Advances in 3D Rendering and Reconstruction

The field of 3D rendering and reconstruction is moving towards more efficient and accurate methods for capturing and representing complex scenes. Recent developments have focused on improving the performance of bidirectional path tracing, stochastic rasterization, and 3D Gaussian splatting. These advancements have enabled faster rendering times, improved visual fidelity, and increased control over render cost and quality. Notably, the integration of physical priors and illumination-agnostic methods has shown promise in addressing challenging lighting conditions and improving novel view synthesis results. Additionally, the use of hierarchical attention networks, mesh compression, and implicit neural representations has demonstrated potential for efficient and high-quality 3D attribute compression and surface reconstruction. Noteworthy papers include: StochasticSplats, which combines 3D Gaussian splatting with stochastic rasterization for efficient and accurate rendering. LITA-GS, which introduces a novel illumination-agnostic novel view synthesis method via reference-free 3DGS and physical priors, achieving state-of-the-art results and faster inference speed.

Sources

Proxy Tracing: Unbiased Reciprocal Estimation for Optimized Sampling in BDPT

StochasticSplats: Stochastic Rasterization for Sorting-Free 3D Gaussian Splatting

LITA-GS: Illumination-Agnostic Novel View Synthesis via Reference-Free 3D Gaussian Splatting and Physical Priors

Hierarchical Attention Networks for Lossless Point Cloud Attribute Compression

Feature-Preserving Mesh Decimation for Normal Integration

Mesh Compression with Quantized Neural Displacement Fields

3D Gaussian Inverse Rendering with Approximated Global Illumination

Luminance-GS: Adapting 3D Gaussian Splatting to Challenging Lighting Conditions with View-Adaptive Curve Adjustment

A Survey on Physics-based Differentiable Rendering

Built with on top of