Thursday, 3 September 2026 - 14:15 – 15:15 

The fields of photogrammetry, 3D computer vision and computer graphics are undergoing a rapid paradigm shift and a clear convergence. Historically constrained by offline sensors, discrete geometric structures and isolated perception pipelines, modern 3D frameworks are evolving into unified ecosystems capable of generating, representing and understanding complex real-world environments, even in real-time. 

This talk provides a holistic overview of these transformations, exploring how recent advancements are redefining the boundaries of 3D data acquisition, processing, representation and interpretation.

We will first examine innovative aspects of 3D data acquisition and processing, highlighting the transition from traditional acquisition and pure geometric processing to multi-modal sensing and learning-based 3D pipelines. 

Next, we delve into the evolution of 3D scene representations, tracing the journey from classical point clouds and meshes to continuous implicit Neural Representations (NeRFs) and explicit representations like Gaussian Splatting. We will discuss how these representations bridge geometric fidelity with photorealistic rendering while optimizing runtime performance. 

Finally, we address 3D scene understanding and semantic segmentation, showcasing how traditional supervised deep learning methods are now paired to semantic feature distillation (e.g. DINO, CLIP) enabling machines to perceive, segment and reason about spatial concepts without manual supervision.