Summator-Subtractor Network: Modeling spatialand channel differences for Change Detection

Here, we provide the pytorch implementation of the paper: Summator-Subtractor Network: Modeling spatial and channel differences for Change Detection.

For more ore information, please see our published paper at IEEE TGRS.

Requirements

Python 3.8.2
pytorch 1.10.1
torchvision 0.11.2
einops  0.6.1

Train

You can find the training script run_cd.sh in the folder scripts. You can run the script file by sh scripts/run_cd.sh in the command environment.

The detailed script file run_cd.sh is as follows:

gpus=0
checkpoint_root=checkpoints 
data_name=LEVIR  # dataset name 

img_size=256
batch_size=8
lr=0.01
max_epochs=200  #training epochs
net_G=base_transformer_pos_s4_dd8 # model name
lr_policy=linear

split_train=train  # training txt
split_val=val  #validation txt
project_name=CD_${net_G}_${data_name}_b${batch_size}_lr${lr}_${split}_${split_val}_${max_epochs}_${lr_policy}

python main_cd.py --img_size ${img_size} --checkpoint_root ${checkpoint_root} --lr_policy ${lr_policy} --split ${split} --split_val ${split_val} --net_G ${net_G} --gpu_ids ${gpus} --max_epochs ${max_epochs} --project_name ${project_name} --batch_size ${batch_size} --data_name ${data_name}  --lr ${lr}

Evaluate

You can find the evaluation script eval.sh in the folder scripts. You can run the script file by sh scripts/eval.sh in the command environment.

The detailed script file eval.sh is as follows:

gpus=0
data_name=LEVIR # dataset name
net_G=base_transformer_pos_s4_dd8_dedim8 # model name 
split=test # test.txt
project_name=SSCD_LEVIR # the name of the subfolder in the checkpoints folder 
checkpoint_name=best_ckpt.pt # the name of evaluated model file 

python eval_cd.py --split ${split} --net_G ${net_G} --checkpoint_name ${checkpoint_name} --gpu_ids ${gpus} --project_name ${project_name} --data_name ${data_name}

Dataset Preparation

Data structure

"""
Change detection data set with pixel-level binary labels；
├─A
├─B
├─label
└─list
"""

A: images of t1 phase;

B:images of t2 phase;

label: label maps;

list: contains train.txt, val.txt and test.txt, each file records the image names (XXX.png) in the change detection dataset.

Data Download

LEVIR-CD: https://justchenhao.github.io/LEVIR/

WHU-CD: https://study.rsgis.whu.edu.cn/pages/download/building_dataset.html

DSIFN-CD: https://github.com/GeoZcx/A-deeply-supervised-image-fusion-network-for-change-detection-in-remote-sensing-images/tree/master/dataset

GOOGLE-CD: https://drive.usercontent.google.com/download?id=1DAlxuqalNIPopt-WgtDmCYO98_jWM3ER&export=download&authuser=0

Citation

If you use this code for your research, please cite our paper:

@ARTICLE{10380600,
  author={Wang, Leiquan and Fang, Ye and Li, Zhongwei and Wu, Chunlei and Xu, Mingming and Shao, Mingwen},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={Summator–Subtractor Network: Modeling Spatial and Channel Differences for Change Detection}, 
  year={2024},
  volume={62},
  number={},
  pages={1-12},
  doi={10.1109/TGRS.2024.3349638}}