Reproduce “Single-view 3D reconstruction via inverse procedural modeling” 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
Albert Garifullin, Nikolay Maiorov, Vladimir Frolov
An inverse procedural modeling approach reconstructs structured, editable 3D models from minimal input images, leveraging optimization strategies tailored for both differentiable and non-differentiable procedural generators. The method generates artifact-free results for complex objects like trees and buildings, offering practical utility superior to general-purpose deep learning methods by providing models with inherent structural information.
Key insights
- Directly reconstructing procedural parameters from images inherently yields structured and editable 3D models, solving the lack of utility found in monolithic meshes from other methods.
- A hybrid optimization framework, combining gradient-based methods with evolutionary algorithms, effectively handles both differentiable and non-differentiable procedural generators.
- Even from a single viewpoint, high-quality mesh reconstruction is achievable, with additional views primarily enhancing texture accuracy rather than geometric fidelity.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.