Modeling by Clipped Furniture Parts: Design with Text-Image Model with Stability Understanding
Hironori Yoshida, Seiji Itoh · ACM International Conference on Interactive Media Experiences · 2024
Text input in MR(Mixed Reality) provides options for users to model in details instead of just placing objects, however, 3D modeling with text input costs computation and takes time. To overcome this hurdle, we let the text-image model judge 3D layout of furniture parts. Since vanilla text-image model can not judge furniture stability, we tested two approaches: 1. combine with geometric loss, and 2. fine-tuning the model. We report the comparison of these two approaches and discuss further development for MR integration of our system.