Google has created a browser-based version of MediaPipe framework that allows you to create and implement a modular machine learning algorithms, reported in the blog Google Developers. The developers have created a websiteon which you can try out a few basic computer vision algorithms.
Over the last decade, researchers in the field of machine learning has made huge progress in this area and their algorithms are increasingly used in practice. However, until now the spread of machine learning algorithms is largely constrained by technical problems associated with their use. One of the key problems is that many algorithms, including computer vision algorithms require too many resources to operate in real-time on smartphones and desktop computers. In addition, as in other areas of programming, developers raises the question of portability of their programs, reuse of open source and rapid deployment.
In 2019, the developers from Google created the framework MediaPipe, which largely solves these problems. In it, a developer can take the already written by other programmers basic computer vision algorithms and a simple way to combine them in the desired pipeline. In addition, the framework was originally designed with the expectation that the program should turn on all major platforms, without the need for adaptation to each of them separately.
The authors tried to optimize the code to work on common user devices. As an example, they previously showed a demo version of the app for smartphones that in real time builds a model in hand that can be useful for the qualitative recognition of finger gestures using the camera.