search

tzt101 / MichiGAN

MichiGAN: Multi-Input-Conditioned Hair Image Generation for Portrait Editing (SIGGRAPH 2020)
MIT License
294 stars 39 forks source link

how to generate orientation map #4

Closed ValentiaSulli closed 2 years ago

ValentiaSulli commented 4 years ago

Thanks for your works！ I want to train michigan on my own dataset, but I can not find the codes about extract orientation map. Could you please provide the relevant codes ?

tzt101 commented 4 years ago

This part of the code belongs to Sanp Inc. I have applied for this code and will make it public as soon as I can get it.

utc1205 commented 4 years ago

I see a class "orient" in cal_orient_stroke.py , so i want to ask if it's the correct way to generate orientation map? thanks.

tzt101 commented 4 years ago

This code is used to calculate the orientation map of generated image (see models/networks/loss.py) and is a little different from the code which is used to generate orientation ground truth (we do not smooth the orientation map when calculating orientation loss).

wasupandceacar commented 4 years ago

Could you elaborate what is applied to orientation ground truth when do smoothing?

mirfan899 commented 2 years ago

any update on this?

ValentiaSulli commented 2 years ago

已收到您的邮件，谢谢

JaksonYee commented 2 years ago

Thanks for your works！ I want to train michigan on my own dataset, but I can not find the codes about extract orientation map. Could you please provide the relevant codes ?Thank you very much!

tzt101 commented 2 years ago

You can use cal_orientation.py to generate new orientations on your own dataset.

ValentiaSulli commented 2 years ago

已收到您的邮件，谢谢