TreeSRNF: Square-Root Normal Fields for Generative Modelling of the Geometric and Structural Variability in Tree-like 3D Objects

TreeSRNF introduces a new way to represent and compare tree-shaped 3D objects by treating them as points in a Riemannian tree-shape space. That matters because vegetation is awkward for existing shape pipelines: trees deform not just in silhouette, but in branching topology, branch thickness, and surface geometry all at once. The framework extends Square Root Normal Fields beyond genus-0 surfaces so it can model both geometry and structure in a single system.

The method computes point-wise and branch-wise correspondences, then uses geodesic paths to describe deformations between tree shapes. In practical terms, that gives developers a more principled way to measure how one plant differs from another, rather than relying on skeletal approximations or thickness-only proxies. The authors also use the space to compute statistical summaries such as means and modes of variation, which is useful for building datasets, analyzing asset families, or driving procedural generation.

The generative side is the part game teams will care about most. By fitting probability distributions to a population of tree-shaped objects, the system can synthesize new variants that preserve both structure and surface detail. The work was tested on real and synthetic plants and botanical trees, and it claims better performance than prior methods. For studios building forests, biomes, or simulation-heavy worlds, that points toward more controllable variation with less manual cleanup.