search

creotiv / hdrnet-pytorch

Unofficial PyTorch implementation of 'Deep Bilateral Learning for Real-Time Image Enhancement', SIGGRAPH 2017 https://groups.csail.mit.edu/graphics/hdrnet/
224 stars 43 forks source link

the model cannot converge #29

Closed Sissuire closed 2 years ago

Sissuire commented 2 years ago

I came across the problem that the model can not converge..... the training loss is getting bigger and the psnr is decreased every epoch. Any idea about this situation?

creotiv commented 2 years ago

This problem persisted for a few years already, many questionable things were changed but still result the same. I compare it with originals i thinking more the 100 times, and still doesnt see much difference. You can try compare block by block with original tensorflow network to find the problem block, cause that the only thing that we didnt make.

Kindly yours, Andrey Nikishaev

Areas ML/DS/CV/Soft Dev/BizDev/Growth Hacking/Customer Rel/IT LinkedIn http://ua.linkedin.com/in/creotiv GitHub http://github.com/creotiv Slideshare https://www.slideshare.net/anikishaev/ AnimalHelp https://patreon.com/uah Telegram @anikishaev Mobile +380632410666

On Tue, Aug 23, 2022 at 5:35 AM Yongxu Liu @.***> wrote:

I came across the problem that the model can not converge..... the training loss is getting bigger and the psnr is decreased every epoch. Any idea about this situation?

— Reply to this email directly, view it on GitHub https://github.com/creotiv/hdrnet-pytorch/issues/29, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAB5CDOZ5RONBULQ47QA63DV2Q2GXANCNFSM57JX4HPQ . You are receiving this because you are subscribed to this thread.Message ID: @.***>