UA-BCD

The official implementation of "Overcoming the Uncertainty Challenges in Detecting Building Changes from Remote Sensing Images" Paper Link.

We are delighted to share that our paper has been successfully accepted by the ISPRS Journal of Photogrammetry and Remote Sensing (ISPRS 2024).

This repository contains the full implementation of our model, including training, testing, and a large-scale inference framework.

---

📦 Pretrained Backbones

We provide the pretrained backbone PVT-v2-b2 for your convenience.
You can download it via Baidu Disk:

• Download Link
  Code: abmg

---

🏋️‍♀️ Training Instructions

To train the UA-BCD model, follow these steps:

1. Set the hyperparameters for training.

2. Run the following command:

   python train.py --batchsize 32 --data_name LEVIR
   

---

🧪 Testing Instructions

To evaluate the trained UA-BCD model, follow these steps:

1. Ensure the model is properly trained and paths are set.

2. Run the following command:

   python test.py --data_name LEVIR --batchsize 32
   

---

🌍 Large-Scale Inference Instructions

For large-scale applications of UA-BCD, use the following steps:

Specify the paths to the pre-change and post-change images, along with the model paths.

Run the inference script:

 ```bash
 python inference.py --uabcd_model <path_to_uabcd_model> \
                  --eue_model <path_to_eue_model> \
                  --A_path <path_to_pre_change_image> \
                  --B_path <path_to_post_change_image> \
                  --Pos XX --batchsize 64

---

🖼️ Visualizations for Large-Scale Applications

Here are examples of large-scale building change detection results:

---

📜 Citation

If you use our work in your research, please cite:

  @article{li2025overcoming,
    title={Overcoming the uncertainty challenges in detecting building changes from remote sensing images},
    author={Li, Jiepan and He, Wei and Li, Zhuohong and Guo, Yujun and Zhang, Hongyan},
    journal={ISPRS Journal of Photogrammetry and Remote Sensing},
    volume={220},
    pages={1--17},
    year={2025},
    publisher={Elsevier}
  }