DMBNet

Official Pytorch Code base for "A DINOv3-Driven Framework with Multiscale Adaptation and Boundary Refinement for Martian Slope Streak Change Detection" [Paper]

Introduction

We propose a novel pixel-level slope streak change detection approach, named DMBNet. It aims to extract fine change streak information by integrating a VFM (i.e., DINOv3) with boundary refinement and multiscale adaptation. To the best of our knowledge, this is the first work to detect pixel-level slope streak changes using a tailored deep learning network.

Using the code:

The code is stable while using Python 3.9.22, CUDA >=12.1

• Clone this repository:

git clone https://github.com/long123524/DMBNet
cd /DMBNet

To install all the dependencies using conda or pip:

pip install -r requirements.txt

Data Format

Make sure to put the files as the following structure:

inputs
└── <train>
    ├── A
    |   ├── 001.tif
    │   ├── 002.tif
    │   ├── 003.tif
    │   ├── ...
    |
    └── B
    |   ├── 001.tif
    |   ├── 002.tif
    |   ├── 003.tif
    |   ├── ...
    └── label
    |   ├── 001.tif
    |   ├── 002.tif
    |   ├── 003.tif
    |   ├── ...
    └── boundary
    |   ├── 001.tif
    |   ├── 002.tif
    |   ├── 003.tif
    |   ├── ...
    

For testing and validation datasets, the same structure as the above.

Datasets

We are further expanding the Martian slope streak change detection dataset, which will be available for use.

Training

python train.py

Test

python prediction.py

Evaluation

python eval_CD.py

A pretrained weight

A pretrained weight of PVT-V2 on the ImageNet dataset is provided: https://drive.google.com/file/d/1uzeVfA4gEQ772vzLntnkqvWePSw84F6y/view?usp=sharing A pretrained weight of DINOv3 on the LVD dataset is provided: https://drive.google.com/file/d/1q4OoQAyxFd6zZmWQlqIXqV2KFuiVDxzZ/view?usp=sharing

Citation:

If you find this work useful or interesting, please consider citing the following references.

@article{long2026dinov3,
  title={A DINOv3-Driven Framework with Multiscale Adaptation and Boundary Refinement for Martian Slope Streak Change Detection},
  author={Long, Jiang and Liu, Sicong and Wan, Bo and Xu, Yusheng and Xie, Huan and Tong, Xiaohua},
  journal={IEEE Geoscience and Remote Sensing Letters},
  volume={23},
  pages={6022205-6022205},
  year={2026},
  publisher={IEEE}
}
@article{long2026cghu,
  title={CGHU-CD: A contrast-guided hierarchical unsupervised method for detecting Martian anthropogenic changes},
  author={Long, Jiang and Liu, Sicong and Hu, Yuxuan and Wu, Jiayi and Liu, Tao and Xiong, Yonggang and Luo, Kaiyue and Zhang, Jie and Du, Kecheng and Xie, Huan and others},
  journal={International Journal of Applied Earth Observation and Geoinformation},
  volume={153},
  pages={105556},
  year={2026},
  publisher={Elsevier}
}