Introduction

This article provides an overview of the Maivin dataset workflow and links to detailed articles for each step in the workflow.

Overview

The Maivin Recorder application allows the user to record the various topics available on Maivin into an MCAP format.  This format allows multiple sensors to be streamed into a portable recording with synchronized timestamps.  Supported sensors include the camera, gps, imu, and any other sensors which publish to a Maivin Topic on the system.

The MCAP format is commonly used for automotive and robotics applications through ROS (Robot Operating System) to capture what are known as "rosbags".  These can be played back using various tools such as Foxglove Studio and RViz.

Maivin Recorder can include annotations generated by a model running on the device.  These annotations can be used to verify and audit the performance of the model on the device and to improve the dataset and further refine the model being deployed back to the device.

The Maivin Publisher application consumes the MCAP files generated by Maivin Recorder and publishes them into a format suitable for model training.  Maivin Publisher supports publishing directly into Deep View Enterprise or exporting to JPEG images with EXIF annotations with optional support for Darknet-style datasets.

Getting Started

To get started capturing datasets with Maivin the first step is to refer to the Maivin Dataset Recording article.