Reproduce “GaussianPlant: Structure-aligned Gaussian Splatting for 3D Reconstruction of Plants” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yang Yang, Risa Shinoda, Hiroaki Santo, Fumio Okura
GaussianPlant introduces a hierarchical 3D Gaussian Splatting approach to jointly reconstruct the photorealistic appearance and explicit internal structure of plants from multi-view images. This method achieved improved novel-view synthesis quality and quantitatively more accurate branch and leaf instance segmentation compared to baselines.
Key insights
- A hierarchical representation combining coarse structural primitives with fine appearance primitives within 3DGS can effectively disentangle and jointly optimize both aspects of complex objects.
- Semantic guidance from pre-trained 2D foundation models (e.g., DINOv3) can be leveraged to drive consistent 3D structural classification (leaf vs. branch) without requiring explicit 3D semantic ground truth.
- Integrating explicit structural graph optimization, including axis-consistency and void-crossing penalties, significantly improves the recovery of accurate and coherent tree-like branching topologies from noisy 3D primitives.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.