0.4

• Switched from old mediapipe 0.8 FaceMesh solution to new 0.10 FaceLandmarker.
  • Number of landmasks increased from 468 to 478!Thus, the output shape of the landmark infomation numpy file is changed to [N][478][3].
• Removed cv2 dependency from face extraction.
  • Now OpenCV is needed only for the computation of the motion energy.
• Added code to extract blendshape data from faces. 52 blendshapes are optionally saved into a numyp array of shape [N][52]. See option --outblendshapes in extract_face_data module.

0.3 - Python 3.11, mediapipe 0.10.x, mtcnn 1.0.0

Updated dependencies to support Python 3.11, mediapipe 0.10.x, mtcnn 1.0.0.

0.2

Freezing code before switching to modern versions of MediaPipe and MTCNN (thus supporting Python 3.11+)

SignLanguage-VideoProcessingTools

First release with tools for trimming, identifying faces, cropping, extracting face mesh data.