ykdai / BasicPBC

Official Implementation of "Learning Inclusion Matching for Animation Paint Bucket Colorization"
Other
259 stars 23 forks source link

Can't perform a expected result for my animation clip. #19

Closed slayeryahu closed 5 months ago

slayeryahu commented 6 months ago

B_ U7W}ANKVS4HRY98G)$%4 汤圆动画.zip

1.Critical Issue: The output lines have altered the style of my original input lines, resulting in a jagged appearance. This is intolerable. 2.Color Output: The error rate is exceptionally high. (These sketches are specifically created by my artisitor for the baiscpbc test.)

Sarthak-999 commented 5 months ago

Hey buddy ,could you please tell the steps to run this program on google colab? I am really keen to test it ,and would be grateful towards you if you tell me the steps to run it.Thank you in advance.

slayeryahu commented 5 months ago

Hey buddy ,could you please tell the steps to run this program on google colab? I am really keen to test it ,and would be grateful towards you if you tell me the steps to run it.Thank you in advance.

I ran it on my own PC, with graphic card 2080ti 22g.

Sarthak-999 commented 5 months ago

Can you please tell about the steps to predicate colorized frames?Also if can I do it on google colab?

ykdai commented 5 months ago

Hi, I am a little bit busy recently and sorry for replying late.

  1. I have adjusted the threshold algorithm to decrease the line width. I think the current result is now acceptable. Then, you will have 2 solutions:

    • If you want to keep the original line, you can pull the code and use --keep_line. It will blend the original line to the colorized image.
    • Since the colors are discrete, you can edit it and correct the mis-colorized regions using other softwares like Retas PaintMan or Clip Paint Studio. Then, they all have anti-aliasing tool to remove the jagged appearance. If you are using the Adobe After Effect, you can also try this plugin (OLM Smoother): https://olm.co.jp/rd/technology/tools/?lang=en for anti-aliasing.
  2. Thank you so much for your feedback. I have tested it and the error rate is indeed very high. We are now training a new version of the model, and will try our best to fix these situations in our next version. I will inform you when it is released. Please stay tuned.

slayeryahu commented 5 months ago

Hi, I am a little bit busy recently and sorry for replying late.

  1. I have adjusted the threshold algorithm to decrease the line width. I think the current result is now acceptable. Then, you will have 2 solutions:
  • If you want to keep the original line, you can pull the code and use --keep_line. It will blend the original line to the colorized image.
  • Since the colors are discrete, you can edit it and correct the mis-colorized regions using other softwares like Retas PaintMan or Clip Paint Studio. Then, they all have anti-aliasing tool to remove the jagged appearance. If you are using the Adobe After Effect, you can also try this plugin (OLM Smoother): https://olm.co.jp/rd/technology/tools/?lang=en for anti-aliasing.
  1. Thank you so much for your feedback. I have tested it and the error rate is indeed very high. We are now training a new version of the model, and will try our best to fix these situations in our next version. I will inform you when it is released. Please stay tuned.

Thank you, I look forward to your achievement.