deepx

DEEPX NPU SDK (DXNN) for edge AI: compile, quantize and deploy ONNX models on DEEPX NPUs

DEEPX NPU tutorials in Jupyter: install the SDK, compile ONNX to .dxnn, run inference and build Python/C++ edge AI apps. Demos: YOLO26, PP-OCR, CLIP, hand landmarks, 36-channel video.

DEEPX NPU runtime and Linux driver for edge AI inference: DX-RT, firmware, DX-APP C++/Python examples and DX-Stream GStreamer plugins. Runs .dxnn models on DX-M1 M.2 accelerators.

DEEPX Model Zoo: 490+ pretrained computer vision models for DEEPX NPUs.

DEEPX NPU compiler (DX-COM): convert ONNX models into optimized .dxnn binaries with INT8 quantization for DEEPX NPUs such as DX-M1. Part of the DXNN SDK

List of NPUs

A PyTorch-based Convolutional Neural Network (CNN) project for MNIST digit classification

Two-node bare-metal K3s cluster on LattePanda 3 Delta with DeepX DX-M1 NPU inference, local registry, CI/CD and Prometheus/Grafana

Qualcomm IQ-9075 EVK 한 대로 하는 엣지 AI 예제 42개: 내장 Hexagon NPU와 DEEPX DX-M1 NPU (설치부터 실습·응용까지)

같은 ONNX 모델을 DEEPX DX-M1(.dxnn)과 Qualcomm Hexagon NPU(.tflite)용으로 바꾸고 변환 전후 정확도·속도를 비교하는 스크립트