Skip to content

Chunks and merging

Working with multiple chunks is one of the most architecturally nuanced parts of a Metashape project. The mechanics of Align Chunks and Merge Chunks are well-documented at the GUI level in the official manual, but the implications — what gets preserved, what gets lost, and what optimisation can and cannot do across chunk boundaries — are scattered across forum threads spanning a decade.

The articles in this section consolidate the architectural rules:

  • What mergeChunks actually does — Why merging produces orientation, not connection. The merge_markers and merge_tiepoints flags are the only mechanisms that actually link sub-chunks for downstream optimisation.
  • The three alignChunks methods — Point-based, marker-based, and camera-based: when each fits the problem, what their failure modes look like (mostly silent), and how to spot a failed alignment before merging.
  • The merge_tiepoints option — Why the GUI checkbox is greyed out by default, what keep_keypoints=True does, and when retaining cross-chunk tie points is worth the processing cost.
  • mergeChunks does not deduplicate cameras — The architectural reason post-merge deduplication is destructive, a deduplication script (with caveats), and three cleaner architectural strategies.

Articles

Quick decision table

Goal Method
Place sub-chunks in their relative positions only mergeChunks defaults — no link options
Have Optimize Cameras refine cross-chunk geometry merge_markers=True (when GCPs/markers shared) or merge_tiepoints=True (when keypoints stored)
Re-introduce tie points across chunks at merge merge_tiepoints=True + keep_keypoints=True at align time
Avoid duplicate cameras Don't share images across source chunks; or use marker-based linkage
Dedupe cameras after a camera-based merge Use the 1.0-era deduplication script — but only after building all dependent dense products