MISGNet: Multilevel Intertemporal Semantic Guidance Network for Remote Sensing Images Change Detection

Authors: Binge Cui, Chenglong Liu, Haojie Li, Jianzhi Yu

We are excited to provide the PyTorch implementation of the paper: MISGNet: Multilevel Intertemporal Semantic Guidance Network for Remote Sensing Images Change Detection.

🛎️Updates

🎉 Exciting News! 🎉 Nov. 26th, 2024，We are thrilled to announce that MISGNet has been accepted for publication in IEEE JSTARS! 🎉 You can check it out here.

If you find the project interesting, please consider giving it a ⭐️ star ⭐️ to support us! Stay tuned for more updates! 🔥

🔭Overview

🌟Semantics Guidance Module (SGM)

🌟Multilevel Difference Aggregation Module

📝 Requirements

To run this project, you need to install the following dependencies:

albumentations>=1.3.0
numpy>=1.20.2
opencv_python>=4.7.0.72
opencv_python_headless>=4.7.0.72
Pillow>=9.4.0
Pillow>=9.5.0
scikit_learn>=1.0.2
torch>=1.9.0
torchvision>=0.10.0

🛠️ Installation

To clone this repository and get started, use the following commands:

git clone https://github.com/JackLiu-97/MISGNet.git
cd MISGNet

🗝️Quick Start

1. Download Pretrained Models

You can download the pretrained models for the following datasets:

• LEVIR-CD: Baidu Drive, code: itrs
• SYSU-CD: Baidu Drive, code: itrs

After downloading, place the model in the output folder.

2. Run the Demo

Once the model is in place, you can run the demo to get started:

python demo.py --ckpt_url ${model_path} --data_path ${sample_data_path} --out_path ${out_data_path}

🚀 Training

To train a model from scratch, run the following command:

python train.py --data_path ${train_data_path} --val_path ${val_data_path} --lr ${lr} --batch_size ${batch_size}

🔍 Evaluation

To evaluate a model on the test subset, use:

python predict.py --ckpt_url ${model_path} --data_path ${test_data_path}

⚗️Result

We have also provided inference results for easier comparison with our model:

• LEVIR-CD: Baidu Drive, code: itrs
• SYSU-CD: Baidu Drive, code: itrs

📚 Supported Datasets

WHU-CD: The WHU Building Change Detection Dataset contains two aerial images taken at different time phases, with significant land-use changes over a $20.5km^2$ area.

• Size: $32570\times15354$
• Resolution: $0.2m$
• Train/Validation/Test Split: $6096/762/762$

LEVIR-CD: Consists of $637$ very high-resolution ($0.5$m/pixel) Google Earth image patch pairs.

• Size: $1024\times1024$ pixels
• Time Span: 5-14 years
• Contains 31,333 individual change building instances.

SYSU-CD: Contains $20000$ pairs of $0.5$m aerial images from Hong Kong, covering changes such as new urban buildings, suburban expansion, groundwork, vegetation change, road expansion, and sea construction.

• Image Size: $256\times256$

Dataset	Name	Link
LEVIR-CD building change detection dataset	LEVIR-CD	website
SYSU-CD building change detection dataset	SYSU-CD	website
WHU building change detection dataset	WHU-CD	website

📄 License

The code is released for non-commercial and research purposes only. For commercial use, please contact the authors.