2024.3.18 The paper "Transformer-based Multimodal Change Detection with Multitask Consistency Constraints" is accepted by Information Fusion
2023.9.20 🚧 This repository is currently under active development and construction. We appreciate your interest, but please note that it may not yet be fully functional or complete.
This repository contains the code and resources for multimodal change detection, a technique for detecting changes between different data modalities, such as satellite images and digital surface models (DSMs).
We provide a DSM-to-image multimodal dataset, which detecting multi-category building change from height data and aerial images, called Hi-BCD. You can download the dataset via: ITC-sever with password itc2023 or BaiduNetdisk or GoogleNetdisk.
It is constructed for detecting 2D and 3D changes simultaneously from cross-dimensional modalities. Some samples are as follow:
It includes 1500 pairs of high-resolution tiles emcompassing three cities in the Netherlands. The details of Hi-BCD dataset are as follow: | Attribute | Category | Amsterdam | Rotterdam | Utrecht |
---|---|---|---|---|---|
changed objects | newly-built | 389 | 510 | 458 | |
demolished | 251 | 229 | 187 | ||
changed pixels | amount | 6.625M | 5.139M | 7.73M | |
$prop_{/total}$ | 1.3% | 1.0% | 1.5% | ||
samples | total | 500 | 500 | 500 | |
with change | 40.8% | 34.2% | 43% |
src/
: Source code for multimodal change detection.data/
: Example datasets or links to datasets.models/
: Pre-trained models (if applicable).results/
: Folder to store results.docs/
: Documentation and tutorials.git clone https://github.com/your-username/multimodal-change-detection.git
Install the required dependencies: pip install -r requirements.txt
You can download the model weights via BaiduNetdisk or GoogleNetdisk, including our method and the other compared change detection methods. Next, put the model weights in /weights and run the following command:
bash reproduce.sh
More information about the compared methods can be found in AwesomeChangeDetection.
To train your own multimodal change detection model, follow the instructions in the Training documentation.
To perform change detection on your own data, check out the Testing tutorial.
If you find this code or dataset useful in your research, please consider citing our paper:
Feel free to customize this repository according to your specific project's details and needs. It is easy to use and extend follow the instruction of Pytorch-Lightning.