You could skip this chapter and page straight on to the camera settings. But then you would miss the best part — and the most useful. Because the technology you are about to work with is no box of tricks that fell from the sky five years ago. It is the most recent chapter of a very old story: the attempt to grasp the world from vantage points. Anyone who knows this story understands not only why the capture rules are the way they are — they also lose the last remnant of awe before the supposed black box. It is geometry. It is old. And it is beautiful.
The timeline of this chapter. Each station is told in what follows — and each leaves a trace right up to your camera of today.
Let us begin with a picture you can still see today from an aeroplane over northern Italy: fields, roads and ditches that follow a strict grid for kilometres on end. This grid is roughly two thousand years old. Roman land surveyors laid it out — with a device that from today's perspective looks touchingly simple: the groma, a cross of wooden arms with hanging plumb lines, with which exact right angles could be staked out. With it they divided conquered territories into even squares (the centuriatio), aligned roads dead straight over hundreds of kilometres, and laid out towns whose ground plans survive to this day.
Their toolbox held more: the dioptra, developed in Greek antiquity and described by the engineer Heron of Alexandria, could measure angles — an early forerunner of the theodolite that surveyors still use today. With the chorobates, a long spirit level, Roman engineers checked the fall of their aqueducts to fractions of a degree — otherwise water conduits over fifty kilometres would simply have been impossible.
What has this to do with photography? Everything — for here the core idea is born from which all that follows descends: you can measure the world by sighting it from several fixed positions. Two observers who see the same tower know its distance from their two viewing angles and the gap between them — without ever going there. That is at heart triangulation, and it is exactly the principle with which DFX SplatCore computes the position of every image point in space from your photos. The difference lies not in the principle but in the scale: where two Roman surveyors sighted one tower, today hundreds of photos sight millions of points.
Triangulation, the primal principle: what two surveyors here do with one tower, DFX SplatCore does with every image point of your photo series — a million times over.
Honesty in the detail — were the Romans photogrammetrists? No — and such things are taken seriously here, precisely because Part 5 sets out to clear up myths. The Romans were masters of surveying: positions, angles, sight lines. But they never measured from images — for that they lacked both: the mathematical understanding of perspective and an image one can trust. Both arrive only in the next two stations. The Romans are the root of surveying, not of photogrammetry — but without this root the whole tree would not exist.
A millennium and a half later, Florence, around 1420. The master builder Filippo Brunelleschi — the same man who would later raise the famous cathedral dome — stages an experiment that changes the history of art: he paints the Baptistery exactly as it appears from a particular position, bores a hole in the panel and lets viewers look through this hole onto a mirror reflecting the painting. Painted image and real building become indistinguishable. The message: an image is not a free invention — it follows laws, and whoever knows them can depict reality exactly.
What Brunelleschi demonstrated, Leon Battista Alberti wrote down in 1435 in “De pictura” — the first mathematical description of central perspective. His famous image for it: the painting is an open window through which one sees the world; the lines of sight from the eye to the objects pierce the window plane, and where they pierce it lies the image. Piero della Francesca deepened the mathematics behind it, and Albrecht Dürer finally built apparatus that demonstrated the principle mechanically — his woodcuts of 1525 show draughtsmen sighting through viewfinders and thread grids to transfer reality point by point onto the image plane.
Let us note what — centuries before the first camera — was already fully understood here: a perspectival image is a geometric projection with precisely determinable properties. Every camera you pick up today is at heart Alberti's window: a plane (the sensor) through which lines of sight (the light through the lens) depict the world. When DFX SplatCore speaks of “camera parameters”, it means exactly the quantities that describe this window — where it stands, where it points, how wide it opens.
Alberti's window (1435). Replace “eye” with “sensor” and “window” with “lens image plane” — and you have the camera model that DFX SplatCore computes with.
So far the calculation ran in one direction: world → image. The next step is the decisive one, and it is easily overlooked. First the triangulation of the old surveyors was built up into a science: Gemma Frisius described in 1533 how a whole country is mapped over chains of triangles, and Willebrord Snellius at the start of the 17th century surveyed an entire stretch of land this way for the first time — the method with which continents were later mapped.
But then, in 1759, Johann Heinrich Lambert — mathematician, physicist, autodidact — published a work with the lovely title “Die freye Perspektive” (“Free Perspective”). In it lies, almost in passing, a revolutionary thought: perspective can be computed backwards. From a perspectivally correct image one can determine from where it was taken — the draughtsman's position and viewing direction follow from the image itself. The field calls this today resection, and it is exactly the operation DFX SplatCore performs for each of your photos: where did the camera stand when this picture was made?
