#belarus #lukashenko #deepfake

Instagram: @tech_solo_official
Website: https://techsolo.cc/
- - -
Жыве́ Белару́сь!

With XSeg you only need to mask a few but various faces from the faceset, 30-50 for regular deepfake. It will take about 1-2 hour.

Already segmented faces can be reused from your previous deepfakes, they will help for training XSeg. Use "fetch" script to gather segmented faces into separated folder.

You can mask 4000 various faces (from dataset which is used for pretraining) and train XSeg, thus you will get an automatic masker for any face.
The problem of such approach is that XSeg will learn your own rules of masking, which are not suitable for other deepfakers, for example jaw line, hair line, beard line, etc.

If 4000 faces will be masked by community, result will be awful, because every man has his own masking rules, thus XSeg neural network will be embarrassed by the lack of segmentation logic in samples.


DeepFaceLab project page : https://github.com/iperov/DeepFaceLab

В Ростове-на-Дону я успел: провалиться тёмным вечером в обрыв у родника Гремучий, уклонившись от гонщика нелегального, побывать на руинах старой цивилизации в Большой Мартыновке, вкусить смак чамбо, укусить пирожок с лебедой и увидеть у моря таганрогскую Японию (Богудонию).