2/5/2025 | 3:30 PM - 5:30 PM | Regency B

A NeRF for all seasons

Author(s)

Michael Gableman | Air Force Research Lab
Avinash Kak | Purdue University

Abstract

Due to the work accomplished by Shadow NeRF and Sat-NeRF, a framework exists to render a scene at novel viewpoints given a set of satellite images of the scene. Our work extends that framework to capture seasonal features, allowing the scene to reflect specific times of the year. The primary challenge in extending the Neural Radiance Field (NeRF) framework to include seasonal features is to ensure the seasonal features remain independent of other features, such as shadows, within the image. A secondary challenge involves accounting for the small number of satellite images available of the scene. To address these challenges, we provide the network an additional input, the time of the year. Also, we expand the loss function to discourage the network from using solar or seasonal features to explain structural features. We demonstrate our network on eight regions and evaluate the network’s ability to render novel views, seasons, shadows, and create height maps from sets of images captured by the Maxar WorldView-3 satellite.