How does FaceSwap handle changes in the B input library, or training model mid training?

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
CProdDigital
Posts: 14
Joined: Fri Nov 04, 2022 4:22 pm
Has thanked: 1 time

How does FaceSwap handle changes in the B input library, or training model mid training?

Post by CProdDigital »

I'm using footage from a head mounted facial capture camera to train a swap right now, so I have a LOT of dead on frontal frames that I'm training from, like 12,000. This seems like overkill considering its just them talking so there's probably alot of "redundant" frames, and I've got some more dynamic angle footage I'd like to add.

Can I just extract more footage to the B folder and then continue training? would it speed up training to remove some of the 12000 front-facing frames? It seems like as its getting more iterations, the training is slowing down.

Also, I'm using the original trainer, but I heard Dfaker is better if you have a good GPU (I've got a 3080) can I swap Trainer mid training or will that mess things up?

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

Re: How does FaceSwap handle changes in the B input library, or training model mid training?

Post by torzdf »

Yes, you can add/remove faces at will (stop training, change your dataset, then resume. Don't do this whilst training is running). The better distributed the training data across variety/pose/expression/lighting etc, the better it will adapt to each of these conditions.

You cannot change the model used once commenced. This is locked, as the structure of each model is fundamentally different.

If you haven't already, you should ready your guides as a lot of this information (including what can/can't be changed for an existing model) is already covered:
app.php/tag/Guide

My word is final

User avatar
CProdDigital
Posts: 14
Joined: Fri Nov 04, 2022 4:22 pm
Has thanked: 1 time

Re: How does FaceSwap handle changes in the B input library, or training model mid training?

Post by CProdDigital »

I cant make another post rn for some reason, but do you know how to reuse data from B source for a new A video, or vice versa? like a new B souce but keep data from the A vid training?

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

Re: How does FaceSwap handle changes in the B input library, or training model mid training?

Post by torzdf »

I'm not entirely sure I understand what you mean.

You can just point a new model at the same B folder you used before and a new A folder for the new identity.

My word is final

Locked