Integrating MediaPipe into an Android video conferencing application

The process of migrating to the MediaPipe ML framework from higher-level tools (ML Kit and OpenCV) that were used to process the video stream between the device camera and WebRTC-based video conferencing. Moving away from the existing solution will solve the high load problem, reduce the size of the application and allow for more flexibility in frame processing. The main task in the transition to MediaPipe is to develop an architecture of interaction with the framework that allows switching modes of operation in real time, as well as to give a parallel video stream to several consumers. During the transition process, MediaPipe is additionally adapted to the specifics of the platform and application: refinement of the JNI layer, creation of calculators, and implementation of custom graphs.