Scripts description

1. preprocess.py: Training dataset augmentation offline. Support up-sampling strategy based on positive and negative.
2. rooftop_dataset.py:Create training, validation and test dataset in pytorch format.
3. train.py: Training of rooftop extraction model.
4. predict.py: Batch prediction based on ensemble strategy and expansion prediction method.
5. predict_trt.py: Batch prediction based on TensorRT.
6. evaluation.py: Evaluation of dataset.
7. deeplab_resnet.py: Deeplabv3+ model based on resnet backbone.
8. deeplab_xception.py: Deeplabv3+ model based on xception backbone.
9. ensemble_config.json: Configuration of ensemble model. (The path of checkpoints)

Model weights download:

1. (Baidu Netdisk) URL: https://pan.baidu.com/s/1nnn8P4zqEUf2U9HXt2QPwQ Code: w7t2

Citation: Zhang, Z., Qian, Z., Zhong, T., Chen, M., Zhang, K., Yang, Y., Zhu, R., Zhang, F., Zhang, H., Zhou, F. and Yu, J., 2022. Vectorized rooftop area data for 90 cities in China. Scientific Data, 9(1), pp.1-12.