Cornell University researchers have developed a robotic photography system that can automatically rove an indoor space and shoot aesthetically appealing photos.
The researchers said AutoPhoto employs a learned aesthetic machine learning (ML) model, and trailblazes new computer vision and autonomous photography applications by integrating current ML models with customized deep learning models.
The system is based on a learned aesthetic estimation model trained on over 1 million human-ranked photos.
Cornell's Hubert Lin said, "To guide the robot, we trained a separate model to move around in an environment and find a place that looks good."
After scanning dozens of three-dimensional photos of indoor scenes and correctly selecting the best compositional angles, AutoPhoto explored a common space on a Clearpath Jackal robot; the team said the system roamed the space and shot three quality photos in a few minutes.
From Cornell Chronicle
View Full Article
Abstracts Copyright © 2022 SmithBucklin, Washington, DC, USA
No entries found