Component Selection Guide#

This page provides clear, task-oriented guidance on what components to choose for your edge AI solution. It focuses on the following eight core technologies used in Open Edge Platform:

Model Development

Training or fine-tuning a model?

Train models with:

PyTorch

Model development and research workflow.

Choose it when: You need to design, train, fine-tune, validate, or experiment with neural networks and export the resulting model for deployment.

Do not choose it as the main tool when: You already have a prepared model and you want to optimize it and deploy on Intel hardware.

Model Deployment

You already have a model?

Run optimized inference inside your application with:

OpenVINO

Model conversion, optimization, compilation, and runtime inference.

Choose it when: You need optimized, in-process model inference on Intel hardware from C++, Python, or other supported runtime integrations.

Run inference over REST or gRPC with:

OpenVINO Model Server (OVMS)

Production model serving powered by OpenVINO.

Choose it when: Multiple applications need remote inference through REST or gRPC, model lifecycle management, scalable deployment, or separation of clients from model execution.

It uses OpenVINO runtime, offerting its inference capabilities.

Multimedia Pipelines

Need image, video, or audio?

Use custom live multimedia graph without built-in AI analytics with:

GStreamer

General multimedia pipeline foundation.

Choose it when: You need a modular, plugin-based framework for custom live or file-based audio/video capture, transformation, transport, playback, or streaming.

It does not offer AI-based analysis.

Use AI-enabled analytics with:

DL Streamer

AI-enabled media analytics pipelines.

Choose it when: You are building real-time or batch audio/video analytics and want ready-to-use GStreamer elements for inference, tracking, classification, metadata, and hardware-aware media flow.

It is based on GStreamer and offers its media processing features.

Use REST-managed analytics pipelines with:

DL Streamer Pipeline Server

Deployment and lifecycle management of media analytics pipelines.

Choose it when: You want DL Streamer pipelines exposed as a configurable service, managed through REST APIs, and deployable as a reusable microservice.

It is based on DL Streamer, extending its capabilities with pipeline serving.

Image and Media Processing

Need to process image or video data?

Apply frame-level image processing or custom computer vision with:

OpenCV

Computer vision algorithms and image manipulation.

Choose it when: You need frame-level image processing, calibration, geometry, feature extraction, classical computer vision, visualization, or custom pre/post-processing.

Implement codec, container, probing, recording, or transcoding with:

FFmpeg

Media ingest, conversion, packaging, and codec operations.

Choose it when: You need broad codec and container support, probing, muxing/demuxing, recording, transcoding, filtering, or reliable media conversion through command-line tools or libraries.

When selecting the software stack appropriate for your solution, consider what features you need and pick the building blocks accordingly. The components described here do not exclude each other and may be complementary.