Virtual 360-degree tours have become the norm in real estate, more so since the pandemic made it impossible for buyers to visit homes in person.
Manual recording
Today, virtual tours not only distinguish a real-estate agent from competitors, but they also help weed out window shoppers from serious buyers. In fact, research shows that virtual tours can help lower-priced homes sell for 3% more and higher-priced homes sell for 1.6% more.
But shooting a video is no easy feat. Placing, orienting, and moving a camera to capture the essence of a space plays an integral role in photography and videography.
A carefully composed frame provides the viewer with information about a scene and conveys the story the narrator wants to tell, instilling a desired emotion in the viewer.
Preserving a home's aesthetics
Professional photographers and videographers tasked with creating real estate virtual tours spend a lot of time perfecting the camera framing, which often leads to extensive cost footprints, and while this provides a higher degree of control, it’s desirable to frame a scene automatically and not by hand.
Of course, real estate agents can also take the DIY route. But that would involve the increased strain of selecting and buying equipment, planning how to capture the tour, cleaning the house, testing equipment and lighting, and learning to use virtual tour software, which, more than anything else, demands additional time and energy.
To help resolve these issues, Stony Brook University’s Ph.D. students Desai Xie and Ping Hu, along with Arie E. Kaufman, Distinguished Professor and Chair of Computer Science at the university, developed GAIT. Their framework can teach itself to move the camera and generate trajectories that show the most aesthetic views of a given indoor space, while also ensuring that the video isn't shaky.
An AI-generated trajectory for recording an aesthetic video
The team’s work was recently presented at ICCV (International Conference on Computer Vision), one of the top conferences in computer vision.
Ph.D. student Desai Xie says, “Manually defining trajectories of camera poses in a 3D space is a challenging task that even takes professional users a couple of hours. But once trained, GAIT can generate multiple tours for the same scene, unlike an artist, who has to manually define every single trajectory.”
“This technology can tremendously reduce manual workload,” says Dr. Kaufman. “It allows the artist or the DIY-inclined real estate agent to better focus on capturing the essence of the space.”
The project, which was supported in part by NSF grants, and by IBM, SUNY and Adobe, holds much promise for the future of capturing indoor and outdoor spaces in an aesthetically pleasing way, not only for real estate virtual tours, but also for several other industries, including photography, videography, and cinematography.
Communications Assistant
Ankita Nagpal