search

pfnet / PaintsChainer

line drawing colorization using chainer
https://paintschainer.preferred.tech/
MIT License
3.77k stars 561 forks source link

How to create & learning color hints to model... #144

Open dev6969 opened 5 years ago

dev6969 commented 5 years ago

would like to keep learning the code while continuing to use another network model. But I do not understand how to learn color information in this project ....

  1. I think one of the four input channels is a line channel and the remaining three channels are color information. However, I can not understand how to create color hints from the original image. What I understood = 4ch -> 1ch (line) + 3ch (color hint)[how create color hint!?]

  2. Why does "img2imgDataset.get_example ()" randomly add noise to the image?

taizan commented 5 years ago

1.randomly leak original color information 2.data augmentation for stable training

dev6969 commented 5 years ago

thank you for your reply. !!!!(comments did not come up.) I understood the way. and 

    image1 = np.insert(image1, 1, -512, axis=2)  # -512?!
    image1 = np.insert(image1, 2, 128, axis=2)
    image1 = np.insert(image1, 3, 128, axis=2)

Adjusting the color range in this way for recognize a range does not have color information?

not adjusting the range -1 ~ 1