Photo overlap is the share of a scene that two neighbouring photos have in common, and photogrammetry software needs it to match the same surface points across images and triangulate them. The published minimums are close to each other: Meshroom asks for at least 80% frontal and 60% side overlap, Agisoft Metashape uses the same 80/60 ratio, and Pix4D recommends at least 75% frontal and 60% side for the general case.
For a small object you photograph by walking around it or by using a turntable, percentages are hard to measure. Pix4D’s rule for buildings, the closest it gets to a 3D object, is angular: take one image every 5 to 10 degrees, then go round again at higher viewpoints. That is 36 to 72 photos per ring (360 divided by 10 or 5). Below you will find the figures from each vendor, how to adjust them for difficult surfaces, and what to check when the model comes out with holes.
The consumer toolchanger 3D printer: the SnapSwap 4-toolhead system swaps heads in 5 seconds for true multi-color and multi-material printing, with smart calibration and auto filament loading.
- 4 toolheads · 5 s swap
- Up to 500 mm/s · 20,000 mm/s²
- 270 mm cube build volume
- Multi-color & multi-material
Overlap figures from the software makers
Frontal (or forward) overlap is the overlap between consecutive photos along one track, such as along one ring of an orbit. Side overlap is the overlap between neighbouring tracks, such as between a low ring and a higher ring. Here is what each vendor states:
| Source | Frontal overlap | Side overlap | Notes |
|---|---|---|---|
| Meshroom manual (capturing) | at least 80% | at least 60% | Move to a new position (or rotate the object) for every shot; do not shoot several images from the same spot |
| Agisoft Metashape tips | 80% | 60% | Stated for aerial photography and used as the general ratio; do not go below it. The close-range object page gives no percentage |
| Pix4D (general case) | at least 75% | at least 60% | Regular grid pattern at constant height |
| Pix4D (flat, homogeneous terrain such as fields) | at least 80% | at least 80% | Few unique features to match |
| Pix4D (trees, dense vegetation) | at least 85% | at least 85% | Appearance changes a lot between photos |
| Pix4D (building) | one image every 5 to 10 degrees | extra rings | Second and third rings at higher flight height, with the camera angled down more each time |
| RealityScan Mobile | not given as a percentage | not given as a percentage | Cover every part with high overlap; circles and arches at several heights, at least half of each image should be the object |
Treat these as minimums and starting points, not guarantees. Pix4D itself notes that its software may work with lower overlap than recommended, but suggests using the recommended values to get the best results. Metashape adds a reason to over-capture: it can reconstruct only geometry visible from at least two cameras, so each blind zone becomes a hole.
Overlap for an object you walk or turn around
The percentage idea comes from aerial mapping, where the camera moves along flat tracks. For a round object the shared area between two photos shrinks quickly as you move around it, because surfaces turn away from the lens. That is why Pix4D switches to degrees for objects and buildings. A practical routine, based on those figures:
- Ring 1, camera at about object height: one photo every 5 to 10 degrees. On a turntable, that means rotating by 5 to 10 degrees between shots.
- Ring 2, camera higher and tilted down: the same angular step. Pix4D raises the viewpoint and lowers the camera angle with each round.
- Ring 3, camera lower or higher still, to catch undersides, undercuts and tops. One EinScan workflow used three rings at 10-degree intervals and ended with nearly 100 images.
- Close-ups of fine detail, with the same overlap between them and the wider shots so the software can tie them together.
- Check the flip test: open two consecutive photos and flip between them. Most of the object should still be recognisable and shifted only a little. If a feature jumps across the frame, the step is too big.
The flip test is our rule of thumb, not a published figure. A useful minimum for framing comes from RealityScan Mobile: at least half of each image should be the object. Metashape and Meshroom both say the object should fill most of the frame.
Photo counts
There is no ideal number, only the angular step multiplied by the rings. A 10-degree step on three rings is about 108 photos, and a 5-degree step on two rings is 144. Meshroom’s manual says more images are better because you can filter out repeated or poor ones; Agisoft says the same (you can disable excess images later). RealityScan Mobile, in its default mode, caps a capture at 300 images and starts analysing after 20. More photos cost processing time, and photos taken from nearly the same spot add little new geometry.
