1st Workshop on Computer Vision for Earth Observation (CV4EO) Applications – WACV2024
Hybrid event: all presenters in person, with live streaming of the event for registered participants
Geospatial Artificial Intelligence (GeoAI) integrates methods from spatial sciences (e.g., geographic information systems – GIS) and AI to enable knowledge extraction from big geospatial data. GeoAI is extensively applied in conjunction with Earth Observation (EO) data, which entails capturing information about the Earth’s surface using sensors mounted on e.g. satellites and in-situ instruments. By incorporating advancements from computer vision (CV) on EO data, GeoAI finds extensive applications in human dynamics, precision agriculture, disaster management, humanitarian assistance, and national security. Unlike traditional natural images used in CV benchmarks, EO data presents unique characteristics and challenges that include spatial & temporal awareness, data volumes & diversity, and multimodal reasoning. Importantly, appropriately addressing these challenges hold the potential for groundbreaking applications benefiting human and environmental well-being.
Objectives
Outcomes targeted by this 1st Workshop on Computer Vision for Earth Observation (CV4EO) Applications at WACV’24 include promoting the exchange between computer vision researchers with experts from geoscience domains, as well as bridging the gap between computer vision base research with government agencies (problem owners), national laboratories (applied science) and industry (data providers and solution deployment) in the context of challenges and opportunities related to image understanding methods for EO applications. Since applications include humanitarian assistance, disaster response, precision agriculture, national security missions, environment monitoring, promoting the collaboration across all involved parties can greatly benefit the development of CV-enabled tools that can effectively inform decision making for such cases potentially having direct impact on lives and the environment.
The workshop aims to achieve the following goals:
- Promote Multidisciplinary Interaction: CV for EO is a multidisciplinary space with diverse data sources, application domains, and involved disciplines. This workshop will encourage cross-disciplinary interactions, fostering knowledge exchange and collaborative efforts.
- Address Challenges in Multimodal Data: Data from existing remote sensing modalities are heterogeneous and complementary in many aspects. Passive imagery sources vary significantly in number of channels (e.g., multispectral vs. hyperspectral data), while active imagery such as Synthetic Aperture Radar (SAR) include both amplitude and phase components. Variations in data representation formats are another challenge, as geospatial data are often represented in vector format and their rasterization for consumption by co-opted CV models lead to information loss. Given this variety of remote sensing modalities, the workshop will focus on strategies for learning while leveraging multiple EO data sources effectively. We invite discussions that address data challenges related to heterogeneity, spatial and temporal resolution, satellite view angle, data fusion and representation formats.
- Enable scaling of CV4EO applications: Current EO satellite constellations collect 100+TBs of data a day, and images can be billions of pixels large. While these volumes impose challenges for data management and require customization of model training pipelines, the volume and diversity of remote sensing archival data represent a great potential for a wide variety of applications and the development of large-scale models.
- Benefit the Computer Vision Community: Tackling the data volumes and multimodal reasoning challenges in CV for EO also has broader implications for the CV community. Similar challenges are faced in domains such as biomedical image analysis. The workshop will explore how advancements in CV for EO can contribute to addressing these shared challenges.
- Explore Impactful Applications: The workshop will expose members of the CV community to the exciting and impactful applications within the EO domain. Decision-making processes in disaster response, national security, and environmental protection will be discussed, showcasing the real-world applications of CV4EO.
- Foster Talent Formation and Recruitment: The workshop aims to attract and engage talents in CV for EO, as institutions conducting geospatial data analysis face fierce competition for CV and AI expertise. By introducing attendees to applied research opportunities and practical applications, the workshop will contribute to talent formation and recruitment efforts.
Program
Hawaii Standard Time (UTC/GMT -10 hours)
January 7th (Sunday), 2024 – Afternoon (PM)
1:00-1:10 Opening remarks
1:10-1:55 Keynote
Transforming Earth Observation Analytics: Advancements and Applications of Foundation Models in Remote Sensing – by Dr. Hamed Alemohammad, Director of the Center for Geospatial Analytics, Clark University
There is a rapid growth in developing foundation models for Earth observations (EO) including multi-modal models that combine inputs from various sensors in Earth orbit. These developments build on the success of deploying various supervised and semi-supervised techniques for EO in the last several years, and publication of a large number of benchmark datasets. In this presentation, I will provide an overview of these benchmarks and their successes. Next, I will review recent advancements in using transformer-based architectures and developing foundation models for EO. In conclusion, I will present the results from the Prithvi foundation model for various downstream tasks including cloud gap filling, flood mapping, land cover/crop type mapping, and burn scar detection.
