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.
After extraction process, when i try manually tune extraction result i find out that detected face somehow shifted from its correct position (example in attachment fie). Because of this i must fix manually literally every frame, and this is very difficult.
Extraction settings:
Detector: s3fd
Aligner: Fan
Normalization: Clahe
Re Feed: 7
I have AMD RX580 8Gb, Win 10, and old version of faceswap (which only support AMD graphyc)
https://github.com/deepfakes/faceswap/r ... tag/v2.2.0
This, more likely, is an issue with the alignment tracking being out by 1 frame. It happens sometimes (and with some videos).
Sometimes closing and re-opening the manual tool will remedy this.
If the issue persists, try changing your video to an image sequence (ie, use something like ffmpeg or the EFFMPEG tool in faceswap) and then running extract on the image sequence.
My word is final
I've found that certain videos appear to extract properly but I cannot run the alignment manual tool on them (changing an alignment manually on one face applies to the previous frame) and as you mention the actual alignments are out when extracted and checked in some frames.
This seems to correlate with the file itself not having any details in terms of its framerate and compression in the file properties. I have solved this from happening by simply re-encoding the file, I use capcut, which is free and doesn't add a watermark.
