Google has introduced the magritte library, which is designed to automatically hide faces in photos and videos, for example, to ensure the privacy of people accidentally caught in the frame. Hiding faces makes sense when creating collections of images and videos that are submitted for analysis to third-party researchers or posted publicly (for example, when publishing panoramas and photos on Google Maps or when exchanging data to train machine learning systems). The library uses machine learning methods to detect objects in a frame and is designed as an add-on to the MediaPipe framework, which uses TensorFlow. The code is written in C++ and distributed under the Apache 2.0 license.
The library is characterized by low consumption of processor resources and can be adapted to hide not only faces, but also arbitrary objects, such as license plates of cars. Among other things, magritte provides handlers for reliably detecting objects, tracking their movement in the video, determining the area to change and applying an effect that makes the object unrecognizable (for example, pixelization, blurring, and sticker attachment are supported).
Source: opennet.ru