Gaussian Splatting
Representing a 3D scene as a cloud of colored, oriented Gaussians - efficient rendering of novel views.
Think of it like
Instead of tracing rays, you paint with blurry paint drops (Gaussians) positioned in 3D space.
Example
Multi-view photos -> optimize Gaussian positions, colors, covariance -> render novel viewpoints in real-time.
How it actually works
Each Gaussian: center, covariance (orientation/scale), color. Volume rendering sums contributions along each ray. Differentiable rasterization enables optimization. Much faster than NeRF (10-100x speedup), real-time quality, fewer views needed. Trade-off: requires careful initialization (SfM points or coarse NeRF), can have artifacts (floaters, oversplatting). Emerging as practical alternative to NeRF.
For product teams
Real-time 3D rendering, VR/AR, game engines, streaming 3D content.
For engineers
Differentiable splatting + sorting-based rendering. Faster than NeRF. Requires CUDA. Metric: same as NeRF (PSNR, LPIPS).
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome