search

Li-Chongyi / Zero-DCE

Zero-DCE code and model
774 stars 187 forks source link

[Question] If trained on custom data-set, what are the requirements(or suggestions)? #26

Closed CaptainEven closed 3 years ago

CaptainEven commented 3 years ago

Thank you for your excellent work! I want to train Zero-DCE on a custom data-set, what should be considered for a good-performance model?

Li-Chongyi commented 3 years ago

Hi,

You may tune the weights of different losses. Its important to balance the losses.

Even @.***> 于2021年5月17日周一 下午2:31写道：

Thank you for your excellent work! I want to train Zero-DCE on a custom data-set, what should be considered for a good-performance model?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/Li-Chongyi/Zero-DCE/issues/26, or unsubscribe https://github.com/notifications/unsubscribe-auth/AIEVQEIKYY447QEA2DMXHU3TOCZ43ANCNFSM447ZFMYQ .

CaptainEven commented 3 years ago

@Li-Chongyi Thanks for your advice! I found severe mosaic blocks in some enhanced output like:

1_HCA10315_A103026_20210516200100_939_1 1_HCA10315_A103026_20210516200100_939_1

CaptainEven commented 3 years ago

Do you have any advice for mitigating such mosaic?

Li-Chongyi commented 3 years ago

Hi, This kind of curve-based method cannot denoise and remove mosaic. So, you may need a preprocessing algorithm. Commonly, these issues should be solved in the ISP pipeline. Btw, our method is for academic use only. Thanks!

Even @.***> 于2021年5月18日周二 上午9:40写道：

Do you have any advice for mitigating such mosaic?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/Li-Chongyi/Zero-DCE/issues/26#issuecomment-842758307, or unsubscribe https://github.com/notifications/unsubscribe-auth/AIEVQEPMLKP2FX2DQMXBXYLTOHARXANCNFSM447ZFMYQ .

CaptainEven commented 3 years ago

Thanks for replying!