Part-level 3D shape generation driven by user intention inference with preferential Bayesian optimization

Abstract

Advancements in generative artificial intelligence have introduced state-of-the-art models capable of producing impressive visual shape outputs. However, when it comes to supporting decisions during the three-dimensional shape creation process, prioritizing outputs that align with designers' needs over mere visual craftsmanship becomes crucial. Furthermore, designers often intricately combine three-dimensional parts of various shapes to create novel designs. The ability to generate designs that align with the designers' intentions at the part-level is pivotal for assisting designers. Hence, we introduced BOgen, a novel system that empowers designers to proactively generate and synthesize part-level three-dimensional shapes and enhances their overall user experience by reflecting designer intentions through Bayesian optimization. We assessed BOgen's performance using a study involving 30 designers. The results revealed that, compared to the baseline, BOgen fulfilled the designer requirements for three-dimensional shape part recommendations and shape exploration space guidance. BOgen assists designers in navigation and development, offering design suggestions and fostering proactive design exploration and creation during early-stage design ideation.