Status after 900k + Iterations

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
sniperpike
Posts: 1
Joined: Thu Sep 30, 2021 10:24 am

Status after 900k + Iterations

Post by sniperpike »

Hi there,

Im fairly new to this, it is my 5. attempt, but I hope I can get some feedback.
This time I think I have some good image sources. I've got 3000+ images of face B, that I would like to swap to Face A.
Face B has a lot of different expressions, from different angles, and are all sharp with no blur.

I've just passed 900K Iterations in training, but some (Not all) of face B images are still very blurry. Please see below:

Image

Could I expect theese images to get better, if I continue training beyond 1000K of iterations or should they already been better due to the high amount of Iterations?

I use "Original" trainer and have set Batch Size to 16.

Sincerely,
Sniperpike

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

Re: Status after 900k + Iterations

Post by torzdf »

I would say your B data is not varied enough. Ultimately the angles/expressions in the swap are not matching up, which tends to be a sign of not having enough variety of data for the NN to be able to resolve the differences.

My word is final

Locked