Glossary

35 terms

alpha blending: A compositing technique that accumulates colour from ordered, semi-transparent primitives, each weighted by its own opacity and by how much light nearer primitives have already absorbed. Splat rasterisers use it to blend overlapping Gaussians into a final pixel colour.
anisotropic: Having properties that differ by direction, as opposed to isotropic (the same in every direction). An anisotropic Gaussian's covariance can stretch it into a thin disc or needle to match a surface, rather than staying a uniform sphere.
bundle adjustment: The global optimisation step that jointly refines every estimated camera pose and every triangulated 3D point to minimise total reprojection error. COLMAP re-runs it repeatedly during incremental reconstruction as new images register.
covariance matrix: A matrix describing a Gaussian's spread and orientation in space: its terms define how far the ellipsoid extends along each axis and how those axes are tilted, together encoding size, shape and orientation as a single object.
densification: Training-time process that clones and splits Gaussians in under-reconstructed regions and prunes low-opacity ones.
draw call: A single CPU-issued command asking the GPU to render a batch of geometry. Splat renderers draw an entire scene as instanced quads in very few draw calls, so frame cost concentrates in sorting and blending rather than call count.
epipolar geometry: The geometric constraint that, given two calibrated views of a rigid scene, a point's match in one image must lie along a specific line in the other image. COLMAP uses it to verify proposed feature matches and reject geometrically inconsistent ones.
exposure lock: Fixing exposure, and ideally white balance, to one value for an entire capture pass rather than letting the camera re-meter every frame. Prevents auto-exposure from making the same surface look different across photos, which otherwise confuses feature matching.
feature matching: Comparing feature descriptors across images to propose which points likely correspond to the same physical location, prior to geometric verification. COLMAP supports exhaustive, sequential, vocabulary-tree, spatial and transitive matching strategies.
floater removal: The post-training cleanup step of deleting floaters from a trained scene, either manually by box or lasso selection in an editor such as SuperSplat, or automatically via a statistical outlier filter such as splat-transform's floater and cluster filters.
floaters: Small, semi-transparent Gaussians left floating in empty space, disconnected from any real surface, that appear as speckling or haze from viewpoints the optimiser wasn't trained against; typically caused by insufficient view coverage, reflections or moving objects the optimiser had no other consistent way to explain.
Gaussian splat: A single anisotropic 3D Gaussian, an ellipsoid with position, covariance, opacity and view-dependent colour, used as the basic rendering primitive of a splat scene. Millions of these together, optimised against training images, make up the scene.
gradient descent: An optimisation method that repeatedly nudges parameters in the direction that most reduces a loss function, following that loss's gradient with respect to each parameter. Splat training uses it, via backpropagation through the differentiable rasteriser, to adjust every Gaussian's position, shape, opacity and colour.
LOD: Level of detail: rendering or serving a reduced-detail version of an asset where full detail would be wasted, such as lower Gaussian-count representations of splat scene regions far from the camera.
loop closure: Matching images from non-adjacent points in a capture path, most usefully the end of a loop back to its start, so bundle adjustment can correct pose error that accumulates along a long sequential walkthrough instead of letting it compound one-directionally.
LPIPS: Learned Perceptual Image Patch Similarity: a perceptual distance metric computed from pretrained deep-network feature activations rather than pixel values, correlating substantially better with human judgements of image similarity than PSNR or SSIM.
NeRF: Neural Radiance Field: a scene representation that encodes colour and density as the weights of a neural network, rendered by marching rays through the volume and querying the network at each sample point.
opacity pruning: Removing Gaussians whose opacity has fallen below a small threshold during training, on the assumption that a near-transparent Gaussian contributes nothing to any rendered pixel. Keeps splat counts from growing unboundedly alongside densification.
overlap: The fraction of a photo's frame shared with its neighbours in a capture sequence. Structure-from-motion needs each surface point visible from several overlapping viewpoints to triangulate its position, making overlap the primary lever for reliable COLMAP registration.
parallax: The apparent shift in a feature's position between two photos caused by the camera's optical centre moving through space. It is the signal structure-from-motion triangulation depends on; rotating a camera on a fixed point produces overlap without usable parallax.
PLY: Polygon File Format: the element/property container every 3DGS trainer writes its raw output to, one row per Gaussian holding position, scale, rotation, opacity and spherical-harmonics colour. Lossless and universal as an interchange format, but too large to ship straight to a browser.
PSNR: Peak Signal-to-Noise Ratio: a pixel-level fidelity metric, ten times the base-10 logarithm of the maximum pixel value squared over the mean squared error, expressed in decibels; higher is better but it correlates poorly with perceived quality since it weighs every pixel equally regardless of what a viewer actually notices.
quantisation: Storing a value at lower numeric precision than it was computed at, for example rounding a float32 position to 16 bits or mapping a scale to one of a fixed set of codebook entries, to shrink storage size at the cost of some accuracy.
radiance field: A function that maps a 3D position and viewing direction to a colour and density or opacity, describing how a scene looks from any viewpoint. Both NeRFs and Gaussian splats are representations of a radiance field, one implicit, one explicit.
rasterisation: Converting geometric primitives directly into pixels on screen, rather than tracing or marching rays through a scene. 3D Gaussian Splatting adapts a tile-based GPU rasterisation pipeline to render Gaussians instead of triangles.
segmentation: Assigning Gaussians in a trained scene a discrete object or category identity so a specific object can be selected, isolated or edited independently of the rest of the scene. Current methods derive these identities by lifting 2D masks, typically from the Segment Anything Model, into 3D, either as a whole-scene labelling pass or as a per-click prompted query.
SfM: Structure-from-Motion: recovering camera poses and a sparse 3D point cloud simultaneously from a set of overlapping photographs, using matched features, epipolar geometry and triangulation. COLMAP's incremental mapper is the SfM implementation most splat pipelines rely on.
SH bands: The successive groups of spherical-harmonics coefficients, band 0, band 1, band 2 and so on, each adding 2l+1 coefficients per colour channel; band 0 gives flat diffuse colour and higher bands add increasingly directional, specular-like variation at increasing storage cost.
SOG: Compressed splat format storing Gaussian attributes as sorted 2D image grids (typically WebP), giving large size reductions for web delivery.
spherical harmonics: Basis functions on the sphere used to encode view-dependent colour per Gaussian; band 0 is diffuse colour, higher bands add specular-like variation.
SPZ: Niantic's open, MIT-licensed compressed splat format: roughly a tenth the size of the equivalent raw PLY through per-attribute quantisation and parallel entropy coding, while keeping full spherical-harmonics detail per Gaussian rather than collapsing it into a shared palette.
SSIM: Structural Similarity Index: an image-quality metric that compares luminance, contrast and structure within local windows rather than raw per-pixel error, better matching how human vision judges similarity between two images.
streamed loading: Loading a scene progressively, showing a coarse but complete version almost immediately and refining as more data arrives, instead of blocking on one atomic download of the full asset.
triangulation: Computing a 3D point's position from two or more matched, pose-known camera views by finding where the rays cast through each observation intersect, or pass closest to each other once pose and pixel noise are accounted for.
WebGPU: The modern web graphics API succeeding WebGL, whose compute-shader support makes GPU-side splat sorting possible in a browser. Shipped by default in Chrome and Edge since 2023 and, as of the 26-generation Apple OS releases, in Safari.