Gaussian Splatting
Overview
Gaussian Splatting represents a scene as many learnable Gaussian primitives, which are projected and composited to render new views efficiently. Methods refine these primitives through differentiable rendering, often using densification and pruning, while newer approaches add geometry priors, planar primitives, or feed-forward prediction. The field spans static novel-view synthesis, dynamic 4D reconstruction, and SLAM, with evaluation on benchmarks such as Mip-NeRF 360, DTU, N3DV, and KITTI. A central challenge is balancing rendering quality and geometric accuracy with memory, speed, and reliable reconstruction from sparse or changing views.
Latest papers
83 papers3 papers in the last four weeks, down 25% on the four weeks before. 0.0% of all new papers.