Hello team,
I begin using deepfacelab and after some tryings, I have a question: is it better to have thousands of pictures with very good and medium quality, or less pictures with only very good quality?
And after extract pictures from a video, I have a lot of similar pictures. Is is better to get all of them, or delete all double pictures?
Thanks a lot for your answer.
face samples: quality vs number
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 for Extracting and preparing face sets for training a model in Faceswap.
If you have found a bug are having issues with the Extraction process not working, then you should post in the Extract Support forum.
Please mark any answers that fixed your problems so others can find the solutions.
face samples: quality vs number
Re: face samples: quality vs number
We do not directly support DFL, however the core principles remain the same and are covered in our Guides:
My word is final
Re: face samples: quality vs number
Oops ! I thought DFL was the new version of FaceSwap. I use it because FaceSwap crashes for me at the extracting.
But you're right. It's similar.
I looked at the guide. I assumed "multiple faces" means identical faces? So if I delete all identical faces to keep only one, will the training faster?
And for my other question, should I use a lot of pictures with little poor quality added to less pictures with good quality, or should I keep only good quality pictures? I actually have 1000 aligned pictures.
Thanks for your answer, and congratulations for all your guides.