Shown above are 5 different outputs from the colorization model. Source image (top left) was drawn by my friend @kuronaken.

Using generated images as training data for object detection of traffic signs and cones