Greg Eisenhauer, Celine Lin, and Ling Liu

Researchers Awarded Nearly $1.2 Million to Revolutionize 3D Reconstruction Technology

Generating 3D reconstructions of 2D images promises to become the next technology to revolutionize human life, with possibilities of enhanced automated cars and more accurate medical diagnoses. The field is expected to be worth $1.5 billion by 2028.

Researchers from Georgia Tech’s School of Computer Science hope to make the technology more efficient and widely accessible. The group, consisting of Associate Professor Yingyan (Celine) Lin, Professor Ling Liu, and Senior Research Scientist Greg Eisenhauer, was recently awarded a $1.198 million grant from the National Science Foundation to accomplish this goal.

3D reconstruction technology takes 2D images of a real-life scenario. It then generates images to create an accurate 3D reconstruction of the scene that can be viewed digitally.

While there have been significant breakthroughs in 3D reconstruction quality in recent years, further improvements are necessary to make it practical for real-world applications.

“Distributed learning of 3D reconstruction from limited 2D images needs to address two open challenges,” said Eisenhauer. “First, efficient on-device 2D synthesis of imagery of multiple scenes, and second, the technical challenge of distributed learning on a large population of edge clients with private imagery data and heterogeneous, possibly limited, computing resources.”

Lin hopes this project can help enhance the technology's efficiency and processing speed while preserving image quality, ultimately enabling real-time 3D reconstruction on various everyday devices.

“We cannot currently run 3D reconstruction technology on devices with limited resources, such as mobile devices. Our project aims to overcome this and other efficiency bottlenecks,” Lin said.

While 3D reconstruction is often associated with autonomous vehicles and virtual reality, Lin emphasizes its broader implications for everyday life, including in the medical field.

"For instance, imagine a physician in a remote area collaborating with a specialist in Atlanta during a surgical procedure," Lin said. "With more efficient 3D reconstruction technology, the specialist could view and interact with a three-dimensional representation of the patient, providing real-time guidance. Such detailed 3D imagery is vital for more precise diagnoses and better outcomes."

The joint research project will also bring research innovations to Georgia Tech educational programs to train the next generation of computer science and engineering students at the graduate and undergraduate levels. This will include hands-on experience in optimization methods and techniques, including hardware-algorithm co-design and system-algorithm co-design for complex learning problems.