Advice: considering of 3D relation instead of mixture in network

[YacratesWyh]

The main problem of how to process FaceID now relies merely on the difference of details strength between eyes and others. Yet the purely mixture of texture inversion would mix all the 3d distortion together adding artifacts into your model. Thus I suggest, maybe adding 3d relation to the training would help.

[Paper99]

It's really a nice suggestion! Thank you!