From photo restoration to computer vision

By Celine Gomes

At the recent New England Instrumentation and Technology Symposium (NEITS), Ph.D. student Laura Kaplan from Tufts’ Department of Electrical and Computer Engineering shared her research on improving how artificial intelligence (AI) understands real-world traffic scenes. Working in the Vision and Sensing Systems Lab under the mentorship of Distinguished Professor and Dean of Graduate Education Karen Panetta, Kaplan is helping expand one of the largest datasets of roadway camera footage used to study vehicle collisions. Her work aims to make AI systems better at recognizing and responding to collisions, advancements that could help emergency responders reach crash sites faster and save lives. 

Kaplan’s work in multimodal AI extends beyond traffic safety, reaching into fields like medical training and simulation. Her research is driven by a commitment to make technology that works in the real world — not just in theory.

But she didn’t always plan on becoming an engineer.

Before ever stepping into a computer science classroom, Kaplan worked as a photographer and photo restorer, carefully reconstructing images and paying attention to image data in ways that foreshadowed her later research. Her work behind the camera lens gave her an early appreciation for how people see—and how fragile and complex visual information can be. Yet it wasn’t until later that she realized she wanted to take those instincts in a new direction. 

“After having my own business I realized that I was taking on jobs, not based on payment or artistic fulfillment, but on whether or not they would give me an excuse to use the latest tools from Adobe. I would over-invest every time in reverse engineering them,” Kaplan recalls. Encouraged by friends who recognized her technical aptitude and pushed her to aim higher, Kaplan decided to return to school. She enrolled at the College of Staten Island, where she earned a B.S. in computer science and mathematics, graduating valedictorian of her class.

Kaplan first began working on a dataset of roadway collision camera views during her senior year. She quickly noticed that most datasets were either synthetic, too limited, or unrealistically clean. Researchers were reporting strong model performance based on these idealized conditions, then suggesting real-world implementation. "I felt that was an unfair way to test these models – and then to suggest that you're going to implement that in a real-world scenario? So, I wanted to make a messy data set." That instinct to resist idealized data and instead push toward complexity has stayed with her.

Today, Kaplan develops ways to teach computers how to “see.” By combining deep learning and AI with traditional image processing techniques, she integrates data from image, video, audio, infrared, and hyperspectral sensors. Collaboration is central to her approach: she works with colleagues across Tufts—from the School of Dental Medicine to international relations—and mentors undergraduates in the lab. “The systems we work on are so big, it just wouldn’t make sense for only one person to touch them,” she says.

Looking ahead, Kaplan acknowledges the uncertainty surrounding AI and offers guidance: build strong foundations. “If you want to find the good in AI, you should first fall in love with math, algorithms, and problem solving,” she says. “Start at the start. Invest slow and steady, enjoy all those fundamental pieces, and you’ll have the flexibility to choose exactly where you want to go.”

Kaplan’s journey, from photographer to Ph.D. candidate at Tufts, is a reminder that careers rarely follow a straight path. Sometimes, the choice to reassess your interests and embrace a new direction is what makes the greatest impact.