Chainer implementation for Learning to Simplify: Fully Convolutional Networks for Rough Sketch Cleanup

[ishanrai05]

Code of the paper Learning to Simplify: Fully Convolutional Networks for Rough Sketch Cleanup which is tested and trained on custom datasets based on Chainer.

[soskek]

Thank you! The code works in my environment with your uploaded images.
But, somehow the generated images by a trained model are much worse than ones shown in README. I used the default setting: python main.py --train=True -g 0.
Can you implement a good setup as default? Or did I something wrong?

[ishanrai05]

@soskek Hi. Thanks for review. For Chainer Adam seems to do better work than AdaDelta. I have included the default settings as close to those mentioned in the paper. I'll change it to produce good results.