VISIMO has won an Air Force Phase I Small Business Technology Transfer (STTR) award to conduct research on the synthetic generation of landscape training data. Work will be conducted in partnership with Colorado State University (CSU), and will play a vital role in advancing technologies reliant on vast quantities of landscape imagery such as flight simulators, unmanned aerial vehicles (UAVs), and even video games.
The goal of the project, “Generative Adversarial Network (GAN) for Synthetic Data Generation,” is to use artificial intelligence (AI) and machine learning (ML) to create synthetic landscape images and 3D environments. The technology will be trained on real-world images, and then produce original images that look real. This unlimited supply of landscape data can be used to train ML models, and significantly reduces the cost of data.
Training on the synthetic data will give technologies like autonomous vehicles the ability to “see” in real world environments where visibility is limited.
“Our research has the potential to advance a number of human-centered technologies,” said James Julius, President and CEO of VISIMO.
“Our solution can help firefighting UAVs see through smoke, improve targeting for humanitarian aid drops in fog, and create realistic flight simulations for pilots, as just a few examples.”
GANs are most commonly used to generate realistic images of human faces. Today, many advertisements use GAN-generated faces rather than images of real human models.
“Human faces are incredibly complex,” said Dino Mintas, VP of Data Science and Software Development at VISIMO. “If GANs can create realistic human faces, they certainly have the capacity to create realistic landscapes. We saw the connection and are happy the Air Force sees the same potential.”
VISIMO’s work is supported by Steve Simske, professor of Systems Engineering at CSU. Simske holds advanced degrees in biomedical and electrical engineering and is a former director at Hewlett-Packard Labs.