./exp/lightcdnet_s_levir_workdir/latest.pth --eval mFscore mIoU

Visualization:

python tools/test.py configs/lightcdnet/lightcdnet_s_256x256_40k_levircd.py ./exp/lightcdnet_s_levir_workdir/latest.pth --format-only --eval-options "imgfile_prefix=tmp_infer"

Citation

If you find this project useful in your research, please consider cite:

@ARTICLE{10214556,
  author={Xing, Yuanjun and Jiang, Jiawei and Xiang, Jun and Yan, Enping and Song, Yabin and Mo, Dengkui},
  journal={IEEE Geoscience and Remote Sensing Letters}, 
  title={LightCDNet: Lightweight Change Detection Network Based on VHR Images}, 
  year={2023},
  volume={20},
  number={},
  pages={1-5},
  doi={10.1109/LGRS.2023.3304309}}