Giters

pixpark / gpupixel

Real-time image and video processing library similar to GPUImage, with built-in beauty filters, achieving commercial-grade beauty effects. Written in C++11 and based on OpenGL/ES.

Home Page:https://gpupixel.pixpark.net/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

[BUG] VNN model face landmark index information and coordinate transformation

luckypapa opened this issue · comments

commented

I'm trying to add a new filter that inherits FaceMakeupFilter, but I think I need to add information to faceIndexs. However, I understand that it uses a VNN model(PC model) that provides 278 landmark pointers, but it seems that 105 landmark points are actually provided. (An additional 5 are created by calculating the middle value of the two points)

Why are there 105 landmarks instead of 278?

Is there any data that can confirm the facial location information of the landmark index provided in the model?

If there is no separate data, I would like to convert the x, y coordinate information of the landmark into the x, y coordinates of a 1280x720 2d window and check the location through opencv. Can you tell me how to transform coordinates?