[A-10] Mathematical Systems Science and its Applications

1:45 PM - 2:00 PM JST(4:45 AM - 5:00 AM UTC)Candidate for Academic Incentive Award

[A-10-25] Local Deformation of Rigid Structures Using a Flow Matching Model

◎Tomoki Oya¹, Hitoshi Nishimura¹, Mori Kurokawa¹ (1. KDDI Research, Inc.)

Keywords:

Flow Matching,Image Editing,Rigid Structures

This study proposes a method for achieving local deformation of rigid structures based on a Flow Matching framework for image editing. While existing approaches mainly target non-rigid deformations such as facial expressions and pose changes, local deformation of rigid structures (e.g., steel towers or truss frameworks) remains insufficiently explored. To address this, we introduce spatially selective control using a mask that applies editing guidance only within the target region, while preserving structural consistency outside the mask. This leads to a redefinition of the velocity field in Flow Matching, enabling controlled local deformation. Experimental results on synthetic images of steel structures demonstrate that the proposed method successfully induces localized structural changes. We further quantitatively evaluate the degree of deformation using the ratio of changes inside and outside the mask.