Closed xuguozhi closed 2 years ago
Hi, I think this article explains it best: https://google.github.io/mediapipe/solutions/face_mesh.html#key-concepts
The default keypoints you get from mediapipe are in 2d normalized camera space and that function (which uses the mediapipe canoncial face model and the OpenCV SolvePnP function), the get_metric_landmarks()
function will give you normalized 3d coordinates. You can visualize those coordinates with the --show_3d
parameter (which shows those 3d coordinates in a 2d window, but you'll how stable they are compared to the default 2d ones).
yes , much stable than 2D direct points
Hi I am reading your code. I don't know what is the purpose for get_metric_landmarks https://github.com/JimWest/MeFaMo/blob/341813f8b33c57ef9eabdcb7c20c02c70963675b/mefamo/custom/face_geometry.py#L2482