Advances in 3D Reconstruction and Rendering

The field of 3D reconstruction and rendering is rapidly advancing, with a focus on improving the accuracy and efficiency of surface characterization and geometry learning. Recent developments have introduced novel frameworks and methods for incorporating photoclinometry techniques, surface normal and albedo estimation, and specular-suppressed inputs to accurately represent complex surfaces. The use of Gaussian splats and neural networks has also become increasingly popular, enabling the reconstruction of high-frequency surface textures and the generation of zoomable maps with rivers and fjords. Furthermore, researchers have made significant progress in addressing the transparency-depth dilemma in 3D reconstruction, enabling the accurate reconstruction of transparent surfaces. Noteworthy papers include PMNI, which achieves state-of-the-art performance in the reconstruction of reflective surfaces, and TSGS, which significantly outperforms current leading methods in transparent surface reconstruction. Additionally, the introduction of methods such as ARAP-GS and CAGE-GS has enabled flexible and efficient deformation of 3D Gaussian Splatting scenes, while CompGS++ has achieved substantial compression of Gaussian splatting data for static and dynamic scenes.

Sources

Stereophotoclinometry Revisited

Generation of Zoomable maps with Rivers and Fjords

PMNI: Pose-free Multi-view Normal Integration for Reflective and Textureless Surface Reconstruction

GaSLight: Gaussian Splats for Spatially-Varying Lighting in HDR

3D Gabor Splatting: Reconstruction of High-frequency Surface Texture using Gabor Noise

SDFs from Unoriented Point Clouds using Neural Variational Heat Distances

ARAP-GS: Drag-driven As-Rigid-As-Possible 3D Gaussian Splatting Editing with Diffusion Prior

TSGS: Improving Gaussian Splatting for Transparent Surface Reconstruction via Normal and De-lighting Priors

CAGE-GS: High-fidelity Cage Based 3D Gaussian Splatting Deformation

AAA-Gaussians: Anti-Aliased and Artifact-Free 3D Gaussian Rendering

Second-order Optimization of Gaussian Splats with Importance Sampling

CompGS++: Compressed Gaussian Splatting for Static and Dynamic Scene Representation

Single-Shot Shape and Reflectance with Spatial Polarization Multiplexing

Built with on top of