DXNN SDK

DXNN SDK is the core software stack of DX-AllSuite, providing everything you need to compile, deploy, and run AI models on DEEPX NPUs — from model optimization to real-time application integration.

Modules

Compiler

DX-COM

Convert ONNX models into DEEPX-optimized .dxnn binaries using the DX-COM compilation pipeline with INT8 quantization.

Runtime

DX-RT

Deploy and run .dxnn models on DEEPX NPU with the C++ and Python runtime APIs.

Streaming

DX-STREAM

Build multi-stream AI video pipelines with DEEPX NPU inference using GStreamer-based custom plugins.

Application

DX-APP

Create C++ and Python applications with ready-to-use examples for detection, classification, segmentation, and more.

Module Overview

Module	Role	Environment
DX-COM	Compiles .onnx → .dxnn with quantization	Host PC (x86_64)
DX-RT	Runs .dxnn on DEEPX NPU via C++/Python API	Host / Target Board
DX-STREAM	GStreamer pipeline for real-time video AI	Target Board
DX-APP	Ready-to-use inference application examples	Host / Target Board

Development Workflow

1. Compile — Use DX-COM to convert your ONNX model into a .dxnn binary.
2. Deploy — Transfer the .dxnn file to your target device.
3. Run — Execute inference via DX-RT APIs or build a full pipeline with DX-STREAM.
4. Integrate — Use DX-APP examples as a starting point for your custom application.