Improving accuracy of computer vision models, Voxel51 raises $12.5M
Were you unable to attend Transform 2022? Check out all of the summit sessions in our on-demand library now! Watch here.
Computer vision AI models rely on having properly labeled data in order to infer the correct object. The challenge of helping to verify that data used for a model is accurate is one that Ann Arbor, Michigan-based startup Voxel51 is aiming to solve with open-source tools and a commercial service called FiftyOne Teams.
Ann Arbor is home to the University of Michigan, which is where Voxel51 cofounder and CEO Jason Corso works as a professor, and where he got the idea to build the new company. Corso’s research focuses on computer vision applications like the relationship of video to natural language. In recent years, as computer vision adoption has grown so, too, has the size of the datasets.
“When I was a grad student, I had datasets that numbered in the dozens and I could look at every sample,” Corso told VentureBeat. “Now my students came along and they can’t look at a million samples; it’s just not possible, so the need for Voxel51 was born out of that.”
It’s a need that has found a reception in the marketplace and with investors. Today, the company announced that it has raised $12.5 million in series A funding from Drive Capital, Top Harvest and Shasta Ventures, as well as from existing investors eLab Ventures and ID Ventures, and the University of Michigan.
MetaBeat will bring together thought leaders to give guidance on how metaverse technology will transform the way all industries communicate and do business on October 4 in San Francisco, CA.
The challenge and opportunity of unstructured data for computer vision
Unstructured data takes many forms and includes any type of data that doesn’t fit into a specific data structure format (e.g., columns and rows).
Among the most common forms of unstructured data is video content, which is growing exponentially as the number of cameras continues to grow globally. Getting value out of unstructured video data can happen in a number of different ways. Corso noted that there are technologies that help users to extract semantically meaningful information from images, such as simple tools that allow users to look for images taken in a certain location.
While there is no shortage of unstructured image data and large datasets used to help train computer vision models, ensuring accuracy is a challenge.
“Our whole shtick is that when datasets grew to be over 10 million samples, no one bothered to look at the images anymore,” Corso said.
What Voxel51 is doing is acting as a bridge between what a data engineer does when creating datasets, and what either that same engineer or their partner does when they’re training models. The Voxel51 technology supports visualizing annotations on image data and can be used to identify potential errors as well enabling users to compare the performance of different models.
Corso explained that Voxel51 enables users to semantically slice data to understand the correctness of a model. For example, via a Python API, a user can execute a query on a computer vision dataset to find all the images in which one model outperforms another, for images where there is a child running into the street.
Open source and the enterprise
Voxel51 started as an open-source product, but alongside the funding announcement, the company is officially launching its FiftyOne Teams enterprise offering, which provides commercial support and additional capabilities.
The Voxel51 open-source project was first launched in August of 2020 and has grown over the past two years, with up to 150,000 monthly users. “The open-source project is built for a user with local data, where all the data is on a single system,” Corso said.
In contrast, the commercially supported FiftyOne Teams offering provides support for cloud data, as well as role-based access control (RBAC) to enable multiple users to use the same platform securely. Currently the commercial service is not offered as a fully managed cloud service, instead organizations will still need to run the technology on-premises or in their own cloud instances.
“We are envisioning a future in which, at least for certain types of customers, maybe startups who don’t want to go and deploy locally into their ecosystem, a managed service, but that will not be coming out for some time,” Corso said.
VentureBeat’s mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.