When to push overlap higher
| Subject | Why matching gets harder | What to change |
|---|---|---|
| Plain, untextured surfaces (white plaster, painted prints) | Few unique features | Raise overlap (Pix4D uses 80% for homogeneous terrain), add texture or markers, soft light |
| Shiny, glass or transparent surfaces | Reflections move with the camera | Overlap alone will not fix it: see photogrammetry of shiny objects |
| Foliage, hair, fine lattices | Complex, changing appearance | Pix4D: 85% frontal and side, more oblique angles |
| Repeating details (identical windows, bricks) | Software matches the wrong parts | Keep overlap high and add unique markers (Agisoft) |
| Corners and edges where surfaces meet | Fewer shared points between walls | More-than-average overlap at corners (Agisoft) |
| Deep recesses and undercuts | Blind zones | Extra low and high angles aimed into the hollow |
Camera settings that protect overlap
Overlap does not help if photos are blurry or inconsistent. Agisoft’s tips give these starting points: use a fixed lens or a focal length in the 20 to 80 mm range (35 mm equivalent), the lowest practical ISO, an aperture of f/8 to f/11 for enough depth of field, and a shutter speed fast enough to avoid blur, with a tripod if light is poor. Avoid flash and light that changes between shots, and keep light sources out of the frame. Metashape also processes only unmodified photos: cropped or warped images are likely to fail, although mild brightness or contrast changes are fine. Meshroom adds: do not change the zoom while shooting, and avoid moving objects in the scene.
If you rotate the object while the camera and background stay still, the static background confuses the alignment. Meshroom recommends a plain-coloured background with no distinguishable features on a rotating rig; Agisoft suggests shooting blank frames of the empty background so Metashape can mask it automatically. For more on gear see camera setups for photogrammetry.
Diagnosing a failed scan
| Symptom | Likely cause | Fix |
|---|---|---|
| Alignment splits into several pieces | A gap in coverage between rings or sides | Add transition photos where the pieces meet, with 60% or more shared view |
| Holes in the mesh | Surface seen from fewer than two cameras | Shoot extra angles at the hole, including low and high rings |
| Warped or flat result | Zoom changed, or many shots from one spot | Lock the focal length; move for every shot |
| Noisy or blobby surface on glossy areas | Reflections, not overlap | Use cross-polarisation or a matte coat (see the shiny-objects guide) |
| Soft texture | Blur or shallow depth of field | Higher f-number, tripod, lower ISO |
| Part of the background merges into the model | Static background while turning the object | Mask the background or use a plain backdrop |
What to do after the scan
Overlap affects the raw model. Scans usually still need cleanup before printing: check the mesh error checklist, close gaps (the hole-closing guide covers one approach), and reduce the triangle count with mesh decimation if the file is too heavy. Photogrammetry is one of several types of 3D modeling you can use as a starting point for a print.
Frequently asked questions
What overlap percentage should I use for photogrammetry?
Aim for at least 80% between consecutive photos and at least 60% between neighbouring rings or tracks (Meshroom and Agisoft). Pix4D uses 75% and 60% for the general case and raises it to 80% or 85% for flat or leafy scenes. For small objects, think in angles: a photo every 5 to 10 degrees.
Can I use too much overlap?
Not really: Meshroom and Agisoft both say more photos are better than too few, because you can filter or disable the extras later. The costs are capture time, storage and processing time. Photos from almost the same position add little new information, so shoot more angles rather than more of the same view.
How many photos do I need for a small object?
One photo every 10 degrees on one ring is 36, and three rings are about 108, which matches the roughly 100 images in one cross-polarised EinScan workflow. Every 5 degrees doubles that. Check the alignment result and add photos where the model has gaps.
Does the overlap percentage matter on a turntable?
The idea still applies, but the useful unit is the angle between photos. Use 5 to 10 degrees per step as a starting point, keep the object large in the frame, and use a plain or masked background so the software does not track the static backdrop.
Sources
- Meshroom manual: Capturing
- Agisoft Metashape: Image capture tips, equipment and shooting scenarios
- Agisoft Metashape: Close-range object scanning tips
- Agisoft Metashape: Rotor-wing drone tips (building corners, identical details)
- Pix4D: Selecting the image acquisition plan type
- Epic Games: RealityScan Mobile step-by-step guide
- EinScan: Dealing with specular reflections by cross-polarization photography
