Texas A&M professor hosts first End-to-End 3D Learning Workshop

Dr. Zhiwen Fan, assistant professor in Texas A&M University’s electrical and computer engineering department, hosted the first End-to-End 3D Learning Workshop in October at the 2025 International Conference on Computer Vision (ICCV) in Honolulu. Sponsored by the IEEE, ICCV is one of the world’s premier computer vision conferences, featuring cutting-edge research and workshops led by industry and academic experts.

Fan’s workshop attracted roughly 150 researchers eager to explore the rapidly evolving landscape of 3D learning systems. It introduced new AI foundation models for 3D reconstruction, generation and their applications in robotics, extended reality and scientific imaging. Following the workshop, participants submitted 20 research papers, helping to establish best practices for 3D learning.

During the conference, Fan’s team also earned third place in the ICCV 2025 COGS Challenge on compact 3D representation, hosted by META. The team was led by Zihao Zhu, an undergraduate student at Texas A&M, marking his first research project and recognition at an international competition.

The strong response to the workshop reflects an incoming shift in AI system design. Until recently, many AI systems were built as modular pipelines, with different components handling specific tasks rather than relying on a single end-to-end model. A similar evolution is now underway in 3D data analysis. 

“We were motivated to host this workshop by the emergence of the geometric foundation model in 2024,” Fan said. “This topic is timely and relevant to new advancements being made in AI.” 

These new breakthroughs have created quite a buzz, with important application opportunities for anyone involved in developing AI systems. 

While many current approaches rely on end-to-end learning — such as autonomous driving software or ChatGPT — scholars are beginning to explore geometric or 3D foundation models that incorporate a more advanced design to learn spatial structure. These approaches could greatly improve AI capabilities and help achieve a more robust and reliable performance in real-world tasks, such as robotic manipulation. This new technology provides the basis of Fan’s workshop: introducing new methods and tools to engineers who are shaping the future of 3D learning. 

Following the success of the first workshop, Fan and his team plan to continue their efforts at the 2026 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) in Denver this June. They will be hosting a second edition of the End-to-End 3D Learning Workshop, and co-hosting three additional workshops, including Foundation Models Meet Embodied Agents, Multi-Agent Embodied Intelligent Systems in the Agentic AI Era: Opportunities, Challenges, and Future Directions, and 3D Geometry Generation for Scientific Computing. 

Fan looks forward to these next iterations and hopes to see even more researchers apply 3D foundation models to their work. 

“There are many new opportunities to advance both robotics and generative AI with Geometric Foundation Models,” he said. “At CVPR, we will continue our efforts and invite more experts in this domain, including researchers working on advancing generative AI, autonomous driving, and more.”