Special Session on Realistic 3D Deep Learning at ICME 2025

From June 30 to July 4, 2025, the IEEE International Conference on Multimedia & Expo (ICME) 2025 will take place in Nantes, France, bringing together leading researchers and industry experts to discuss the latest advancements in multimedia technologies.

A highlight of the conference is the Special Session SS07: “Towards Realistic 3D Deep Learning with Limited Supervision,” co-organized by Jürgen Gall, Principal Investigator for Embodied AI at the Lamarr Institute and professor and head of the Computer Vision Group at the University of Bonn. This session will explore cutting-edge methods for 3D deep learning, focusing on overcoming data limitations while ensuring model robustness and scalability.

Topics will include:

• 3D reconstruction from sparse observations
• Generative model-assisted 3D reconstruction and novel view synthesis
• Semi-supervised and weakly-supervised learning for 3D data
• Self-supervised learning for 3D representation learning
• 2D model-assisted 3D deep learning
• Domain adaptation and generalization for 3D models in real-world environments
• Leveraging unlabeled or sparse data for 3D scene understanding
• Applications of limited supervision in real-world 3D tasks (e.g., AR/VR, robotics, medical imaging, multimedia)
• Challenges in evaluating and benchmarking 3D deep learning models under limited supervision

By bringing together researchers and practitioners, this session aims to advance the field of 3D deep learning and tackle the challenges of working with limited supervision.

You can still register to attend the conference here: ICME 2025 Registration