search

andreas128 / SRFlow

Official SRFlow training code: Super-Resolution using Normalizing Flow in PyTorch
Other
829 stars 112 forks source link

train error: svd_cuda #21

Open vkl-git opened 3 years ago

vkl-git commented 3 years ago

Hello, thank you for your work.When I tried to train with my own datasets, I used the SRFlow_DF2K_4x model, but during the training process, "svd_cuda: the updating process of SBDSDC did not converage (error:11)" error appeared.Have you ever encountered the same reason in your training?Is there a solution? image

flybiubiu commented 3 years ago

which cuda and pytorch version?

vkl-git commented 3 years ago

Thank you for your response.  I used CUDA 10. 0 and pytorch1.7. Are there any problems? 

---Original--- From: "flybiubiu"<notifications@github.com> Date: Sat, Feb 20, 2021 22:20 PM To: "andreas128/SRFlow"<SRFlow@noreply.github.com>; Cc: "viki-li"<2368943151@qq.com>;"Author"<author@noreply.github.com>; Subject: Re: [andreas128/SRFlow] train error: svd_cuda (#21)

which cuda and pytorch version?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

flybiubiu commented 3 years ago

I train it with same problem before, But I run setup file all problem are display. If GPU is 3090,cuda version should be cuda 11.0

andreas128 commented 3 years ago

Hi viki-li, this error usually occurs in the back propagation step of the 1x1 convolution, when the learning rate is too high.

Does it help to reduce the learning rate?