Enabling the Computer Vision Revolution

Computer vision — the ability for computers to “understand” visual content like pictures and videos — is revolutionizing various industries, from healthcare to retail and insurance. And part of the reason it’s now being widely adopted is that startups are popping up to help innovators apply computer vision simply and without in-depth machine learning or deep learning expertise.

Ordinarily, one of the most painful things when working on computer vision is the annotation, organizing, pre-processing and augmentation of images. You can go the route of writing your own scripts to accomplish a lot of this, but that is a lot of heavy lifting and very time-consuming.

Roboflow is a startup that aims to simplify the process of building computer vision models so you can ditch those scripts and accomplish what you need to do in a fraction of the time. Roboflow allows you to tackle every step of the computer vision lifecycle, all the way from uploading images to annotation, organizing those images, training and even deploying your computer vision models.

On the latest episode of AI & U, Mark McQuade sits down with Joseph Nelson, Co-founder and CEO of Roboflow, to discuss computer vision and its real-life use cases.

Tune in to hear about:

• The evolution of Roboflow, from its humble beginnings as Magic Sudoku
• Making machine learning more accessible
• Roboflow and computer vision use cases
• Computer vision predictions for 2021

Joseph discusses the advanced work being done by developers as a hobby. “You know that famous quote, right? ‘What the nerds are doing in the nights and weekends is what the Fortune 500 will be doing in five years.’ And we see that being true across the platform, so we welcome all sorts of hobbyist students to try and build stuff with computer vision just as much as large enterprises that are dipping their toes into things.”

Joseph also discusses the limitations of computer vision. “Computer vision is pretty powerful, but also pretty brittle. In other words, we can do a pretty good job of training the model to recognize people that are entering a store on cameras or just count how many customers you have enter your store on a given day. But if you took that same camera perspective and then you try to use it at a restaurant at a slightly different angle, or maybe you used it in the winter and people have thicker coats on, that model would start to degrade. And so what's interesting is that computer vision today does a really good job of learning one scenario, but has a lot to learn in terms of generalizability as well as adapting to a domain problem.”

Joseph gives a number of use cases for computer vision, including one on the golf course. “We see a lot of interesting work when pairing vision with drones. I was talking to a company yesterday that is using a drone to fly across golf courses to see what the presence of clover grass, dallisgrass and types of weeds is like. In a golf course, it's a pretty nasty problem. And so this business flies drones over the top of golf courses and helps better maintain and get control of weed outbreaks before they become too serious. The intersection of aerial inspection work and computer vision yields all kinds of impressive opportunities.”

Join the Conversation: Find Solve on Twitter and LinkedIn, or follow along via RSS.

Stay on top of what's next in technology

Learn about tech trends, innovations and how technologists are working today.