Google has updated its Mendel Linux release for its high-performance Coral developer board and its add-on Coral system-on-module (SoM).
Google's Coral hardware came out of beta last month and at the time it promised to deliver a new version of Mendel based on Debian 10 Buster — the same version of Debian that the Raspberry Pi Foundation built on for the latest version of its standard Raspbian OS. Coral developers can now test version 4.0 of Google's customized Debian OS.
The new Mendel Linux 4.0 Day also comes with upgraded GStreamer pipelines and support for Python 3.7, OpenCV, and OpenCL. And it's updated the Linux kernel to version 4.14 and U-Boot to version 2017.03.3.
The Coral line of products are aimed at getting AI workloads on to edge devices so that pre-trained AI models can process data closer to data collecting sensors, outside of the data center and offline.
SEE: Six in-demand programming languages: Getting started (free PDF)
The $149 Coral Dev Board includes the Coral SoM, which features a GPU, video processing unit (VPU), a Google-developed Edge TPU ASIC with the NXP IMX8M SoC, Wi-Fi and Bluetooth, memory, and storage.
According to Carlos Mendonça, product manager on the Coral Team, it's now possible to use the Dev Board's GPU to convert YUV to RGB pixel data at up to 130 frames per second on 1080p resolution. That's supposed to be "one to two orders of magnitude faster" than devices running Day's predecessor, Chef, could convert YUV to RGB.
"These changes make it possible to run inferences with YUV-producing sources such as cameras and hardware video decoders," he explained.
Google has provided instructions for flashing a new system image from a Linux or Mac system on its updated documents and tools page.
Google today also announced MediaPipe for Coral, its framework for bringing perception capabilities to Coral boards. It's called MediaPipe because it allows developers to build "machine learning perception pipelines" for processing media, like video and audio on the device.
SEE: Best Raspberry Pi alternatives (September 2019 edition)
Machine learning solutions Google has for MediaPipe include hand tracking and gesture recognition, multi-hand tracking, face detection, hair segmentation, and object detection.
"Developers and researchers can prototype their real-time perception use cases starting with the creation of the MediaPipe graph on desktop. Then they can quickly convert and deploy that same graph to the Coral Dev Board, where the quantized TensorFlow Lite model will be accelerated by the Edge TPU," explained Mendonça.
Additionally, Google has released a new tutorial for the Teachable Sorter, a physical sorting machine that uses the Coral USB Accelerator's capabilities to quickly sort different objects as they fall past the camera lens. The Coral Accelerator is a two-finger-sized accessory with an Edge TPU coprocessor that connects to Debian Linux machines via a USB 3.0 Type-C cable.
MORE ON RASPBERRY PI AND SINGLE-BOARD COMPUTERS