Developer Guide

• Metadata
• Model Preparation
  • 1. Model files format used by OpenVINO™ Toolkit
  • 2. Model pre- and post-processing specification file
  • 3. Specify model files in GStreamer elements
  • 4. Workflow diagram
• Model-proc File
  • Contents
  • Overview
  • Pre-processing description (input_preproc)
  • Post-processing description (output_postproc)
• How to Create Model-proc File
  • Content
  • Theory
  • How to define pre-processing
  • How to define post-processing
  • Practice
• Python Bindings
  • 1. GStreamer Python bindings
  • 2. Video-analytics specific Python bindings
  • 3. gvapython element
  • 4. Performance considerations
• Custom Processing
  • 1. Consume tensor data and parse/convert it on application side
  • 2. Set C/Python callback in the middle of GStreamer pipeline
  • 3. Insert gvapython element and provide Python callback function
  • 4. Insert new GStreamer element implemented on C/C++
  • 5. Modify source code of post-processors for gvadetect/gvaclassify elements
• Object Tracking
  • Object tracking types
  • Additional configuration
  • Sample
  • How to read object unique id
  • Performance considerations
  • DeepSORT tracking
• GPU device selection
  • 1. Media (VAAPI based) elements
  • 2. Inference (OpenVINO™ based) elements
• Profiling with Intel VTune™
  • 1. Install VTune™
  • 2. Configure VTune™ target platform.
  • 3. Source OpenVINO™ toolkit and Intel® DL Streamer environment variables
  • 4. VTune™ configuration
  • 5. Results Analysis
• Converting NVIDIA DeepStream Pipelines to Intel® Deep Learning Streamer (Intel® DL Streamer) Pipeline Framework
  • Preparing Your Model
  • GStreamer Pipeline Adjustments
  • Multiple Input Streams
• How to Contribute
  • Coding Style
• Latency Tracer
  • Pipeline
  • Sample Output
  • Pipeline latency flag
  • Sample Output
  • Element latency flag
  • Sample Output
  • Setting interval example
  • Sample Output