Faces for Training Question

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
363LS2GTO
Posts: 30
Joined: Fri Jul 08, 2022 7:06 pm
Has thanked: 1 time
Been thanked: 4 times

Faces for Training Question

Post by 363LS2GTO »

I would like to know if there is any way to view the masks and landmarks for the combined "faces used for training." These images are usually pulled from multiple alignments files and combined into one folder for training.

Although there is no alignments file associated with the faces during the training process, I am guessing that all of this information is already 'baked in' to the file for the purposes of training.

When using "Tools, Alignments, Extract" to pull every N face from the edited alignments file, it seems that random choice will pull a few imperfectly edited faces from the potentially thousands to tens of thousands of images.

A work around would be to do a separate extract of the source (setting N to about 15 for example). This would allow for more precise manual work on a smaller number of images. This is alluded to in the guide, but I wanted to make sure that this is the only way as it adds a lot of extra folders to keep track of in addition to the extra extract and manual work. Is this the usual way for anyone looking for a higher quality training set?

User avatar
torzdf
Posts: 2649
Joined: Fri Jul 12, 2019 12:53 am
Answers: 159
Has thanked: 128 times
Been thanked: 622 times

Re: Faces for Training Question

Post by torzdf »

I'm not really sure what you mean by the first part of the question... You can use the mask tool to output a visual representation of the masks stored.

EEN doesn't pull out a random choice, it extracts every N frames (so if you set EEN as 10, it will extract faces from every 10th frame, assuming that frame has faces).

TBH, I don't normally spend too much time worrying about masks for training sets. I find they are 99% correct (assuming a correctly aligned face), which I think is within tolerance for training a model.

My word is final

Locked