Question about sizes

Want to understand the training process better? Got tips for which model to use and when? This is the place for you


Forum rules

Read the FAQs and search the forum before posting a new topic.

This forum is for discussing tips and understanding the process involved with Training a Faceswap model.

If you have found a bug are having issues with the Training process not working, then you should post in the Training Support forum.

Please mark any answers that fixed your problems so others can find the solutions.

Post Reply
User avatar
susanvilla
Posts: 2
Joined: Thu Jun 27, 2024 10:10 am

Question about sizes

Post by susanvilla »

Hello,
I want train using the Lightweight model
As you know, Lightweight model is 64px input, 64px output
When I extract the face from the video, I wonder if there was a difference between 64px and 128px, 256px.
Hope that someone can answer me.
Thanks in advance!

User avatar
torzdf
Posts: 2734
Joined: Fri Jul 12, 2019 12:53 am
Answers: 159
Has thanked: 139 times
Been thanked: 639 times

Re: Question about sizes

Post by torzdf »

All extracts are done at 'Head' Centering, but training is normally done at Face centering. A head centered extracted face is 1.33x larger than a face centered image.

Assuming that you are training at face centered (the default), then you can multiply the model output size by 1.33 to get the optimal extract size, so 64px x 1.33 = 85.33. In this case I would round up to 96px and extract at that.

However, as long as the extracted face patch is larger than your model output size, there is no real benefit to changing the extract size from the default.

My word is final

User avatar
susanvilla
Posts: 2
Joined: Thu Jun 27, 2024 10:10 am

Re: Question about sizes

Post by susanvilla »

All extracts are done at 'Head' Centering, but training is normally done at Face centering. A head centered extracted face is 1.33x larger than a face centered image.

Assuming that you are training at face centered (the default), then you can multiply the model output size by 1.33 to get the optimal extract size, so 64px x 1.33 = 85.33. In this case I would round up to 96px and extract at that.
However, as long as the extracted face patch is larger than your model output size, there is no real benefit to changing the extract size from the default.

Thanks for your answer. I got it.

Last edited by torzdf on Tue Jul 09, 2024 3:16 pm, edited 2 times in total.
Post Reply