Face landmark recognition is the task of detecting and identifying specific points on a human face, such as the corners of the eyes, the tip of the nose, and the corners of the mouth. MediaPipe is a framework for building multimodal machine learning pipelines, which includes a neural network for face landmark recognition.
The MediaPipe face landmark recognition neural network uses a combination of convolutional neural network (CNN) and regression layers to predict the coordinates of facial landmarks. The network takes as input an image or video frame of a human face and outputs a list of coordinates for each facial landmark.
The ReInHerit toolkit Face-fit app uses the MediaPipe network to understand the pose and expression of users and evaluate how similar they are w.r.t. those depicted in an artwork.