search

Ysz2022 / NeRCo

[ICCV 2023] Implicit Neural Representation for Cooperative Low-light Image Enhancement
https://openaccess.thecvf.com/content/ICCV2023/html/Yang_Implicit_Neural_Representation_for_Cooperative_Low-light_Image_Enhancement_ICCV_2023_paper.html
221 stars 15 forks source link

A question about training results #16

Open xuejuhao opened 7 months ago

xuejuhao commented 7 months ago

Hello, why did the results of training with the same configuration as in the paper turn out to be very poor? The training configuration is as follows: epochs = 300 batchsize = 1 patchsize = 256 Training images consist of 1000 pairs selected from LSRW/nikon/. The testing images are also selected from LSRW/nikon.

May I ask what is the best training configuration. I would be appreciated if you could reply.

The training effect is shown in below with strong drag artifacts. (}J3I@N9V 19 8WTC448MZR

Ysz2022 commented 7 months ago

Maybe you can attempt with bigger patchsize, such as 400 or 512. Empirically, the bigger the patchsize is, the better performance will be.

xuejuhao commented 7 months ago

Maybe you can attempt with bigger patchsize, such as 400 or 512. Empirically, the bigger the patchsize is, the better performance will be.

OK，fine. I will try it. Thank you.