Iteration count lower and upper limits

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
SmellyCheese
Posts: 11
Joined: Wed Jan 08, 2020 3:35 pm
Has thanked: 7 times

Iteration count lower and upper limits

Post by SmellyCheese »

I know this is a question that is largely dependent on variables.

However, in your experience, is there a lower and upper limit for # of iterations that you need for a decent product? (Specifically with Realface)

Edit: And is there significant difference across different trainers for iterations required?

The pursuit of knowledge and competence

User avatar
bryanlyon
Site Admin
Posts: 793
Joined: Fri Jul 12, 2019 12:49 am
Answers: 44
Location: San Francisco
Has thanked: 4 times
Been thanked: 218 times
Contact:

Re: Iteration count lower and upper limits

Post by bryanlyon »

User avatar
SmellyCheese
Posts: 11
Joined: Wed Jan 08, 2020 3:35 pm
Has thanked: 7 times

Re: Iteration count lower and upper limits

Post by SmellyCheese »

I ask because I'm coming up on 48 hrs with 50k iterations and it doesn't look too great. Blurry. I saw in another post 50k is likely too few.

The pursuit of knowledge and competence

User avatar
Immrspy
Posts: 3
Joined: Fri Jan 17, 2020 4:30 am
Been thanked: 2 times

Re: Iteration count lower and upper limits

Post by Immrspy »

The Keanu Reeves video by Corridor (https://www.youtube.com/watch?v=3dBiNGufIJw) is one of the more popular examples of deepfakes, and in their behind the scenes video on the other channel (https://www.youtube.com/watch?v=IzEFnbZ0Zd4), they say that they had the model train for 200k iterations on villain.
Even with 200k iterations, they intentionally filmed vertical to limit the size of the face they were replacing, to make the fake look more convincing. In the end, the blurriness can be a result of not enough training, or not large enough of an output size on the model. with only 50k iterations, it most likely just needs to be put back in the oven.

Locked