Same data and features on CesiumJS — for the side-by-side performance comparison. Keeps URL parameters
Off = model-view in the shader (float32); tiles jitter on ECEF rounding. Automatically off during loader and flight
Diagnostics only: off lowers the cloud, the height uniforms do not follow
Feasibility test: Gaussian-splatting model (Spark) in its own WebGL overlay, 61 MB — click to look around
Protected parcel from GeoJSON. Column = light column through the canopy, X-ray = footprint shines through the trees, Canopy = plate on the crowns, Wall = low glowing wall. Switching re-frames the camera
Exact = the surveyed staircase with every 1 m² cell line — the authoritative boundary. Organic = rounded, area-preserving, a stylised reading of the same parcel
Radius of the disc the outline is opened and closed with, 0 = untouched staircase
Automatic nearby weather cycle
Checking device headroom
Optimizations · Cesium comparison
Everything off — only the point cloud with zoom-dependent density, navigation and basemap. Individual choices below are kept
Reloads with ?compare=1: no device benchmark, no startup resolution cap — the full, fair comparison
Design
Basemap
Allowed error in screen pixels. 1 = pixel-perfect and expensive, higher = softer and far fewer tiles (measured: 23 tiles at 1, 16 at 2, 9 at 4). The point cloud itself runs at 8
Ground fog
Past 100% the fog saturates earlier instead of overshooting its colour — the coverage itself is capped at 100%
Falloff upward from the base (1/e height)
0 = fog reaches the ground at full density · higher = a band hanging in the canopy with clear air beneath
1 = physical (Beer-Lambert) · <1 fog starts early · >1 the near field stays clear and density moves into the distance
Smaller = denser. e-folding distance along the fog base, not a cutoff. 0 = full coverage immediately
0 = follows the daylight automatically · 1 = a fixed colour around the clock
Ground under the point cloud
Replaces the satellite imagery with a solid colour where point data exists — the shape follows a coverage mask built from the tileset cells, not the survey rectangle
Not graded by daylight, but fog and the vignette still apply to it
100% = imagery fully replaced · below that the terrain shows through
Pulls the patch inward. Higher = more basemap stays visible at the edge. Low can let it spill past the points
Width of the transition. 1% = hard edge, higher = a soft ramp across roughly the 50 m mask resolution
Distance fog
Fraction of the view distance where the haze starts. Small = haze begins right at the camera
Fraction of the view distance where everything turns sky-coloured — this hides the map edge. Always kept above the onset
Both are fractions of the current view distance, not metres — so the haze keeps the same proportions over the canopy as it does 90 km out. Colour follows the daylight.
Cloud shadows
Base depth of the shadows. Additionally multiplied by daylight, and halved when the cloud layer is off
Metres per noise period. Small = fine, busy dappling · large = broad continental shadows
Tightens the noise ramp. 0 = soft, washed blotches · high = clearly defined cloud gaps
Vignette & surround
0 = a hard cut, higher = points dissolve toward the edge
>1 thins out early (widely scattered) · <1 holds density up to the edge
Tints the points and the map themselves, not just the overlay
Vignette position
Pitch below which the vignette is fully pinned to the camera (side view)
Pitch above which the vignette fully follows the ground point at screen centre (top-down)
Moves the side-view anchor ahead of the camera; 0 = exactly on the camera
Stops the vignette from clipping nearby points in side view
Below 95% points never vanish abruptly, only through the fog in the distance
Depth of Field
A real post pass with a blur pyramid — the only effect that does not run in the shader. Off = no cost
On = the focal plane follows the point at screen centre · Off = a fixed distance from the camera
With autofocus an offset from the aimed point (negative pulls focus nearer), otherwise the absolute distance
Distance past the focal plane to full blur. Small = shallow and cinematic, large = almost everything sharp
Width of the blur kernel — also the main cost driver
Small = the focal plane eases in, large = it snaps