Deep learning-based total electron content (TEC) map completion

A deep convolutional generative adversarial network with Poisson blending (DCGAN-PB) for TEC map completion.

Prerequisites

This project is run in Python 3 and mainly uses Tensorflow under the guidance of Anaconda to manage the environment. You can use platform other than anaconda at your convenience. The detailed anaconda package list can be found in the file pkgs.yml.

Database

The training data is included here. Please download and unzip to the preferred location. For the mask files, you can get from link1 and link2. They are some examples of the mask files, you are welcome to make your own mask files.

To run the project

After configuring the environment, put your parameters such as learning rate, epochs, etc in train-dcgan.py.

python train-dcgan.py

It will run the project.

After getting the trained model, to inpaint a single mask file.

python3 ./complete.py [path/to/mask/file] --inmaskfile [path/to/mask/file] --nIter 4000 --outDir [path/to/save/output] --lam 0.05

You can tune the parameters either in the command line or in complete.py. After which, the inpainted results would be saved to the output path.

Authors

Y. Pan ([email protected])

M.W. Jin

Y. Deng

S.R. Zhang

References

Taehoon Kim

Brandon Amos

Yeh (semantic inpainting)