LGPNet

This is a official implementation for our paper: "Building Change Detection for VHR Remote Sensing Images via Local-Global Pyramid Network and Cross-Task Transfer Learning Strategy" has been published on IEEE Transactions on Geoscience and Remote Sensing by Tongfei Liu, Maoguo Gong, Di Lu, Qingfu Zhang, Hanhong Zheng, Fenlong Jiang, and Mingyang Zhang.

Result Preview

Model Download Link

Link: https://pan.baidu.com/s/15_gvp9seONXpHK90LDJN0Q
Password：yv9e

Requirements

python=3.7.10
pytorch=1.9
opencv-python=4.1.0.25
scikit-image=0.14.2
scikit-learn=0.24.1
tqdm

Usage

Train

1. Load the pretrain model path
2. Load the train and test(val) data path
   python BCD_train.py

Test

1. Load the model path
2. Load the test data path
   python BCD_test.py

Example(WHU)

BCD_train.py

data_path = "./samples/WHU/train"  
epochs=110, batch_size=4, lr=0.0001, ModelName='DPN_Inria', is_Transfer= True  
BFENet.load_state_dict(torch.load('Pretrain_BFE_'+ModelName+'_model_epoch75_mIoU_89.657089.pth', map_location=device))  

run: python BCD_train.py

BCD_test.py

BFENet.load_state_dict(torch.load('BestmIoU_BFE_DPN_epoch91_mIoU_91.864527.pth', map_location=device))
BCDNet.load_state_dict(torch.load('BestmIoU_BCD_DPN_epoch91_mIoU_91.864527.pth', map_location=device))

tests1_path = glob.glob('./samples/WHU/test/image1/*.tif')  
tests2_path = glob.glob('./samples/WHU/test/image2/*.tif')  
label_path = glob.glob('./samples/WHU/test/label/*.tif')  

run: python BCD_test.py

Note
We recommend importing the complete data set before executing training, otherwise, an error will be reported during the training process (Error: the denominator cannot be zero in the evaluation index calculation).

Get results (Visual and Quantitative)

Visual result: ./samples/WHU/test/results
Quantitative result: ./test_acc.txt

Citation

If you find our work useful for your research, please consider citing our paper:

@article{liu2021building,  
  title={Building Change Detection for VHR Remote Sensing Images via Local-Global Pyramid Network and Cross-Task Transfer Learning Strategy},  
  author={Liu, Tongfei and Gong, Maoguo and Lu, Di and Zhang, Qingfu and Zheng, Hanhong and Jiang, Fenlong and Zhang, Mingyang},  
  journal={IEEE Transactions on Geoscience and Remote Sensing},  
  year={2021},  
  pages={1-17},  
  doi={10.1109/TGRS.2021.3130940},  
  publisher={IEEE}  
}  

Acknowledgement

This code is heavily borrowed from the PSPNet[1], PANet[2], DANet[3], etc. We are very grateful for the contributions of these papers and related codes. In addtion, we are also very grateful for the outstanding contributions of the publicly available datasets (WHU and LEVIR) of the papers [4] and [5].

[1] Zhao H, Shi J, Qi X, et al. Pyramid scene parsing network[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 2881-2890.  
[2] Li H, Xiong P, An J, et al. Pyramid attention network for semantic segmentation[J]. arXiv preprint arXiv:1805.10180, 2018.  
[3] Fu J, Liu J, Tian H, et al. Dual attention network for scene segmentation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019: 3146-3154.  
[4] Ji S, Wei S, Lu M. Fully convolutional networks for multisource building extraction from an open aerial and satellite imagery data set[J]. IEEE Transactions on Geoscience and Remote Sensing, 2018, 57(1): 574-586.  
[5] Chen H, Shi Z. A spatial-temporal attention-based method and a new dataset for remote sensing image change detection[J]. Remote Sensing, 2020, 12(10): 1662.  

Contact us

If you have any problme when running the code, please do not hesitate to contact us. Thanks.
E-mail: liutongfei_home@hotmail.com
Date: Nov 7, 2021