2:00 – 3:00 – Oral Presentations
Full papers – 12 min with Q&A
- 2:00-2:12 – CNet: A Novel Seabed Coral Reef Image Segmentation Approach Based on Deep Learning
Hanqi Zhang (Wuhan University)*; Ming Li (ETH Zurich,Wuhan University); Jiageng Zhong (Wuhan University); Jiangying Qin (Wuhan University)
- 2:12-2:25 – Modernized Training of U-Net for Aerial Semantic Segmentation
Jakub Straka (University of West Bohemia)*; Ivan Gruber (University of West Bohemia)
- 2:25-2:37 – GAST: Geometry-Aware Structure Transformer
Maxim Khomiakov (Technical University of Denmark)*; Michael Andersen (Technical University of Denmark); Jes Frellsen (Technical University of Denmark)
- 2:40-2:52 – TinyWT: A Large-Scale Wind Turbine Dataset of Satellite Images for Tiny Object Detection
Mingye Zhu (University of Science and Technology of China); Zhicheng Yang (PAII Inc.)*; Hang Zhou (Alchemy Insight); Chen Du (PAII Inc.); Andy Wong (PAII Inc.); Yibing Wei (University of Wisconsin – Madison); Zhuo Deng (Tsinghua University Shenzhen International Graduate School); Mei Han (PAII Inc.); Jui-Hsin Lai (PAII Inc.)
Short papers – 8 min with Q&A
- 2:52-3:00 – Conditional diffusion models for land-use and imperviousness forecasting
Philipe Ambrozio Dias (Oak Ridge National Laboratory )*; Christa Breslford (Oak Ridge National Laboratory (ORNL))
3:00-3:30 Coffee Break
3:30-4:10 Keynote
The Role of Earth Observation Image Analysis in Humanitarian Response – by Dr. Kasie Richards, American Red Cross Situational Awareness Lead
A discussion covering the growing need for EO analysis as the frequency of disasters and emergency response increases. With dialogue on the opportunities of EO image analysis in addressing decision support and situation awareness in humanitarian response, including present and future applications. Touching on the challenges in field adoption of EO, CV and other forward leaning research and technology.
4:10-4:55 Roundtable discussion (45min)
Keynote speakers, authors of accepted papers, and members of the audience will have the opportunity to engage in discussions from a predefined set of topics, as well as questions posed by the audience during the event.
4:55-5:00 Closing remarks
Organizers
- Philipe Dias, Oak Ridge National Laboratory (ORNL)
- Dalton Lunga, Oak Ridge National Laboratory (ORNL)
- Katherine R. Picchione, MIT Lincoln Laboratory
- Manil Maskey, National Aeronautics and Space Administration (NASA)
- Ronny Hänsch, German Aerospace Center (DLR)
Advisory Committee
- Nathan Jacobs, Washington University in St. Louis
- Amy Tabb, United States Department of Agriculture (USDA)
- Henry Medeiros, University of Florida
- Fabio Pacifici, Maxar Technologies
- Shawn Newsam, University of California Merced
- Stefano Ermon, Stanford University
Program Committee
- Caleb Robinson, Microsoft AI for Good Research Lab
- Claudio Persello, University of Twente
- Abhishek Potnis, Oak Ridge National Laboratory (ORNL)
- Hunsoo Song, Purdue University
- Jeffrey Liu, MIT Lincoln Laboratory
- Michael Schmitt, University of the Bundeswehr Munich
- Orhun Aydin, Saint Louis University
- Jacob Ardnt, Oak Ridge National Laboratory (ORNL)
- Ribana Roscher, University of Bonn
- Srikumar Sastry, Washington University in St Louis
- Ujjwal Verma, Manipal Institute of Technology
- Lexie Yang, Oak Ridge National Laboratory (ORNL)