Our current edge detection pipeline does not really make coloring book esque images:
Tweak edge detection to improve the output and consider using alternatives such as CycleGAN. This will be a very open-ended ticket; we just want improvements to the end result.
Branch:
improve-edge-detection
Task at Hand