RD8
Data for training language or image models is readily available - the internet provides rich, annotated data in great abundance. In the 3D realm it's a different story. Available datasets, such as grabCAD or Autodesk's Fusion 360 Gallery, are hard-fought and difficult to annotate with spatial and geometric metadata. Despite recent advancements, i.a., Autodesk's B-rep diffusion model, BrepGen, this results in models which continue to lack understanding of their physical and geometric context. Bridging this gap requires not only better data curation but also a fundamental rethinking of how 3D representations are learned, structured, and contextualized in relation to the physical world.
Bio: Rasmus Stavenuiter is senior software developer at RD8, where he is involved in developing computational models for kinematic and statistical analyses of mechanical components and currently is spearheading RD8's AI-research initiative. Rasmus holds a master degree in mathematics from KU.