3D reconstruction is a major challenge in computer vision (CV) research. Especially in recent years, optimization-based methods, such as differentiable rendering, NeuS, and 3D Gaussian splatting, are commonly used to reconstruct the shape and appearance of the estimated scene to match the observation. On the other hand, in many real-world objects, there are very often “constrained” behind the observed shape, such as “plant branches have a tree structure.” However, it is extremely difficult, even with the latest CV methods, to perform 3D reconstruction while guaranteeing these constraints on the object structure and generation rules. This research aims to realize and systematize “constrained” 3D reconstruction, especially focusing on the 3D reconstruction of plants with global and complex constraints.

Grants

• 2025/4～2029/3 JSPS KAKENHI (B)
  PI: Fumio Okura, Grant: 14,500K JPY

Publications

•