Google has released the Mendel Linux 4.0 distribution for Coral platforms

Google Inc. introduced distribution update Mendel Linux, designed for use on boards Coral, such as Dev Board and SoM. The Dev Board is a platform for rapid prototyping of hardware systems based on Google Edge TPU (Tensor Processing Unit) to accelerate tasks related to machine learning and neural networks. The SoM (System-on-Module) is one of the ready-to-use solutions for running machine learning-related applications.

The Mendel Linux distribution established is based on Debian and is fully compatible with the repositories of this project (unmodified binary packages and updates from the main Debian repositories are used). The changes are limited to building an image loaded from eMMC cards and including components to support the hardware components of the Coral platform. Coral-specific components is distributed under the Apache 2.0 license.

Mendel Linux 4.0 became is the first release updated to Debian 10 ('buster'). The build is optimized for embedded systems and does not include unnecessary features, including Debian 10 innovations related to SecureBoot and AppArmor support. New features include support for OpenCV and OpenCL, the use of Device Tree overlays, as well as updates to GStreamer, Python 3.7, Linux kernel 4.14, and the U-Boot bootloader 2017.03.3.

Among the specific innovations is the ability to use the GPU (Vivante GC7000) installed on Coral boards for accelerating the conversion of pixel data from the YUV color model to RGB at a performance of up to 130 frames per second for 1080p video, which may be useful when using the boards for processing video from cameras that generate streams in YUV format. For applying machine learning to process streaming video and audio on the fly, it is recommended to use the open framework MediaPipe. For example, based on it, one can implement an object or face recognition and tracking system in video transmitted from a surveillance camera.

Ready-made and already trained standard machine learning models compiled for Edge TPU processors used on Coral boards continue to be delivered at project site, but are gradually being moved to the general catalog of publicly available models TensorFlow Hub. To simplify the development of custom solutions based on Coral boards and Mendel Linux, prepared guide, demonstrating how to assemble a smart sorter from readily available materials that separates colored and white balls into different bins, using Raspberry Pi and Coral USB Accelerator.

Source: opennet.ru