Coral from Google - GstInference - Introduction

Coral from Google
Error creating thumbnail: Unable to save thumbnail to destination
Introduction
Getting Started SoM Overview
GStreamer
Example Pipelines
GstInference
Introduction Why use GstInference? Example Pipelines Demos
Camera Drivers
Introduction MIPI Camera connection How to flash Google Coral How to recompile kernel Adding a new camera driver Troubleshooting
Reference Documentation
Contact Us

Deep Learning (DL) has revolutionized classic computer vision techniques to enable even more intelligent and autonomous systems. To ease the software development burden for complex embedded visual Deep Learning applications, a multimedia framework, such as GStreamer, is utilized to simplify the task. The Open Source GStreamer audio video streaming framework is a good choice as it separates the complexities of handling streaming video from the inference models processing the individual frames. GstInference is an open-source GStreamer project sponsored by RidgeRun that allows easy integration of deep learning networks into your video streaming application.

Example of use of GstInference in the Coral. The model used was the TinyYolo V2. Video took from https://pixabay.com/videos/road-autobahn-motorway-highway-11018/

GstInference and R2Inference are supported by Coral. To install it, follow these guides:

• Getting R2Inference code
• Building R2Inference: Coral is optimized for Tensorflow Lite. For installation, please follow the Guide to install TFLite for ARM64.
• Getting GstInference code
• Building GstInference

After the installation is completed, you can generate GStreamer pipelines for Coral using different GstInference elements in this Pipeline Generator Tool.