What's new

A reader-facing history of changes to the manual, newest first. The list of entries is generated from the manual's revision history (scripts/gen_changelog.py); the one-line description after each entry is written by hand and preserved when the list is regenerated.

August 2026

New articles

• Choosing camera axes: aerial vs terrestrial (and YPR vs OPK) — Why terrestrial capture should use OPK reference angles today: the YPR gimbal-lock problem, the Sensor.axes fix, and the 2.3.1 .psz bug that keeps OPK the safe choice.

Updated — 15 articles

• chunk.transform.matrix is local→world; camera.transform is local — Added a link to Choosing camera axes.
• AprilTag detection — choosing a variant — Trimmed the Context, moved the variant definitions beside the table, wrapped the confidence callout, and signposted the occlusion-robustness explanation.
• Bundle-adjustment quality: variance factor, overfit testing, and reference detectability — Added the per-observation a-priori σ column and a measured variance-factor benchmark, and clarified that the tie-point σ is per-projection (tiepoint_accuracy × keypoint size).
• Camera reference error: computing per-camera location and orientation residuals in Python — Added a link to Choosing camera axes.
• Coded circular targets: printing, sizing, and choosing a variant — Merged the family-selection tables into one decision section, moved the marker-pool and binary-code tables into a reference appendix, shortened the title, and signposted the explanatory sections.
• Drone metadata: DJI altitude semantics and RTK XMP accuracy tags — Added a link to Choosing camera axes.
• Fixing exterior orientation: skipping Align Photos with known EO — Added a link to Choosing camera axes.
• Gray flags in marker detection: what they mean and how to remove them — Updated the coded circular targets cross-link label.
• Helping alignment when photos don't align: markers, references, and what to use when — Updated the coded circular targets cross-link label.
• Importing camera orientation: EXIF, omega-phi-kappa, and yaw/pitch/roll — Added a link to Choosing camera axes.
• Keypoint-size-normalised reprojection error: the kps metric — Verified the per-projection tie-point weighting (σ = tiepoint_accuracy × proj.size) empirically, promoted the article to verified, and repaired its front matter.
• Programmatic marker placement and pinning — Updated the coded circular targets cross-link label.
• Saving estimated reference values to file: location, rotation, error, and sigma — Added a link to Choosing camera axes.
• Tightening reference accuracies after alignCameras: when the similarity-transform residual isn't enough — Added a link to Choosing camera axes.
• YPR rotation conventions: ypr2mat vs camera.reference.rotation — Added an inline link to Choosing camera axes at the OPK caveat.

July 2026

Updated

• Orthomosaic export — the 4GB / BigTIFF limit and shift-during-export — Added a hedged caveat on disabled cameras and seamline regeneration during export.
• Reference preselection — Documented the measured sequential-preselection window and added a verification script.

