search

DeclanRussell / NvidiaAIDenoiser

A simple implementation of Nvidia's AI denoiser
MIT License
398 stars 52 forks source link

Retraining network #5

Closed jaisonoh closed 6 years ago

jaisonoh commented 6 years ago

Hi, there.

I want to train the network with custom dataset. Is there any way to train again?

Thanks.

Meetem commented 6 years ago

No you can’t. Trained model owned and trained by nvidia, so they only provide functions to use NN as is.

чт, 2 авг. 2018 г. в 5:39, jaisonoh notifications@github.com:

I want to train the network with custom dataset. Is there any way to train again?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/DeclanRussell/NvidiaAIDenoiser/issues/5, or mute the thread https://github.com/notifications/unsubscribe-auth/AHQC8sOE7BG_1CwpjNjdmFqMy5_afQi6ks5uMmZPgaJpZM4Vriac .

DeclanRussell commented 6 years ago

Actually yes you can, heres is a snippet from the OptiX programming guide:

You can also create a custom model by training the denoiser with your own set of images and use the resulting training data in OptiX, but this process is not part of OptiX itself. To learn how to generate your own training data based on your renderer's images you can attend the course Rendered Image Denoising using Autoencoders, which is part of the NVIDIA Deep Learning Institute.

I haven't met anyone yet who has actually managed to create there own training data set yet as it is a very demanding process which requires having a few thousand image pairs (I think nvidia used 20,000). If anyone is seriously interested creating their own training data set I would be happy to add support for it in this application.

jaisonoh commented 6 years ago

Thanks for your answer. I'll try the course.