EGENet: Edge-Guided Enhancement Network for Building Change Detection of Remote Sensing Images with a Hybrid CNN-Transformer Architecture

Here, we provide the pytorch implementation of the paper: Edge-Guided Enhancement Network for Building Change Detection of Remote Sensing Images with a Hybrid CNN-Transformer Architecture. For more information, please see our paper at IGARSS.

Network Architecture

1. Environment setup

conda create -n egenet python=3.8
conda activate egenet
pip install -r requirements.txt

2. Dataset Preparation

Data structure

"""
Change detection data set with pixel-level binary labels；(WHU256 or LEVIR-CD-256)
├─A
├─B
├─label
├─label_edge
└─list
"""

A: images of t1 phase;

B:images of t2 phase;

label: label maps;

label_edge: using the Canny edge detection operator on theusing the Canny edge detection operator on the label maps;

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

3. Download the datesets:

• LEVIR-CD: LEVIR-CD
• WHU-CD: WHU-CD

and put them into data directory. In addition, the processed whu dataset can be found in the release.

4. Download the models (pretrain models):

• resnet18

and put it into pretrain directory.

5. Train & Test

The following are scripts for different networks and datasets, run according to your needs

python main_cd.py --project_name 'EGENet_LEVIR' --data_name 'LEVIR' --net_G 'EGENet'
python main_cd.py --project_name 'EGENet_WHU' --data_name 'WHU' --net_G 'EGENet'
python main_cd.py --project_name 'EGCTNet_LEVIR' --data_name 'LEVIR' --net_G 'EGCTNet'
python main_cd.py --project_name 'EGCTNet_WHU' --data_name 'WHU' --net_G 'EGCTNet'
python main_cd.py --project_name 'BIT_LEVIR' --data_name 'LEVIR' --net_G 'BIT' --loss 'ce'
python main_cd.py --project_name 'BIT_WHU' --data_name 'WHU' --net_G 'BIT' --loss 'ce'
python main_cd.py --project_name 'ICIF_Net_LEVIR' --data_name 'LEVIR' --net_G 'ICIF_Net' --loss 'ce'
python main_cd.py --project_name 'ICIF_Net_WHU' --data_name 'WHU' --net_G 'ICIF_Net' --loss 'ce'
python main_cd.py --project_name 'ChangeFormer_LEVIR' --data_name 'LEVIR' --net_G 'ChangeFormer' --loss 'ce'
python main_cd.py --project_name 'ChangeFormer_WHU' --data_name 'WHU' --net_G 'ChangeFormer' --loss 'ce'

References

Appreciate the work from the following repositories:

• EGCTNet (Our EGENet is implemented on the code provided in this repository)