2025. június 06.
Global Perspectives, Local Innovations
GeoSummit 2025, held on May 12–13 at the Budapest University of Technology and Economics, brought together a diverse international audience to explore how emerging technologies—from AI to advanced remote sensing—are redefining geospatial sciences. The event aimed to foster exchange between research and practice, driving innovation in digital mapping, environmental monitoring, and urban development.
A Diverse and Inspiring Community
Charles Toth, one of the principal organizers of GeoSummit 2025, opened the event with a warm welcome that reflected his connection to the Budapest University of Technology and Economics (BME). An Honorary Doctor of BME and a research professor at The Ohio State University, Professor Toth has spent more than three decades at the forefront of photogrammetry, mobile mapping, and navigation technologies. His academic career has been marked by groundbreaking achievements—such as co-developing the first civilian mobile mapping system—and an exceptional commitment to fostering international collaboration. His long-standing relationship with BME has supported student exchanges, joint research, and leadership in global geospatial science initiatives. GeoSummit 2025 was, in many ways, a reflection of his tireless efforts to bridge continents, disciplines, and generations of researchers. The event was jointly organized by the Departments of Photogrammetry and Geoinformatics and Geodesy and Surveying at BME. Tamás Lovas, head of the Department of Photogrammetry and Geoinformatics, and Szabolcs Rózsa, Dean of the Faculty of Civil Engineering, welcomed participants at the opening session. In the first session, Árpád Barsi, former head of department, presented the current research activities of the Department of Photogrammetry and Geoinformatics, while Szabolcs Rózsa also introduced the research portfolio of the Department of Geodesy and Surveying.
The conference welcomed participants from over a dozen countries, including renowned university professors, researchers from national mapping agencies, innovators from the geospatial industry, and a large number of graduate students.
It was especially rewarding to witness the strong engagement from EELISA students, who joined discussions and gained hands-on exposure to real-world geospatial challenges. Equally important was the presence of our PhD students, who presented their work to a distinguished audience—gaining critical feedback, building international connections, and leaving the event with renewed research momentum.
Keynote Insights: Exploring the Edge of Geospatial Technology
Christian Heipke (Leibniz University Hannover)
From Photogrammetry to Implicit Models: The Deep Learning Turn
Heipke provided a thought-provoking reflection on how deep learning has become a cornerstone in both semantic and geometric image interpretation. Implicit representations like NeRF and Gaussian Splatting now outperform traditional models in complex 3D tasks, but he cautioned that we’re still grappling with challenges such as model trustworthiness, hyperparameter sensitivity, and the generalization gap. One compelling takeaway was the call for better integration between semantics and geometry and a rethinking of data fusion strategies across multiple sensor types—especially in environmentally critical applications like climate monitoring.
Uwe Sörgel (University of Stuttgart)
Crowdsourcing Remote Sensing Ground Truths—At Scale and With Quality
Sörgel’s presentation tackled one of the most pressing bottlenecks in deep learning for remote sensing: labeled training data. He argued convincingly that paid crowdsourcing, if managed properly, can produce vast datasets with comparable reliability to expert labels. The key lies in smart aggregation, visual task guidance, and active learning, which together can cut expert intervention down to under 1%. A memorable example involved annotating complex 3D point clouds—typically a task reserved for specialists—using non-expert workers guided by intuitive tools and interfaces. The talk underlined a crucial shift: volume and structure can sometimes trump precision, provided
Nicolas Paparoditis (IGN France)
From Semantized Meshes to a National Digital Twin
Paparoditis outlined IGN France’s ambitious plan to create a national-scale digital twin by 2028. The future, he argued, lies in semantized 3D meshes—automatically generated, easily updatable, and essential for urban accessibility, infrastructure planning, and ecological policy enforcement. A striking part of his presentation showcased how dual 2D–3D graph models are already being used to map mobility pathways for people with disabilities, bridging the gap between spatial representation and human movement. He emphasized that achieving this vision demands broader access to mobile mapping data, including from autonomous vehicles. As he put it, the conceptual leap from 3D model to national twin is already underway—and it's powered by data, automation, and simulation layers.
