FakeApp 1.1 Tutorial

(Reference)

https://www.youtube.com/watch?v=ghTb2kZSpZE&ab_channel=IrrelevantVoice

 

1. Download any video1 and video2, ffmepg.exe(https://www.ffmpeg.org/)

2. Create folder c:\fakes\data_A

3. Create folder c:\fakes\data_B

4. Create folder c:\fakes\model

5. Copy ffmpeg.exe and video2 file to c:\fakes\data_A

6. ffmpeg -i video2.mp4 -vf fps=1 "a%04d.png"

7. Copy ffmpeg.exe and video1 file to c:\fakes\data_B

8. ffmpeg -i video1.mp4 -vf fps=0.5 "b%04d.png"

9. Extract

* Copy Path Data: and paste to same textbox because of bugs before start

9.1. Path Data: change to 'C:/fakes/data_A'

9.2. Start

9.3. Path Data: change to 'C:/fakes/data_B'

9.4. Start

9.5. After aligned faces extracted, you should delete some non-face images

      because it makes training speed slower and quality lower

10.Train

* Copy Path Model:, Data A:, Data B: and paste to same textbox because of bugs before start

10.1. Start

10.2. When loss under 0.01, press 'q' After 3~5~10 hours later, you will get good model
(it's up to you)

10.3. After Trained, Delete C:\Users\[User]\AppData\Local\Temp\t.exe *** file,

    _MEI88442 Folder (Name is up to system?)

11. Conv...

* Copy Path Model:, Data: and paste to same textbox because of bugs before start

11.1. Start

11.2. After Trained, Delete C:\Users\[User]\AppData\Local\Temp\m.exe *** files

* Tips

- fps: ... 0.5 1 ... 5 (up to you)

- More datasets make good quality