June 2026

New articles — 125 articles

• alignChunks(method=point) cannot break geometric symmetry
• camera.project and camera.unproject: 2D ↔ 3D in Python
• chunk.transform.matrix is local→world; camera.transform is local
• exportPointCloud defaults: GUI vs Python CRS difference
• filter_mask=True starves matchPhotos when masks cover most of the image
• mask_tiepoints cross-view propagation and the foreground-occluder case
• mergeChunks does not deduplicate cameras
• Model.renderDepth: synthetic depth rendering from arbitrary viewpoints
• sensor.calibration vs sensor.user_calib — initial vs adjusted values
• Adaptive camera model fitting
• Adding cameras to an aligned chunk: the keep_keypoints workflow
• Agisoft knowledge base
• AI-assisted mask generation
• Aligning two meshes / point clouds: model-to-model registration in Python
• Applying Patch on multiple shapes (orthomosaic patching by script)
• AprilTag detection — choosing a variant
• Area and volume measurement: Model view vs DEM, Shapes vs cropping
• Auto-export per-shape: orthomosaic, DEM, point cloud, mesh, KMZ
• Automatic sky masking with ONNX: pre-clean for sky-prone aerial captures
• Automating gradual selection in Python
• Bundle-adjustment quality: variance factor, overfit testing, and reference detectability
• Calibration groups: programmatic management in Python
• Camera reference error: computing per-camera location and orientation residuals in Python
• Camera stations: when to use them and the nodal-head requirement
• Change Path: swapping image format / resolution after alignment
• Choosing the master sensor in a multi-camera layout
• Choosing which camera parameters to fix or optimize
• Coded circular targets: printing, sizing, and choosing a variant
• Color calibration: when to use it, what it does, and the white-balance / vignetting knobs
• Comparing chunks for change detection: DEM, mesh, and point-cloud diff workflows
• Components view
• Computing camera direction vectors and look-at points in Python
• Computing per-camera coverage area (image footprint on the model)
• Contributing
• Converting camera.transform to ENU (or any local Cartesian)
• Creating point shapes programmatically: grid placement with DEM-based elevation
• Custom vertical datums: adding a geoid undulation grid
• Declaring a fixed-geometry multi-camera rig in Python
• DEM build options: point cloud vs mesh as source, and the interpolation knob
• Depth-map quality and filter modes: choosing parameters for Build Point Cloud
• Diagnosing CUDA / OpenCL errors: timeouts, OOM, kernel failures
• Diagnosing under-aligned chunks
• Diagnostic mesh visualisation: colorize by overlap or altitude
• Drone metadata: DJI altitude semantics and RTK XMP accuracy tags
• DSM ridge-line artefacts: alignment-quality diagnosis, not DEM-pipeline bugs
• Estimate image quality
• Exclude stationary tie points: when and why
• EXIF focal length: which tags Metashape reads, and the no-crop-factor rule
• Exporting cameras for Gaussian Splatting / Colmap downstream pipelines
• Exporting depth maps from Python (32-bit, scaled)
• Exporting to Cesium 3D Tiles: point cloud vs tiled model
• Filtering cameras by tie-point selection: which photos see this 3D point?
• Fisheye and spherical sensors: the seven projection types and when each applies
• Fixing exterior orientation: skipping Align Photos with known EO
• Generic preselection
• Glossary
• Gradual selection
• Gray flags in marker detection: what they mean and how to remove them
• Ground classification: the erosion radius parameter
• Guided matching
• Helping alignment when photos don't align: markers, references, and what to use when
• How Metashape computes vertex normals on model export
• Importing camera orientation: EXIF, omega-phi-kappa, and yaw/pitch/roll
• Intel i9-13900K / 14900K instability: BIOS workarounds for Metashape crashes
• Keypoint-size-normalised reprojection error: the kps metric
• Linux vs Windows: 10–40% faster processing
• Logging from Metashape Python scripts (and why app.settings.log_* does not work for headless scripts)
• Mapping orthomosaic pixels back to source images
• Marker projection statistics: counts, per-marker errors, and metre-vs-pixel framing
• Mesh and point-cloud editing recipes (Python)
• Mesh surface types: Arbitrary vs Height field, and the source-data choice
• Metashape's distortion model and converting to OpenCV / Colmap
• Multi-GPU setups: SLI, TCC mode, and utilization
• Multispectral imaging: per-file band declaration and master-band change
• Network processing: setup, monitoring, and common failure modes
• Orthomosaic export — the 4GB / BigTIFF limit and shift-during-export
• Orthomosaic in a marker-defined planar projection
• Pair preselection: choosing between Disabled, Generic, and Reference
• Performance tuning: CPU, RAM, GPU, and OS
• Point cloud confidence values: what they mean and how to filter
• Programmatic calibration import / export
• Programmatic chunk.region control: move, scale, rotate the bounding box
• Programmatic marker placement and pinning
• RAM and quality settings: what determines peak memory
• Recovery paths for unaligned cameras
• Reducing camera overlap in over-acquired datasets
• Reference preselection
• Release gate demos
• Removing "blue flag" marker projections cleanly
• Rendering models from synthetic cameras: spherical panoramas and regular views
• Repositioning a chunk: moving the origin to a known point
• Reproducing chunk-info statistics in Python
• Reprojection error analysis: per-camera and per-tie-point
• Rolling-shutter compensation: Regularized vs Full, when to use which
• Sample data
• Saving estimated reference values to file: location, rotation, error, and sigma
• Scalebar distance error: per-scalebar values and RMS aggregation
• Scripting context pitfalls: GUI vs command-line, document handles, and stage validation
• Sensor and camera shared-tie-point graphs: detecting isolated groups
• Setting chunk.region to bound the tie-point cloud
• Style
• Symlink filename — not target — controls the camera label
• Synthetic position priors via ReferencePreselectionSource
• Texture blending modes — what each one actually does
• Texture in Metashape 2.3 — what changed and why your scripts may need updating
• The merge_tiepoints option (and the keep_keypoints prerequisite)
• The chunk's internal coordinate system: arbitrary scale and the chunk.transform.scale factor
• The Clean Tie Points → Optimize Cameras loop
• The slave-sensor transform: composition rule, axis convention, and recipes
• The three alignChunks methods (point / marker / camera)
• Tie-point multiplicity: track length, distribution, and what it tells you
• Tightening reference accuracies after alignCameras: when the similarity-transform residual isn't enough
• Tiled models: when to use, what they replace, and how to export
• Transferring camera orientation between modalities (RGB → thermal)
• Troubleshooting Metashape: diagnostic ladder and where the logs live
• Undocumented tweaks: a partial reference
• Unverified
• Version differences
• Version timeline
• What mergeChunks actually does (and what it does not)
• When does optimizeCameras actually do something?
• When does Metashape use the GPU? (and how to verify)
• When to use chunks: dataset partitioning strategies
• Working with shape geometries (1.7+ API)
• YPR rotation conventions: ypr2mat vs camera.reference.rotation