Appearance
When photogrammetry fails, the cause is almost always in the photographs, not the software. The usual suspects are too little overlap, blurry or inconsistently lit images, or a subject that is reflective, transparent, featureless, or moving. Work the problem in order: first confirm your input is sharp and well-overlapped, then mask distractions, then adjust alignment settings. Tweaking accuracy sliders on bad photos wastes time — fix the input and most "failures" disappear.
Why did alignment fail in the first place?
Alignment works by finding the same feature points across multiple photos and triangulating the camera positions. It fails when the software cannot find reliable matches. Run through this diagnostic order:
- Overlap — does each surface point appear in at least three photos (60–80% overlap)? Sparse coverage is the top cause.
- Sharpness — any motion blur or out-of-focus frames? Blurred features will not match. Remove them.
- Subject features — is the surface textured, or is it a blank white wall, glass, or chrome? Featureless and reflective surfaces have nothing stable to match.
- Movement — did the object, lighting, or background change between shots? Inconsistency breaks matching.
- Lighting — were exposures consistent and diffuse, with no hard moving shadows?
Most failures resolve at steps 1–3 before you ever touch a setting.
How much overlap and how many photos do I need?
Think coverage, not count. Every point on the object should be visible in at least three images, which in practice means 60–80% overlap between consecutive shots. For an object on a turntable, that is roughly a photo every 10–15 degrees, repeated at two or three heights. A pillar or statue needs orbits at low, mid, and high angles plus the top. When in doubt, take more photos at closer spacing; redundancy is cheap, gaps are fatal.
My camera only aligned part of the set — how do I fix split alignment?
Split or partial alignment means the software solved two clusters it could not connect, usually because of a coverage gap or a feature-poor band between them. Fixes, in order:
- Add bridging photos that overlap both clusters — shots taken across the gap so common points tie the halves together.
- Mask moving backgrounds (people, foliage, your own shadow) so the matcher ignores them and focuses on the static object.
- Increase tie points. In Metashape, raise Key point limit (e.g. 60,000) and Tie point limit (e.g. 10,000) and set alignment accuracy to High.
- Re-align the unaligned cameras specifically, rather than resetting the whole project.
text
Metashape alignment defaults that recover split sets:
Accuracy: High
Generic preselection: On
Reference preselection: Off (unless you have GPS)
Key point limit: 60000
Tie point limit: 10000
Then: right-click unaligned cameras > Align Selected CamerasWhy is my surface bumpy, melted, or full of holes?
A melted or holey surface is downstream evidence of an upstream problem: the sparse cloud was weak, so the dense cloud inherited the error. Causes and remedies:
- Holes — that area was under-photographed. Add close, overlapping shots of it and rebuild.
- Melted blobs — glossy or transparent surfaces gave inconsistent appearance; relight diffusely or matte the surface.
- Bumpy noise — too few tie points or poor camera calibration; re-align at higher accuracy, then rebuild depth maps.
Never try to fix geometry by smoothing the final mesh if the alignment was poor; rebuild from better input.
Do reflective and transparent objects always break it?
They are the classic hard case because their appearance changes with viewpoint, so a "feature" is never in the same place twice. Strategies that work:
| Problem surface | Mitigation |
|---|---|
| Glossy ceramic, metal | Cross-polarisation (polariser on lens + lights) |
| Glass, transparent | Temporary matting spray (only on non-precious items) |
| Dark, featureless | Brighter diffuse lighting, raise exposure |
| Repetitive pattern | Add unique markers / coded targets nearby |
Never spray or coat a precious artefact; for those, switch technique to structured-light or laser scanning instead of forcing photogrammetry.
What is the fastest path back to a usable result?
When a run fails, resist re-running with the same photos and a higher accuracy setting. Instead:
- Cull blurred and badly-exposed frames.
- Identify and re-shoot any coverage gaps (bring the object back if you can).
- Mask backgrounds and moving elements.
- Re-align at High accuracy with raised point limits.
- Only then rebuild the dense cloud and mesh.
Input quality dominates outcome; settings are the final 10 percent.
Key Takeaways
- Failed photogrammetry is almost always a photo problem, not a software problem.
- Diagnose in order: overlap, sharpness, subject features, movement, lighting.
- Aim for 60–80% overlap so each point appears in at least three sharp photos.
- Fix split alignment with bridging photos, masking, and higher key/tie point limits.
- Melted or holey surfaces signal weak input — rebuild from better photos, do not just smooth.
- Reflective/transparent surfaces need polarisation, matting, or a switch to scanning.
- Improve the photographs first; raise alignment accuracy only as the final step.
Frequently Asked Questions
Why did my photogrammetry alignment fail?
The most common causes are too few overlapping photos, a featureless or reflective subject, motion blur, or a moving object. Photogrammetry needs ~60-80% overlap between sharp, well-lit images of a static, textured surface to match points.
How much overlap do my photos need?
Aim for 60-80% overlap between consecutive shots, so any surface point appears in at least three photos. Sparse coverage is the number-one reason alignment fails or produces a broken, partial model.
My camera only aligned half the photos — what now?
Split alignment usually means a gap in coverage or a feature-poor region. Add bridging photos that overlap both clusters, mask out moving backgrounds, and re-run alignment with higher accuracy or more key/tie points.
Why is my reconstructed surface bumpy, melted, or full of holes?
That points to weak alignment, glossy or transparent surfaces, or insufficient coverage of that area. Improve the photos and re-align; a dense cloud built on a poor sparse cloud will always look melted.
Do reflective or shiny objects break photogrammetry?
Yes, glossy, transparent, and dark featureless surfaces defeat point matching because the appearance changes with viewpoint. Use diffuse lighting, a polarising filter, or a temporary matting spray on non-precious objects.
Should I increase alignment accuracy or add more photos first?
Add and improve photos first; settings cannot rescue bad input. Once coverage and sharpness are good, raising alignment accuracy and the key/tie point limits gives the final improvement.