Research

SplatLab, a NeurIPS 2026 system from UC San Diego, reproduces a published 3D Gaussian splatting baseline, extends it, and writes a paper tied to logged runs.

SwitchSplat splits Gaussian splatting scenes into a static base and switchable on/off object slots, rendering state combinations never seen during capture.

MLP Splatting from Imperial College London swaps Gaussians for small neural primitives that align with objects, and its code is now out under an MIT license.

ChronoFuseGS from TU Wien merges separately trained Gaussian splatting captures of a changing site into one model and highlights changes at sub-object detail.

aussian Light Transport fits global illumination as a mixture of 13D Gaussians, training 10 to 23 times faster than Neural Radiosity in a few megabytes.

KISS-GS is a modular 3DGS compression pipeline from Fraunhofer HHI that cuts scenes from 738 MB to 3.2 MB, decoded with nothing but ordinary browser images.

One foam to rule them all.

Radiance field reconstruction just got stronger.

A new radiance field representation has just been published!

Another new Radiance Field representation has been published.

DBS is a new radiance field representation that carries some big advantages over 3DGS.

NVIDIA has published a paper greatly helping detail recovery thanks to Generative AI.