Monika Sester (Leibniz University Hannover)
Sustainable Mobility Needs Smart Maps—And Emotionally Intelligent Ones
Rather than focusing purely on technology, Sester explored how spatial visualizations influence behavior, especially in mobility contexts. She introduced the classic triad of sustainable transport strategies—Reduce, Shift, Avoid—and illustrated how geoinformatics can support each with context-aware maps and predictive tools. But more importantly, she urged the community to consider how a visualization feels: does it encourage the user to walk? To share a ride? To avoid unnecessary travel? Her message was clear: the emotional and cognitive impact of spatial data is a technical and societal challenge geographers must embrace.
Csaba Benedek (HUN-REN SZTAKI)
Fusing Sensors, Enhancing Reality: Perceiving the Urban Scene
Benedek’s keynote took us deep into the world of urban scene perception, presenting advanced techniques for synthesizing coherent environmental models from fragmented and heterogeneous data. Drawing on real-time LiDAR, radar, and camera inputs, his team developed methods for automatic calibration, localization, and change detection, even in dense urban environments. What stood out was the innovative use of machine learning to virtually “complete” missing or low-resolution imagery and point clouds, effectively enhancing both the semantic and visual fidelity of urban data. His work opens promising pathways for applications in autonomous navigation, smart city planning, and emergency response scenarios—where every sensor counts, and so does every pixel.
Daniela Poli (AVT Airborne Sensing)
Aerial Mapping in a Changing World: Practical, Precise, and Powerful
Speaking from the perspective of an industry leader, Poli offered a rare behind-the-scenes look at how aerial survey companies combine photogrammetry, LiDAR, and thermal imaging to meet increasingly complex client demands. One of her key messages was that aerial platforms offer accuracy and data richness unmatched by satellites—especially for thermal and oblique imagery—but both are necessary for multi-scale, long-term analysis. She also stressed that mapping companies are more than data suppliers; they’re co-creators in innovation, offering valuable feedback to sensor developers and researchers alike.
Fabio Remondino (FBK Trento)
Giving a Meaning to 3D Point Clouds: Beyond Benchmark Scores
Remondino reminded us that while many recent papers boast small gains in classification accuracy (93.2% vs 93.4%), we should shift our focus to generalization, standardization, and usability. Point cloud semantic segmentation remains one of the toughest problems in 3D geoinformatics, and it's far from solved. His talk called for broader collaboration with neighboring disciplines and increased attention to underrepresented object classes, explainability, and real-world application needs. A standout prediction was that foundation models and multimodal learning (LLMs, VLMs, etc.) will soon enter the 3D space—a game changer in large-scale automated mapping.
Norbert Haala (University of Stuttgart)
SLAM and Gaussian Splatting for Mapping Indoor Worlds
Haala’s keynote bridged robotics and geoinformatics, demonstrating how Gaussian Splatting can enhance visual localization accuracy in complex indoor environments. Using a wheeled robot outfitted with multiple RGB-D sensors and LiDAR, his team created real-time 3D maps enriched with virtual camera views—effectively teaching machines to “see” better by training them on synthetic perspectives. His results showed that these techniques not only improve localization but also strengthen scene understanding for autonomous systems operating in GPS-denied environments like hospitals, tunnels, or historical buildings.
Dorota Iwaszczuk (TU Darmstadt)
Urban Object Extraction in the Age of Foundation Models
Iwaszczuk’s talk focused on the AI-driven extraction of urban features from multi-modal data—particularly 3D point clouds and imagery. She emphasized that while CNNs still dominate, new foundation models like the Segment Anything Model (SAM) are beginning to reshape the field. However, successful application requires adapting these tools to the unique characteristics of geospatial data, such as 3D structure, high dimensionality, and spatial constraints. Her case studies highlighted how semantic segmentation, vectorization, and domain-informed postprocessing can streamline urban planning tasks like roof modeling and vegetation mapping.
