Jetstream2 Fuels Spatial AI Challenge Winning Team
The Spatial AI Challenge brought together researchers, data scientists, and geospatial experts to push the boundaries of spatial data and artificial intelligence. Participants leveraged Jetstream2's computing resources to tackle real-world problems using spatial AI—an advanced form of AI that interprets 3D data including depth and object relationships, enhancing machines' ability to analyze physical environments.
Spatial AI: The Next Frontier
Unlike traditional AI focused on 2D data, spatial AI processes three-dimensional information, enabling new research and industry applications. This technology expands how machines understand geography, distance, and spatial relationships, opening doors to innovations in fields like agriculture, hydrology, and historical mapping.
The NSF-funded Institute for Geospatial Understanding through an Integrative Discovery Environment (I-GUIDE) champions this cause. I-GUIDE supports transformative scientific and societal advances by harnessing vast geospatial datasets, and recently hosted the first Spatial AI Challenge to accelerate progress in this area.
Winning with GeoMapCLIP
The 2024 Spatial AI Challenge was won by a Purdue University team from the Rosen Center for Advanced Computing. Their project, GeoMapCLIP, builds on GeoCLIP to automatically determine the location of unknown geospatial images by analyzing visual cues. Judges highlighted the project’s reproducible methodology, clear presentation, and applicability across diverse domains.
Training GeoMapCLIP required processing millions of satellite images, demanding significant GPU memory and storage. Jetstream2 provided the necessary infrastructure, including a dedicated virtual machine with an A100 GPU offering 20GB of GPU RAM, plus ample disk space for large datasets.
Jetstream2’s Role in the Challenge
- Supported 6 out of 9 teams with GPU resources
- Enabled near-continuous uptime and stability for uninterrupted model training
- Provided root access for streamlined software installation and environment management
- Supplied sufficient storage for hosting extensive satellite image datasets
Jungha Woo, a team member, emphasized that consistent, high-performance resources removed bottlenecks and allowed the team to focus on refining their model. This stability accelerated their progress throughout the competition.
Looking Ahead
GeoMapCLIP aims to improve geographic interpretation by bridging historical and modern map analysis through AI. The team credited collaboration, mentorship, and Jetstream2’s computing resources for their success. They hope their project will contribute to making geospatial data more accessible and usable via intelligent models.
Those interested can sign up for an upcoming online session with the winning team to learn more about GeoMapCLIP.
Jetstream2 continues to provide computing resources to researchers and educators across the US via the NSF-funded ACCESS project and the NAIRR Pilot, supporting advancements in spatial AI and beyond.
Your membership also unlocks:
