WHY IT MATTERS

40.0 million triangles.
In a browser tab.

Open a scan here and you are moving through the complete survey: every triangle the photogrammetry produced, at full texture resolution, with nothing thrown away. No download, no software, no workstation. That sounds ordinary until you look at what is actually being delivered.

WHAT IS IN ONE SCAN

The real numbers.

Measured 2026-09-18 with nxsedit.

ScanTrianglesPointsPublished sizeExtent
Tapadla Chromite Mine40.0 M21.5 M630 MB~86 m across
Maria Concordia Magnesite Mine20.0 M12.1 M544 MB~758 m across
Marie Agnes Mine18.4 M10.1 M412 MB~336 m across

Extent is the width of the sphere that encloses the whole site. Those published sizes are already compressed; the source photographs and the working models behind them run to hundreds of gigabytes.

FOR COMPARISON

How big is 40 million, really?

A whole video game level is usually built from fewer triangles than one of these scans. The difference is that a game is modelled by hand to be cheap to draw; a scan is measured, and measurement does not simplify itself.

WHY IT IS HARD

Why a normal 3D viewer cannot open this.

THE USUAL ANSWER

Throw the survey away, or keep it to yourself.

Cut it down

General 3D galleries cap what you can upload and expect models in the hundreds of thousands of triangles. Getting 40.0 million down to that means keeping roughly one triangle in every 160. What survives is the shape of the passage. What goes is everything that made it worth documenting: chisel marks, bolt holes, the edge of a collapse, the grain of a timber support.

Or keep it on a workstation

The alternative is the full model in specialist software: a multi-gigabyte download, a licence, a capable computer and someone who knows the program. That works for the survey team. It does not work for a museum visitor, a funder, a heritage officer, a journalist or a diver's family.

So the record either loses its detail, or it never leaves the hard drive. Most underwater documentation ends up in one of those two places.

WHAT WE DO INSTEAD

Send only what the eye can use.

  1. The scan is rebuilt as a tree of detail levels

    Before publishing, the model is turned into thousands of small pieces at several levels of coarseness, each piece compressed on its own.

  2. The browser asks only for what is visible

    As the visitor moves, the viewer works out which pieces fall inside the view and how much detail each deserves at its distance, then requests exactly those byte ranges from the server.

  3. Detail arrives while you keep moving

    Coarse pieces show immediately and sharpen as finer ones arrive. Textures are decoded away from the main thread and uploaded in small batches, so the picture keeps moving rather than freezing.

  4. What leaves the view is released

    Pieces you have passed are dropped from memory before it fills, which is what keeps a 16 GB laptop, and a phone, from running out.

Only the parts of the scan near the viewer are loaded at full detail
Full detail near the viewer, coarse pieces further away, nothing at all behind.

The result: a scan begins rendering one to two seconds after you open the link, on an ordinary laptop over a normal connection, and sharpens as you explore. Phones get the same scan with a smaller memory budget. For the technical version, with commands and numbers, read how to stream a 3D model with 100 million triangles.

WHY IT MATTERS

What this changes.

SEE IT FOR YOURSELF

Then judge whether it feels
like a web page or a dive.