Since the early days of artificial intelligence, computer scientists have been attempting to create machines that can see and understand the world as we do.
In recent years, these efforts have led to the emergence of computer vision, a vast subfield of AI and computer science that deals with processing the content of visual data. The field has taken great leaps and has even surpassed humans in many tasks related to detecting and labeling objects.
That said, computer vision is still, in many ways, in its infancy and developers using the technology still face many challenges.
A new Des Moines based company —Roboflow—is looking to change that by removing the pain points that developers come across when creating computer vision and AR apps. Founded by Brad Dwyer and Joseph Nelson, Roboflow offers toolsets that make developing computer vision and AR apps easier.
Roboflow grew out of Dwyer’s award-winning Sudoku solver, Magic Sudoku. Following the success of Magic Sudoku, Dwyer decided to step away from his role as CEO at Hatchlings and focus on AR and computer vision technology.
This past summer, Dwyer hired an intern to help him work on enhancing board games using AR and computer vision. That work led to BoardBoss, an app that enhances board games by using AR
“We had a bunch of hypotheses we wanted to test and we were able to pin things down,” said Dwyer. “What we found out was, while we could build those apps, there’s a lot of pain points in the process of building AR apps that are not solved yet. It’s still this really nascent industry.”
Ultimately, Dywer decided he want to build a developer tool to help other people create computer vision and AR apps and that was the genesis of Roboflow.
“It’s one of those things where a lot of applications will feel obvious in hindsight,” said Nelson. “But there’s just these million paper cuts you face when you try to build a computer vision app that we’re focused on making far more seamless.”
Roboflow was recently accepted into Pioneer, an online, remote accelerator that funds projects and startups built around the world. Roboflow has been at the top of Pioneer’s leaderboard for three straight months.
Roboflow’s first tool, Roboflow Organize, allows users to manage image datasets for computer vision modeling. Using Roboflow Organize, developers can keep track of images and annotations as well as save money by doing preprocess and image training ahead of time
“We’re focusing on this tool first because it’s one of the biggest pain points we had and there’s not really any tools on the market to address this,” said Dwyer.
Currently, teams often manage their datasets by using local file systems and tools like Dropbox which are designed for basic file storage and not designed for machine learning and training data.
Long-term, Dwyer and Nelson see multiple potential paths for Roboflow.
“On the one side is the augmented reality side of things where you can make things smart with software instead of hardware. We think that is going to be the way objects interact in the future and want to enable developers to be able to do that,” said Dwyer. “On the other side, the computer vision part is applicable well beyond making objects smart, from things like cancer diagnoses to self-driving cars to even things like insurance and agriculture.”
One of the biggest challenges Roboflow faces is making sure it is tackling the right problems, Dwyer and Nelson told Clay & Milk.
“When you’re really early to a space, people sometimes don’t realize that the tools they’re using aren’t well attuned to the problems and ways in which they’re using those tools,” said Nelson. “So for us, it’s really important we make sure our vision is clear and that we’re focusing on the highest leverage thing.”
“We think computer vision is both going to transform the economy more than the internet did and that its going to move from the realm of research and development to the realm of small teams of developers being able to create really powerful apps that they couldn’t before,” said Dwyer.