Reproduce “GeoScaler: Geometry and Rendering-Aware Downsampling of 3D Mesh Textures” 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
Sai Karthikey Pentapati, Anshul Rai, Arkady Ten, Chaitanya Atluru, Alan Bovik
GeoScaler, a deep neural network, downsamples 3D mesh textures by integrating geometry, UV parametrization, and a differentiable rendering loss. This method generates textures that maintain higher visual fidelity in rendered 3D scenes, enabling efficient high-quality graphics on resource-constrained devices with up to 3dB PSNR improvement for 8x downsampling over traditional bicubic methods.
Key insights
- Optimizing texture downsampling directly in the rendered 2D view space, rather than solely on the 2D UV plane, significantly improves perceived visual quality.
- Integrating 3D geometric information and connectivity through graph convolutions is effective at mitigating artifacts arising from UV map discontinuities and warping distortions.
- Dynamically adjusting UV parametrization based on both geometry and texture content allows for more efficient texture space allocation, preserving detail in perceptually important regions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.