Every day, Adobe researchers push the boundaries of technology—a process that’s second nature to Principal Scientist Brian Price. Price, who joined Adobe Research after completing his PhD 15 years ago, has now contributed to his 100th patent at Adobe. We talked with him about his work, some of his most notable innovations, and what it’s like to bring so many new ideas into the world.
The secret behind 100+ new inventions
Over his years with Adobe Research, Price has fine-tuned a process for developing innovative ideas into new technologies. The first step, he says, is knowing the state of the art. “Most of my older patents were pre-AI, when you needed to understand the algorithms. These days, you need to understand AI and how it works.”
The next step is to look for places where the current methods fail, where users have to do too much work, or where the current technology can’t handle certain cases. Or, sometimes, Price just focuses on ideas that would improve users’ lives and make them happy.
The tricky part, he says, is deciding how to solve the problem. It’s a matter of trial and error, pushing ideas further, and finding colleagues and collaborators who keep you on your toes. It also helps to have pressing research questions and a driving curiosity to keep you inspired.
Price’s big research questions
Price’s early research questions, back when he first began collaborating with Adobe as a master’s student, focused on turning photos into vector graphics. During his PhD research and later when he joined Adobe Research full-time, he expanded into segmentation, the work of identifying an object within an image and then creating a mask so a user can recolor, delete, or move the object into a new image. It’s a difficult process, and designers have spent countless hours painstakingly tracing objects by hand with varying degrees of success.
“To do segmentation right, you need to spend a whole bunch of time fixing little mistakes and touching it up. It’s an intermediate step that no one likes to do. So we try to make that step go away as much as possible, to make it magical so that users need to do very little work—or none at all—and suddenly have a really good mask. And right now is an exciting time because machine learning is revolutionizing what we can do in the space.”
Segmentation is one facet of the big research questions that drive all of Price’s work: “I’m interested in computer vision, which is about using technology to understand images. And at Adobe, it’s not just about understanding images, but building tools that give people more ways control them.”
Highlights from 100 patents—and a few thoughts on what’s next
One of Price’s first projects as a full-time Adobe researcher focused on stereo images—the pairs of images you look at through 3D glasses. “At the time, people were making 3D movies, and some people predicted that everyone would be wearing 3D glasses in front of their TVs soon. Of course, we weren’t sure if that was accurate, but at Adobe Research we need to anticipate and prepare for possible futures.” Price’s first patent was for a technology that could propagate edits from one stereo image to the other image in the pair so that users wouldn’t have to make nearly the exact same set of edits twice.
AI has been a game changer since then. And one of Price’s favorite patents used early AI for matting. “At the time, AI-based methods for segmentation weren’t very accurate. If you made a mask of an object, the edge might be off by two or three pixels. But users want to get that edge exactly right, with extremely detailed edges for things like hair or fur,” says Price.
So Price and his team created an end-to-end machine learning tool that allows users to provide an image, indicate the foreground and background, and automatically receive high-quality matting results.
“When you do research, you’re always trying to beat out the state of the art so you can have the best in the world, and this was one of those moments when we had something much better than anything else out there,” says Price. The technology became Photoshop’s Object Aware Refine Mode in Select and Mask.
Price’s 100th patent grew out of an intern project. Price and his team wanted to address a nagging issue in segmentation: that there are a lot of ways to do it. Some tools ask users to trace around the boundary of an object, others require you to color the inside of the object, and some require a rough circle around the object. Since this variation can be confusing, the team developed technology that lets a user draw on an image any way they want to. The tool interprets their intentions and segments the correct object.
Innovating into the future
Looking ahead, Price is excited about two big trends: generative AI that allows people to create and edit images, and agentic AI that helps with their creative projects. For his own research, he’s experimenting with ways to use these technologies to make things easier for people with huge albums full of images they want to edit and organize.
As he looks back at 100 patents, Price is also thinking about the collaborators who’ve helped make it all possible. “I don’t do this alone. I have co-inventors on all of my patents. On some, I was the main inventor, and on others I was the ‘helper,’ but they’re all really group projects. It’s important to surround yourself with people who are smarter than you are or have different expertise than you do.”
That’s where Adobe Research really shines. “Coming up with new ideas is literally our day job. We have a group of smart people with a lot of different areas of understanding and expertise. And we bring in interns who come in with new ideas. We let them focus on just one problem and really push on it for several months with our feedback. That helps us come up with some really amazing, novel ideas for new inventions,” says Price.
Wondering what else is happening in Adobe Research? Check out our latest news here.
