Generic preselection¶

Status: documented (2026-05-24).

What it is¶

Generic preselection is a feature-based pair-selection method used by Match Photos / Align Photos to identify candidate image pairs for matching. Unlike reference preselection, which uses camera coordinates, generic preselection uses low-resolution feature comparison: each image's features are matched against every other image's features (or against a shortlist if reference preselection is also enabled), and pairs above a similarity threshold are kept for the full matching step.

Where it lives¶

	Standard	Pro
GUI	Workflow → Align Photos → check Generic preselection	(same)
Python	`Metashape.Chunk.matchPhotos(generic_preselection=True/False, …)`	(Pro only)
Default	`True`	`True`

How it interacts with reference preselection¶

The two preselection methods are independent kwargs that can both be True, both be False, or either one alone. Their interaction is documented in Reference preselection:

Both True: reference preselection picks N nearest neighbours by coordinate; generic preselection runs feature similarity on that shortlist. Recommended default for aerial surveys with reliable GPS.
Generic only (reference_preselection=False, generic_preselection=True): O(N²) feature comparison across all images. Recommended when references are missing or unreliable.
Reference only (reference_preselection=True, reference_preselection_mode=…, generic_preselection=False): introduced in PhotoScan 1.3. Trusts the references to identify pairs without further feature filtering. Useful when references are very accurate (surveyed positions) and feature-based selection is unnecessary overhead.
Both False: matches every image against every other image without any pair filtering. Slowest; rarely the right choice.

When to disable¶

The canonical "clean" alignment recipe (per Diagnosing under-aligned chunks) keeps generic preselection enabled:

Accuracy: High
Generic preselection: enabled
Reference preselection: disabled
...

Disable generic preselection only when you have very accurate reference data and want to trust it exclusively (reference_preselection=True, reference_preselection_mode=ReferencePreselectionSource, generic_preselection=False). This is the "reference-without-generic" mode introduced in PhotoScan 1.3.

Caveats¶

Generic preselection is the default. Disabling it without enabling reference preselection produces the slowest possible alignment (full all-pairs matching).
The similarity threshold is not user-tunable via Python. The internal threshold is calibrated for typical photographic content; performance on synthetic / rendered / non-photographic imagery may be unpredictable.
filter_stationary_points (a separate kwarg) interacts with the matching, not with preselection. It removes tie points that are stationary across image pairs (typical of reflections, dust on the lens, static foreground); see the matchPhotos signature for details.

Reference preselection — the coordinate-based pair-selection partner.
Guided matching — a separate option applied at the match stage, not at pair selection.
Tie points

Articles in this manual¶

Diagnosing under-aligned chunks — recommends generic_preselection=True as part of the canonical "clean" recipe.
Synthetic priors via ReferencePreselectionSource — a deliberate generic_preselection=False case for repeated-feature scenes where generic preselection would produce false matches.

Forum threads worth reading¶

Date	Version	Author	Thread	One-line takeaway
2016-12-20	PhotoScan 1.2	Alexey Pasumansky	Align images - Reference preselection	Generic preselection runs on the reference-preselection shortlist when both are enabled (1.2 era; separable in 1.3+).
2021-02-11	Metashape 1.7	Alexey Pasumansky	Tie points ghosting	The canonical "clean" baseline keeps generic preselection enabled.