Peter J.G. Teunissen (TU Delft)
Pose Estimation at Centimeter Scale—From RTK to Array PPP-RTK
Teunissen, a globally recognized professor in satellite navigation, delivered a mathematically rich but practically grounded lecture on carrier-phase GNSS pose estimation. He introduced participants to the concept of Array PPP-RTK, a technique that fuses precise point positioning with robust attitude estimation using antenna arrays. With applications ranging from autonomous vehicles to maritime navigation and aerospace, his work illustrates how the fusion of geometry and statistics enables high-confidence navigation in even the most challenging environments.
Shared Visions: Panel Discussion on Future Challenges
After the keynote presentations, a thought-provoking panel discussion brought together all keynote speakers for a wide-ranging conversation on the future of geoinformatics and its role in society. The discussion highlighted pressing industrial needs, such as automation-ready data, scalable computing infrastructure, and the integration of sensor networks into real-world decision-making. Speakers also emphasized the growing gap between rapid technological advancement and educational adaptation.
With fewer students entering the geospatial sciences, and a rapidly expanding skill set required—ranging from AI and coding to ethics and domain-specific knowledge—the panelists noted the urgent need to rethink how and what we teach. Particular attention was given to Generation Z's learning preferences, the need for more applied, interdisciplinary, and project-based learning, and the challenge of maintaining engagement in a field that is growing both broader and deeper. The discussion concluded with a shared belief that the future of the discipline depends on a reinvention of its educational models, coupled with stronger ties between academia and industry.
Demos That Made Innovation Tangible
- Robodog Demo (Evosoft Hungary): A nimble robotic platform that showed real-time mobility and obstacle avoidance using onboard sensor fusion—a crowd favorite among students and researchers alike.
- Mixed Reality Demo (BME): Participants explored scanned environments using Meta Quest Pro and Hololens headsets, experiencing firsthand how virtual spaces can enhance design, planning, and teaching.
- Drone Development Demo (VTR Drones): Live demonstrations of LiDAR-equipped UAVs highlighted their versatility in high-resolution terrain modeling and infrastructure inspection.
A Moment of Reflection: Keynote Dinner with the Organizers
After a full day of stimulating sessions, the keynote speakers and the organizing team gathered for a private dinner at a cozy restaurant near the university. The dinner hosted by ISPRS was intended to recognize the local organizers' contributions in securing the venue and coordinating logistical support for the Budapest GeoSummit 2025. This intimate setting allowed for deeper discussions, personal reflections, and the strengthening of professional ties among those who helped shape the event. The relaxed atmosphere provided a welcome break from the intensity of the conference program and served as a fitting close to a successful first day.
Thank You & Access to Keynote Presentations
We sincerely thank all participants, presenters, and partners who contributed to the success of Budapest GeoSummit 2025. Your energy, insight, and curiosity made this conference an outstanding shared experience.
We’re pleased to share the keynote presentation slides below. Feel free to download and revisit the ideas that made this event so impactful:
- Christian Heipke – AI in Photogrammetry and Remote Sensing (PDF link)
- Uwe Sörgel – Paid Crowdsourcing in Remote Sensing: Applications and Quality Control (PDF link)
- Nicolas Paparoditis – National Mapping: Overview (PDF link)
- Daniela Poli – Airborne Mapping Overview (PDF link)
- Csaba Benedek – Urban Scene Perception and Environment Model Synthesis from Multisensorial Spatial Data (PDF link)
- Monika Sester – Sustainable Urban Mobility (PDF link)
- Fabio Remondino – Giving a Meaning to 3D Point Clouds (PDF link)
- Norbert Haala – SLAM-Based Mapping of Complex Indoor Environments (PDF link)
- Dorota Iwaszczuk – AI-based Extraction of Urban Objects from Images and 3D Point Clouds (PDF link)
- Peter J.G. Teunissen – Carrier-Phase GNSS Pose Estimation: From RTK to Array PPP-RTK (PDF link)
- Árpád Barsi – Research Activities of the BME Department of Photogrammetry and Geoinformatics (PDF link)
- Szabolcs Rózsa – Research Activities of the BME Department of Geodesy and Surveying (PDF link)