Need autonomous driving training data? ›

How Does Working With Mighty AI…Work?

How Does Working With Mighty AI…Work?

It’s a fair question, and we get it a lot. At the highest level, our platform enables you to manage your autonomous vehicle perception data—while freeing you from creating the software tools necessary to generate ground truth datasets—so you can focus on getting to production faster. But you’re probably wondering: what does that look like when it comes to life? As a customer, what are you responsible for, and what does Mighty AI take care of?

To illustrate, we made an illustration. (heh)

As a Mighty AI customer, you’re on deck only for the stuff in blue. Mighty AI does all the parts in black, and our annotator community takes care of the green items.

To elaborate: our platform handles all aspects of the data lifecycle, including ingestion, classification, annotation, quality assessment, and dataset export. We begin by collaborating with you to define project requirements and specs, then we take care of designing the annotating/labeling workflows. This includes choosing the appropriate tools and calibrating quality settings. You can upload data to our platform through Mighty Studio and its built-in media manager or programmatically using our API. Our annotators, who we recruit and train, then tackle the tasks. Mighty Quality—our proprietary quality system—ensures datasets meet the quality standards you expect, and you review using Mighty Studio to easily sort, view, and export annotations.

We often encourage customers to run a small batch of images through our platform, which we review together to ensure annotations align with your needs for developing your models. Once you’ve given the thumbs up, we launch the task at scale and notify all qualified annotators that it’s available. Tons of annotators get going on the task, while Mighty Quality continuously runs several layers of monitoring and QA (you might say that’s our secret sauce). After more accuracy checks, we export the data to you. From there, you train your model, maybe tweak your training data requirements, and the cycle starts over. In addition to generating ground truth datasets, we can validate annotations your machine learning systems generate. To get you started quickly, Mighty AI has a number of industry-tested workflows to validate if models are performing as desired and make any necessary annotation corrections to improve their performance.

Does that help? Need anything clarified further? Let us know.

Editor’s Note: This post was originally published in October 2016 and has been revamped and updated for accuracy and comprehensiveness.