dabox-project 0.1.0

The DABOX project

_|_|_|      _|_|    _|_|_|      _|_|    _|      _| 
_|    _|  _|    _|  _|    _|  _|    _|    _|  _|   
_|    _|  _|_|_|_|  _|_|_|    _|    _|      _|     
_|    _|  _|    _|  _|    _|  _|    _|    _|  _|   
_|_|_|    _|    _|  _|_|_|      _|_|    _|      _| 

Building robots is hard. If we want to live in a future where there are robots everywhere, robots need to be a lot easier to build.

Getting neural networks to run with low-latency on video streams is notoriously difficult. dabox is designed to be an easy-to-install, ML-friendly Python application with several features that every robot needs.

Features available out of "dabox"

• Low-latency inference with FFmpeg, ZMQ, and ONNX Runtime
• Web-based 3D visualization by viser
• Real-time RTSP, LL-HLS, WebRTC streams by MediaMTX
• Automatic camera discovery and multi-camera support
• Supported on Mac, Linux, and x86+dGPU systems

Installation

Create environment

dabox requires python >= 3.10. We recommend using conda to manage dependencies. Make sure to install Miniconda before proceeding.

conda create --name dabox -y python=3.10 && conda activate dabox

Install from source

The current recommended way to install dabox is from source.

git clone https://github.com/jefequien/dabox.git && cd dabox
pip install -e .'[dev]'

Usage

Start DABOX!

dabox-up

# Visit http://localhost:8080