Take a moment to grasp it: the complete mathematics of photogrammetry thereby existed eighty years before photography. What was missing was not understanding — what was missing was an image whose perspective one could believe. A hand drawing is never exact enough; the draughtsman errs, interprets, embellishes. The world was waiting, without knowing it, for an image that draws the geometry itself.
In 1839 Louis Daguerre presented the first practicable photographic process in Paris (Nicéphore Niépce had already laid the foundations in the 1820s). For art this was a sensation. For surveying it was the fulfilment of a century-old wish: at last an image that does not interpret perspective but records it with physical exactness. Light falls through a lens onto a plate — that is Alberti's window, only without the erring draughtsman.
It took less than a decade before someone drew the connection. The French officer Aimé Laussedat began around 1850 to make maps from photographs — he called it métrophotographie and is regarded today as the father of photogrammetry. Note where he came from: from land surveying. The new discipline arose not in the photo studios but among the cartographers — as a direct continuation of the line that began with the groma.
The German thread of this story begins with a near-accident. The young master builder Albrecht Meydenbauer surveyed the cathedral of Wetzlar in 1858 — in the way usual at the time: by hand, hanging from a rope against the façade. In doing so he almost fell to his death. The experience led him to a thought that from today's perspective sounds self-evident and back then was bold: why clamber about on the building when one can photograph it and survey it at the desk? Meydenbauer developed his own metric cameras, coined the term “photogrammetry” in 1867 — and founded the Royal Prussian Institute of Photogrammetry in 1885, the world's first archive to record architectural monuments systematically in metric images.
Why this institute was more than a public office is shown by the twentieth century: when wars and catastrophes destroyed buildings, Meydenbauer's glass plates became the basis on which what was lost could be rebuilt or at least documented. Hold on to this thought — a good image is an insurance against loss. It will return when we come later to the digitisation of cultural heritage. What Meydenbauer began with glass plates for cathedrals, you can do today with a camera for anything that matters to you.
AI image
For a century photogrammetry remained a craft for specialists: metric cameras, stereo viewers, room-filling analysis machines at which trained operators drew maps from aerial images point by point. The mathematics was the old one — triangulation and resection — but every point cost handwork.
Then came the computers, and with them the last barrier fell. In the 1990s image processing learned to recognise distinctive points on its own — the corner of a window, a spot on a wall, a thousand times per image, without human involvement. With that Lambert's resection could be automated: the computer finds the same points in many photos and computes from them simultaneously the positions of all cameras and the location of all points in space — a process called Structure from Motion (“structure out of movement”). In 2006 the project Photo Tourism showed something astonishing at the time: from ordinary holiday photos of strangers from the internet, the cathedral Notre-Dame could be reconstructed spatially. In 2016 the free software COLMAP bundled the state of this research — and exactly this COLMAP works today in DFX SplatCore when you import a large image series.
Let the span of this arc sink in for a moment: the resection that Lambert described in 1759 for a single drawn image runs today in minutes over hundreds of photos and millions of points. Nothing about it is magic — it is the same geometry, only unimaginably faster. When DFX SplatCore reports to you after the import that it has computed the camera positions, it has done exactly what a Roman surveyor would have understood in principle. He would only have needed a little longer.
Up to here this story tells, for two thousand years, one thing in ever new guises: measuring. Where is the point? How high the tower, how far the road, how deep the room? The result was numbers, maps, plans, point clouds — images for the mind.
Then, around the year 2020, something fundamentally new happens, and it is important to recognise the break as a break: research asks a different question. No longer “where is each point?”, but “how does this scene look from any arbitrary vantage point?” — not measuring, but reproducing. The result is not a plan but an experience: one moves freely through a photorealistic scene as though standing within it. How this came about, which ideas overtook one another in the process and why of all things a method called Gaussian Splatting won the race — that is the story of the next part.
For you as a photographer this break has a very practical meaning that runs through all the following chapters: some capture rules from the measuring age still hold unchanged (the geometry still needs vantage points and sharpness), others have become obsolete, and yet others are more important than ever. Which rule belongs in which drawer is clarified by Part 5 — the myths chapter. First, though, we take a closer look at the newcomer.
