Want to know about the Faceswap's Face Extraction process? Got tips, ideas or just want to learn about how it all works? Then this is the place for you
I have been working on a project, using The Three Stooges. The videos that I find are all low quality, even the ones that claim to be remastered. That adversely affects the extraction process. If I understand correctly, there is no reason that I need to extract faces and create an alignments file from the video that I intend to use, in order to create a model in the training phase. Can I just collect photographs from the Internet to use for training and use that training with the video?
Using photos as part of your dataset is fine. You can extract a folder full of images just like a video. I recommend you still use the videos as much as possible. In general on the videos, avoid "remastered" unless it comes from original film stock and try to find the best data you can. If you come by the Discord and show examples, we might be able to help with advice. But unfortunately poor quality sources are hard to overcome.