Mohit Gupta brings holistic approach to computer vision

Mohit Gupta

Photography is both old and new, in a sense.  Though it’s been around since the 19th century, it keeps being reinvented through technological advances—and that perpetual transformation shows no sign of slowing down.

Mohit Gupta, an expert in computer vision, says it’s time to rethink some of our basic assumptions about cameras. Gupta is a new assistant professor in the University of Wisconsin-Madison Department of Computer Sciences.

“Cameras were designed to mimic the human eye, and images have been designed for human consumption,” he says.  “But we don’t have to stick to this paradigm now.  We can now think of photography not as capturing images, but as acquiring visual data that may not make sense to a human observer at first.  But through a combination of unconventional optics and decoding algorithms, you can create images that have much more information than what a traditional camera captures.”

While conventional cameras capture only 2D image information, these next-generation computational cameras, as they are called, can capture 3D shape information, a full 360-degree view, and even things invisible to humans, such as ultraviolet, X-rays or polarized light.

What’s more, these cameras and the data-rich images they capture have an almost limitless range of potential applications.

Computer vision and robotics systems enabled by these cutting-edge cameras have the potential to revolutionize our lives through a wide range of technologies:  autonomous cars, augmented reality, hands-free user interfaces, surgical robots, space exploration, and robotics in both industry and the home.

While some researchers in computer vision focus on one side of the equation or the other—either the physical systems to capture images, or algorithms to interpret those images—Gupta believes it’s important to work on those two sides in tandem.

“For a vision system to be successful within challenging, real-world settings, these two components should be designed synergistically,” he says.

Gupta is currently designing computer vision systems, especially 3D cameras, that will perform reliably outside the predictable confines of a lab.

He wants these systems to work “in the wild,” as he puts it, under non-ideal conditions (picture an autonomous car driving on a foggy day, or a drone flying through a dust storm) or involve materials that are tricky to capture in images (imagine a robot performing surgery inside a human body).

Gupta also seeks to build vision systems with capabilities that, until now, were impossible—like systems that can measure not just the appearance or 3D shape of objects, but sense underlying material properties, like what an object is made of.  One way this technology could be used is the non-invasive detection of diseases like skin cancer.  Using low-cost imaging devices, doctors will be able to rapidly detect features deep below the skin that are suggestive of diseases like melanoma.

Gupta, who earned his Ph.D. in robotics from Carnegie Mellon University in 2011, spent five years as a research scientist at Columbia University in New York before joining the UW-Madison faculty.

His wife, Tripti Singh, is a faculty physician with a joint appointment between UW-Madison and the UW Hospital and Clinics, where she specializes in nephrology. The couple have a toddler son, so Gupta is enthusiastic about Madison’s family-friendly side.  “It seems tailor-made for raising a young family,” he says.

Gupta was also drawn to Madison by the caliber of fellow researchers he’ll be able to collaborate with, both in the CS department and across campus.

Working across disciplinary boundaries is important to Gupta, and part of what led him to specialize in computer vision. “The two things I’m most fascinated with are light and the human brain.  I’m not a physicist or neuroscientist, but there is one field in computer science where these intersect, and that’s computer vision.”

The field also excites Gupta since it allows him to work on fundamental scientific problems while also seeing his ideas come to life real-world settings.

“I try to make my research as practically feasible as possible,” he says, noting that some of his work was recently licensed for robotic assembly of machine parts.  “As an academic, it’s satisfying to see your research make a practical impact on a multi-billion-dollar industry like industrial automation.”

Gupta is ready to share his passion for discovery and holistic approach to computer vision with UW students.  “Our primary job is to inspire and make them curious,” he says.

[Photo credit: Sarah Morton, UW-Madison College of Letters & Science]