Scalable GenXR: Generative AI for extended reality in retail, FHNW School of Computer Science

The rapid growth in immersive technologies coupled with increasingly popular digital commerce (i.e., online shopping) has created innovation opportunities for streamlining the creation of high‑quality 3D models of the retail items, ready to deploy on extended (virtual, augmented, mixed) reality, specifically augmented reality (AR) on off-the-shelf smartphones. Yet, despite the recent developments, producing high-quality 3D models remains costly, time‑consuming, and heavily dependent on manual labor. These constraints limit scalability and accessibility of realistic “try before you buy” experiences. GenXR advances generative AI‑driven 3D content creation through latest computer vision and deep learning frameworks for automated, scalable, and sustainable generation of 3D and AR assets from one (or a small set of) 2D product images, ideally without human intervention.

Building on recent breakthroughs in generative AI, neural rendering, computer vision, deep learning and natural language processing, the project tackles image-to-3D and text-to-3D approaches and addresses one of the central bottlenecks in current 3D model creation workflows: the need for extensive manual processing to achieve highly realistic visualizations.

To achieve the project goals, the GenXR team systematically evaluates state‑of‑the‑art approaches through benchmarking, assesses perceptual quality in terms of realism and level of detail, and advances post‑processing techniques toward full automation. In addition to scalability, GenXR tackles sustainability: As demand for AI-generated 3D content grows, so does the environmental impact of computation. Thus, GenXR examines methods for energy-efficient 3D model generation, optimizing both training and inference to reduce the carbon footprint of AI-based content pipelines. GenXR builds on three previously funded projects: an Innosuisse Innovation Cheque feasibility study, an HTZ-funded feasibility study and a project funded by the Forschungsfonds Aargau.