search

royorel / StyleSDF

Other
536 stars 50 forks source link

A question about training #32

Closed Scxw010516 closed 1 year ago

Scxw010516 commented 1 year ago

Hi @royorel, Thank you for your great work! I wonder what is the difference between Training the volume renderer and Training the full pipeline？What is the latter doing more than the former? Looking forward to your reply!

royorel commented 1 year ago

Hi @Scxw010516,

The training is done in two stages. Training the volume renderer is the first phase of training where only the volume renderer is trained in order to generate 64x64 3D consistent faces.

In the second stage, the full pipeline (volume renderer + styleGAN generator) is trained to generate the high resolution 1024x1024 3D aware faces.

It is crucial to train the system in two stages, otherwise it won't converge.

Scxw010516 commented 1 year ago

Hi @Scxw010516,

The training is done in two stages. Training the volume renderer is the first phase of training where only the volume renderer is trained in order to generate 64x64 3D consistent faces.

In the second stage, the full pipeline (volume renderer + styleGAN generator) is trained to generate the high resolution 1024x1024 3D aware faces.

It is crucial to train the system in two stages, otherwise it won't converge.

Hi @royorel , Thank you for your prompt reply！