The hard cases

This chapter exists because trust arises from honesty. There are subjects with which today's technology struggles — not because of bad software but for physical reasons that you now all know: almost every hard case is a violation of multi-view consistency (3.5) or a lack of features (3.3). For every case the same trio stands here: why it is hard, what works today all the same, and what is on the horizon. Whoever knows the opponents loses neither time to impossible shots nor commissions to avoidable mistakes.

7.1 Glossy and reflective — gold, metal, lacquer

Why hard: The travelling highlight (3.5, the sphere diagram). Every vantage point sees the reflections elsewhere — the witnesses contradict one another, and the training answers with ghost points, milky surfaces or artefacts floating in the air. The harder and more metallic the gloss, the worse; it hits polished metals, piano lacquer and gilding hardest.

What works today: First: defuse the gloss at the source — a light tent (light cube) wraps the object in diffuse light from all sides, hard reflections become soft shimmer, and soft shimmer the splatting training can model. Second: photograph more densely — more vantage points give the training more opportunity to separate surface from reflection. Third, only where it is permitted and reversible: matting (professionals use washable matte sprays — with cultural heritage and loans this is of course taboo, there the conservator's clearance always decides). Reckon honestly: a strongly reflective object delivers today a result with visible weaknesses.

What is on the horizon: Research is working exactly here — dedicated method branches model reflection separately from the surface (Part 2.5). DFX SplatCore follows this development actively; the table in 2.6 names the point deliberately as open. If it is solved, it is a software update — your cleanly photographed series of today stay usable for it. That too is a reason to capture well: good photos do not age, software limits do.

Two image halves with the same polished brass pot: on the left the entire living room including windows is mirrored in the surface, on the right the pot in the light tent shows only soft, even shimmer — The same metal, two worlds: on the left the pot wears the whole room on its skin — even the photographer. On the right the light tent turns hard reflection into soft shimmer: the impossible subject becomes a feasible one.

7.2 Glass and transparency

Why hard: The double problem. Through glass every witness sees a different background (violation of 3.5), and on glass the surroundings mirror at the same time (the case from 7.1 on top). The surface itself has almost no features of its own (3.3). Three violations at once — the clear wine glass is the hardest single subject there is.

What works today: Little, and this stands here as plainly as it is meant: for the faithful reconstruction of clear glass the technology today is not the right tool. Frosted, satined or coloured glass is considerably more good-natured (the less see-through, the better). Where it is only about the form, the classical detour leads to the goal: a matte duplicate or a temporary, reversible matting — with the same caveats as in 7.1. What is on the horizon: Transparency, alongside reflection, is the second great open field of research; here the same consolation holds as above — good series wait well.

7.3 Texture-poor — white ceramic, smooth surfaces

Why hard: The witness who has nothing to tell. A structureless white surface looks the same at every point — the feature search (3.3) finds no anchors it could recognise. The camera computation becomes wobbly, surfaces turn wavy or full of holes.

What works today: Astonishingly much, with two honest tricks. First: borrow context. Lay the object on a texture-rich underlay (newspaper, wood grain, a printed cloth) — the surroundings deliver the anchors the object lacks and stabilise the camera computation for the whole. Second, the old professional trick: project a pattern. A fixed-mounted projector throws a fine noise pattern onto the object; since projector and object stand still and only the camera moves, the pattern sticks in the same place for all witnesses — it is consistent and thus valid. One more note for a clear conscience: with very smooth subjects Motion Guard sometimes warns too strictly (its sharpness measurement needs structure — a documented limit). Then check the flagged images yourself at 100 percent before you distrust them.

Without structure the feature search finds no grip — a projected pattern gives it fixed anchor points, the same for every witness.

Two image halves with the same smooth white ceramic vase: left completely featureless and structureless, right with a fine projected dot pattern that gives the surface visible anchor points — Same vase, two states: on the left the feature search has nothing to hold on to. On the right a projected pattern gives it exactly that — without changing the object itself.

7.4 Fine and filigree — hair, grilles, thin struts

Why hard: Structures thinner than what a dab of colour can represent — and through whose gaps a different background shines on every image (again 3.5, in miniature). Spokes, wickerwork, fur and hair go soft, thicken or fray out.

What works today: Concentrate resolution on the detail — closer in or a longer focal length, so that the fine structure gets many pixels. Add a calm, contrasting background (dark wickerwork against a bright ground), so that the gaps at least look consistent. And steer expectations: a basket weave turns out well, individual free-standing hairs go soft — that is the state today, and it is no different at competitors.

7.5 Moving — plants, people, water

Why hard: The witness who leaves his place. The reconstruction presupposes a frozen world — the leaf that nods in the wind between image 12 and 13 is geometrically in two places at once. Result: double edges, streaks, mush on everything that stirred.

If something moves between two shots, the reconstruction asserts both positions at once — the result is double edges and mush.

Close-up of a branch with leaves: most leaves are sharp, but a cluster of leaves in the middle of the frame shows clear doubled, overlapping edges and a soft, cloudy smeared veil — the trace of motion during capture — This is what the contradiction looks like once it lands in the model: sharp all around, and exactly where something moved, a doubled, cloudy ghost image.

What works today: Stop the world as well as you can: outdoors wait for still air or shield it, indoors avoid draught and vibration, work through briskly (the shorter the series, the less time the world has to change). Water: standing works, flowing becomes a soft surface — sometimes an acceptable, almost painterly effect, but no geometry. And the special case of the human being, honestly named: a person breathes, sways, blinks — for serious portrait captures professionals use camera arrays that trigger all vantage points simultaneously. With a single travelling camera at best a short, held pose in record time succeeds — plan that as an experiment, not as a commission.

Circular camera-array rig with two rings of about twenty identical cameras each mounted on a metal frame, all firing their sync lights at the same instant; in the centre, on a round stage platform, a person stands with arms raised, face turned away from the camera — This is what "every vantage point at once" looks like: two rings of cameras firing in the same fraction of a second — no wobble, no waiting, no contradiction in time.

7.6 Oversized and walkable — rooms and buildings

Why hard: Here the capture turns around: it is not you who circles the object — the space surrounds you. With that the familiar opponents pile up: window light mixes with artificial light and changes (3.6), floors mirror (7.1), white walls are texture-poor (7.3), and the chain (3.2) must pass through every door. Doorways are the classical bottleneck: whoever steps through in three strides breaks the chain — the rooms beyond become a second island.

What works today: Systematics. Walk along the walls, view towards the centre of the room, even steps, close the loop. At every doorway extra transition images — before the door, in the door, behind the door, with visible overlap into both rooms. Decide the lighting beforehand (all artificial light or all daylight, not the switch partway through) and set the series to a time of day without a wandering sun. Large projects are planned in sections with deliberately overlapping seams. And: the room case is its own type of capture with its own decisions — the tools in DFX SplatCore continue to develop here, and a later section of this series will devote a dedicated deep dive to the walkable space.

Know the opponents:
almost every hard case is a violation of the six needs — and almost every one has a mitigation.

← Settings Next · Part 8 Field to World The capture is the only link of the chain that cannot be repaired from afar. Whoever can capture holds the key.