This is the source code for the paper:
Luan Dong, Fengling Zheng, Hongxia Chang, Qin Yan. Corner points localization in electronic topographic maps with deep neural networks[J]Earth Science Informatics, 2018,11(1):47-57. https://doi.org/10.1007/s12145-017-0317-3
The provided files are the model(locnet.py), the training and testing program (train.py and test.py respectively). The format of ground-truth information can be figured out from the .pkl file.
- Chainer
- Visdom (for visualizing the training process)
- OpenCV
- Scan your topographic maps. (We use a 200 dpi configuration)
- Using sloth or any other labeling software to label the rectangle objects described in the paper.
- Parse the .json file produced by sloth, and construct a ground-truth file likes the .pkl file in the codebase.