Does it matter if A faces are more than B Faces

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.

Locked
User avatar
MaxHunter
Posts: 193
Joined: Thu May 26, 2022 6:02 am
Has thanked: 177 times
Been thanked: 13 times

Does it matter if A faces are more than B Faces

Post by MaxHunter »

It occurred to me that maybe having too many faces, and not enough balance, could throw everything off. I figured it was better to have a LOT of "A" faces (more than "B") was a good thing, but then I recently started noticing the "halo", double face effect, and no matter how I tried to position the alingments, it wouldn't go away. I was wondering if this is caused from having not enough balance. In this example, my A faces are about 3,000, and my B Faces a little more than 2,000.

How does this unbalance effect the model? Thoughts?

User avatar
torzdf
Posts: 2649
Joined: Fri Jul 12, 2019 12:53 am
Answers: 159
Has thanked: 128 times
Been thanked: 623 times

Re: Does it matter if A faces are more than B Faces

Post by torzdf »

I don't know what the ""halo", double face effect" is.

However, in wider terms, there is no benefit to loading up one side more than the other. There shouldn't be any obvious detrimental effect if they are not matched in volume, as long as the data is similarly distributed in terms of variety.

My word is final

User avatar
MaxHunter
Posts: 193
Joined: Thu May 26, 2022 6:02 am
Has thanked: 177 times
Been thanked: 13 times

Re: Does it matter if A faces are more than B Faces

Post by MaxHunter »

Thanks for the answer. I'll post a picture example of the double face effect.

It's basically where you can see both faces, with the "A" face giving a "halo"/outline effect in the form of a haze on the outside of the the "B" face. I can't tell why. I try to do test conversions as I train and it seems like it happens more as training goes on. Which is why I thought it's possibly from an imbalance of volume, where as the AI is seeing the "A" face more and therefore incorporating that strong presence in the conversion.

Locked