Idea to creat more source Images for Training
Hi,
i use a Idea to create missing angles for the face training.
The 3D Face Reconstruction tool know everyone i think. It works fine but for a faceswap with few images from the person, it's difficult to create a good swap.
My idea is (never se this befor), take some fotos from other persons for the traning.
Yes, this is not new but usually, the other person looks not the same 
.
I search for "twin" person but find only identical mouth or Eyes.
Now the "new" part for my traning (hope this is new for someone) :
I extract the faces with Unet-Dfl Mask, "redact" the eye Area and train the model with the Unet-Dfl Mask. With the Unet-Mask, it's only train the same mouth.
To save time for redact the photos i use a simple trick... Excel....
How to:
Import the extract Images in Excel (a screenshot form the multiple faces save a lot of time, look for 256x256!/Head size), copy boxes over the eyes and then take a Screenshot with the boxes (watch out 256x256! /Head). Extract this image and it's finish.
In my "eyes" it works very good and the training / converts looks incredible.
Here some Pics for "how to" unterstanding. (sorry, bad english skills)
Hope someone like this / can use it.
