KyanChen/BuildingExtraction

1# STTNet2Paper: Building Extraction from Remote Sensing Images with Sparse Token Transformers31. Prepare Data     4   Prepare data for training, validation, and test phase. All images are with the resolution of $512 \times 512$. Please refer to the directory of **Data**.5  6   For larger images, you can patch the images with labels using **Tools/CutImgSegWithLabel.py**.72. Get Data List    8   Please refer to **Tools/GetTrainValTestCSV.py** to get the train, val, and test csv files.93. Get Imgs Infos     10   Please refer to **Tools/GetImgMeanStd.py** to get the mean value and standard deviation of the all image pixels in training set.114. Modify Model Infos    12   Please modify the model information if you want, or keep the default configuration.135. Run to Train    14   Train the model in **Main.py**.156. [Optional] Run to Test    16   Test the model with checkpoint in **Test.py**.17 18 19We have provided pretrained models on INRIA and WHU Datasets. The pt models are in folder **Pretrain**.20 21If you have any questions, please refer to [our paper](https://www.mdpi.com/2072-4292/13/21/4441) or contact with us by email.22 23```24@Article{rs13214441,25AUTHOR = {Chen, Keyan and Zou, Zhengxia and Shi, Zhenwei},26TITLE = {Building Extraction from Remote Sensing Images with Sparse Token Transformers},27JOURNAL = {Remote Sensing},28VOLUME = {13},29YEAR = {2021},30NUMBER = {21},31ARTICLE-NUMBER = {4441},32URL = {https://www.mdpi.com/2072-4292/13/21/4441},33ISSN = {2072-4292},34DOI = {10.3390/rs13214441}35}36```37