The repository is related to the paper Contour Completion with Deep Structure Priors. With no pre-training and no mask (information about which parts of an image are missing) and starting from random noise, our model can complete the missing parts of a 2D shape. Several modules were borrowed from Deep image prior.
First install requirements:
pip install -r requiremetns
You can run a demo in demo.ipynb
notebook.