The function I want to implement:
Use your mobile phone to take a frontal photo of yourself on the web mobile terminal, and then replace the face of the person in the template photo with your own face, and the size and tone of the face must be consistent with the template Faces blend seamlessly into photos.
Problems I encountered:
I use Tencent Youtu’s facial features positioning interface to extract faces, but how can I seamlessly integrate the face size and tone with the template.
The easiest way is to make it so white that you can ignore the shooting angle and ambient light. In this way, the portrait recognition is OK~hmm~"similar". If you want to be realistic, prepare a few more sets of templates.