Reproduce “GS-2M: Material-aware Gaussian Splatting for High-fidelity Mesh Reconstruction” 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
Dinh Minh Nguyen, Malte Avenhaus, Thomas Lindemeier
A material-aware Gaussian Splatting framework, GS-2M, jointly optimizes geometry and intrinsic material properties to reconstruct high-fidelity meshes from multi-view images, specifically addressing challenging reflective surfaces without relying on complex neural components. This framework attained a mean Chamfer Distance of 0.53 on the DTU benchmark and achieved a PSNR of 34.22 for novel view synthesis, outperforming several state-of-the-art explicit methods.
Key insights
- Joint optimization of geometry and intrinsic material properties within a 3D Gaussian Splatting framework enables robust mesh reconstruction of highly reflective surfaces.
- A self-supervised roughness strategy, driven by multi-view photometric variation, effectively guides material decomposition without requiring complex neural architectures or external priors.
- Enhancements to multi-view constraints, including multi-view normal consistency and occlusion-aware filtering, contribute significantly to improved geometric accuracy and novel view synthesis quality.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.