User:Stephenli2000/sandbox

What is OpenVisionCapsules?
OpenVisionCapsules is a format for encapsulating Computer Vision or Deep Neural Network algorithms, initially proposed & implemented by Aotu.ai. It was released through OpenCV.org in early January 2020 as a BSD licensed Open Source Project. Source code is available here.

The typical input of a VisionCapsule is a video frame, the output is semantic information or structure data of the frame, e.g. car, motorcycle, and truck detection.

The goal of the project
OpenVisionCapsules was created to address the issue of the fragmented smart vision device market. OpenVisionCapsules specifies the input, the output, how to package, and the entry points of the program. It does not replace the format of any Neural Network or algorithms or a runtime like OpenVINO. It is simply a packaging format, including all necessary code, files, and metadata required to describe and implement an algorithm. By establishing a standardized communication, the output of an algorithm, or a so-called VisionCapsule, can be consumed by another VisionCapsule.

Because the OpenVisionCapsules is a self-contained, portable format, it allows compatible devices to distribute algorithms cross devices, and communicate with consistent semantic information. By combining many simple VisionCapsules, on either one device or across multiple devices, sophisticated data can be extracted from video streams to achieve complicated artificial intelligent tasks.

How does it work
Introducing OpenVisionCapsules

How to build an OpenVisionCapsules compatible Hardware

Download and test some prebuilt VisionCapsules