I've been trying faceswap for a while, however I haven't got an satisfying swap yet. I was wondering if the problem is on my training sources. I know that different angles and light conditions can improve the result, however I only have limited sources.
My data is like extracting 30sec videos from about 4 or 5 different sources to get nearly 3000 faces each; however there are still hundreds of similar faces in each 30sec videos, I was wondering if this is the problem that too many similar faces causes the low efficiency of training.
If I can get the same angles in 5sec as in 30sec videos, do they get similar results?(Although if choosing 5 sec, I will only have under 1000 faces training) Or which may be better?
Thank you!
training sources question
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.
- ericpan0513
- Posts: 23
- Joined: Wed Jul 22, 2020 3:34 am
- Has thanked: 6 times
training sources question
- bryanlyon
- Site Admin
- Posts: 793
- Joined: Fri Jul 12, 2019 12:49 am
- Location: San Francisco
- Has thanked: 4 times
- Been thanked: 218 times
- Contact:
Re: training sources question
I don't recommend extracting from extremely short videos. I usually prefer setting a high een (extract every number) and feeding a full movie or other long content. That gives me large variety of data with minimal overlap.
Re: training sources question
Look at this. should shed some light on your question.
I dunno what I'm doing
2X RTX 3090 : RTX 3080 : RTX: 2060 : 2x RTX 2080 Super : Ghetto 1060