Save the face mesh points animation video file.

save_mesh_animation.py

import tkinter

import cv2
import matplotlib
from matplotlib import animation, pyplot
from mpl_toolkits.mplot3d import Axes3D

# The module mesh_detector could be found here:
# https://github.com/yinguobing/head-pose-estimation/tree/google-face-mesh
from mesh_detector import MeshDetector

matplotlib.use("TkAgg")


if __name__ == "__main__":
    # Construct a face mesh detector.
    md = MeshDetector(
        "/home/robin/Desktop/head-pose-estimation/assets/face_landmark.tflite")

    # Read in the image file.
    image_file = '/home/robin/Desktop/lfpw-trainset-image_0871.jpg'
    image = cv2.imread(image_file)

    # The mesh detector is sensitive to the face box size. The sample image I
    # used is smaller than expected.
    epn_width = 20
    image = cv2.copyMakeBorder(image,
                               epn_width, epn_width, epn_width, epn_width,
                               cv2.BORDER_CONSTANT, value=[0, 0, 0])

    # Then, get the mesh.
    mesh, _ = md.get_mesh(image)

    # Draw the face mesh points in 3D, and save the animation.
    fig = pyplot.figure(1, figsize=[12.8, 9.6], dpi=80)
    ax = Axes3D(fig)
    x, y, z = [], [], []
    face = ax.scatter3D([], [], [], c='#ae7181')

    def init():
        ax.view_init(260, 270)
        ax.set_xlim3d(0, 400)
        ax.set_ylim3d(0, 400)
        ax.set_zlim3d(-200, 200)
        ax.set_xlabel('x')
        ax.set_ylabel('y')
        return face

    def update(point):
        print(point)
        x.append(point[0])
        y.append(point[1])
        z.append(point[2])
        face._offsets3d = (x, y, z)
        return face

    ani = animation.FuncAnimation(
        fig, update, frames=mesh, init_func=init)

    # Save the animation to video file.
    # ani.save('animation.mp4')

    pyplot.show()