NVIDIA Jetson Orin - GstInference Integrating deep learning inference into GStreamer

Overview

GstInference is an open-source project from RidgeRun Engineering that provides a framework for integrating deep learning inference into GStreamer. Either use one of the included elements to do out-of-the box inference using the most popular deep learning architectures or leverage the base classes and utilities to support your own custom architecture.

GstInference objective is to put inference in the hands of the developer without having to worry about frameworks, platforms, or media handling. It also has a strong focus on performance, which most Python frameworks don't really consider.

Supported frameworks:

TensorFlow
NCSDK

Provided plugins:

TinyYOLO version 2 detection
Inception version 2 classification
Inception version 4 classification
FaceNet face recognition
Detection overlay for detection metadata visualization
Classification overlay for classification metadata visualization
Embedding overlay for embedding metadata visualization

Promo/Demo Video

Running inference on GStreamer pipelines

It is easy to add inference elements to an existing GStreamer pipeline. By combining inference and overlay elements you can perform deep learning tasks and then easily visualize the results.

Image classification