Agisoft Metashape for Heritage Photogrammetry: Complete Workflow Guide

Why Metashape for Heritage Work

Metashape is not the fastest SfM platform (RealityCapture is significantly faster), and it is not the cheapest (COLMAP is free and open-source). But it is the most practical for professional heritage documentation in India for three reasons: the perpetual licence model fits project-based billing without subscription commitment; its GCP workflow, accuracy reporting and coordinate system handling are production-quality; and its chunk-based multi-model workflow handles the common Indian heritage scenario of a large temple complex requiring separately processed sub-projects merged into a single georeferenced model.

Before You Open Metashape

The quality ceiling of a Metashape project is set in the field, not in the office. Processing cannot recover images with motion blur, insufficient overlap, or surfaces that do not feature-match (uniform colour, transparent glass, highly reflective polished stone).

Before loading images, verify:

• Image sharpness — open a selection of images at 100% zoom in any photo viewer; any motion blur or focus error in more than 10% of images is a problem; reshoot if required.
• Overlap — estimate: for drone nadir surveys, 80% front/70% side; for facade photogrammetry, 70% overlap minimum; carved surface close-range work needs 80%+ overlap to resolve depth in recesses.
• GCP targets — confirm that the GCP targets are visible and unambiguous in at least 5 images per target; remove any target that is partially obscured or blurred.
• File naming — Metashape sorts images by filename for display; consistent naming by camera station makes reviewing alignment results much faster.
• Hardware — dense cloud generation on High or Highest quality requires significant GPU VRAM; for a 2,000-image project on High quality, budget 16GB VRAM minimum; use GPU processing for dense cloud (not alignment); Metashape handles large projects better with NVMe SSD project file location.

Image Alignment

After alignment, review the alignment report: check that the reprojection error is below 1.0 px (ideally 0.5–0.7 px) and that the percentage of aligned cameras is above 95%. Cameras that failed to align — shown greyed out — usually indicate blurred images, images with no overlap with the rest, or images taken of featureless surfaces. Disable failed cameras before proceeding to dense cloud.

For multi-chunk projects, align all chunks separately, then use the Merge Chunks function before GCP import — or import GCPs per chunk and align chunks using the GCPs as tie points across chunks.

GCP Import and Optimisation

GCP import and optimisation is the most consequential step in a georeferenced heritage photogrammetry project. Errors at this stage propagate into all output products.

1. 1Import coordinates — Tools > Reference > Import Reference; specify the coordinate system (WGS84 for GPS-measured GCPs; local site grid for total-station-measured GCPs). Ensure the CRS matches exactly how your GCPs were measured.
2. 2Mark GCPs in images — open the Reference panel; for each GCP, double-click to open the filter view; in each image where the target is visible, right-click and place the marker at the exact centre of the target cross. Minimum 5 images per GCP; 8–12 is better for stable optimisation.
3. 3Run optimisation — after marking all GCPs, click Optimise Cameras; the optimisation adjusts camera positions and orientations to minimise the difference between measured GCP positions and their projected positions in the model.
4. 4Review residuals — in the Reference panel, examine the GCP residuals (XY and Z errors); all GCPs should show residuals below your target accuracy (for 1:50 drawing, residuals should be < 20mm; for 1:20, < 10mm); if a GCP shows high residual (>2× the average), review its marking in images — likely marking error.
5. 5Designate check points — convert 2–3 GCPs to check points (not used in optimisation) to provide an independent accuracy check; their residuals after optimisation represent the true accuracy of the model at uncontrolled points.
6. 6Re-optimise if needed — if residuals are high, review the worst-residual GCPs first; re-mark in images; re-optimise; do not increase accuracy by increasing the GCP marker accuracy parameter — this only suppresses the reported error, not the actual error.

Dense Cloud Generation

Depth filtering — for heritage carved surfaces, set depth filtering to Disabled or Mild. The Aggressive and Moderate settings remove points that are outliers in local depth — which is exactly what the edges of carved recesses look like to the algorithm. Aggressive depth filtering smooths out carved relief geometry. For flat or simple surfaces (orthophoto surveys of flat roofs, walls), Moderate depth filtering is acceptable.

Point cloud classification — after dense cloud generation, run Tools > Dense Cloud > Classify Ground Points if you need to separate ground from structure for DEM generation. For facade elevation work, classification is not needed.

Mesh and Texture

For most heritage documentation purposes, the dense point cloud is the primary deliverable, and mesh generation is optional. Generate a mesh when: the client needs a 3D model for visualisation or VR; you need a watertight surface for 3D printing or CNC reproduction; or the output requires a textured model.

Outputs: Orthophoto, DEM and Point Cloud

Reading the Accuracy Report

Metashape's generated PDF report is the quality certificate for a heritage photogrammetry project. Every professional deliverable should be accompanied by a report. Know what to look for:

• Camera alignment: percentage of aligned cameras should exceed 95%; reprojection error should be below 1.0 px (ideally 0.5–0.7 px); values above 1.5 px indicate alignment quality problems.
• Ground control: GCP residuals (X, Y, Z errors) in the markers table; for 1:50 output, all residuals should be below 20mm; for 1:20, below 10mm; report the RMS (root mean square) error across all GCPs.
• Check points: the check point residuals are the true accuracy indicator — these are GCPs not used in optimisation; if check point error is 2× the GCP residuals, there is likely a systematic error (scale, rotation, CRS mismatch).
• Ground sampling distance: reported GSD should match your capture distance and camera specifications; if the reported GSD is significantly larger than expected, images were captured too far from the subject.
• Camera model: review the calibrated lens parameters (f, cx, cy, k1–k4); if radial distortion coefficients are very high (k1 > 0.3), the camera has significant lens distortion; check that the camera model was calibrated in-project (auto-calibration) or pre-calibrated.

Settings Quick Reference by Project Type

Common Processing Mistakes

• Using Aggressive depth filtering on carved surfaces — smooths out the carved relief the project was designed to capture; always use Disabled or Mild for any carved stone or timber surface.
• Not reviewing alignment quality before dense cloud — running dense cloud on a poorly aligned model wastes hours of processing time; always check reprojection error and camera alignment percentage first.
• Marking GCPs only in the minimum required number of images — 5 images per GCP is the minimum; 8–12 produces significantly more stable optimisation; marking in more images than necessary has no downside.
• Importing GCPs in the wrong coordinate system — the most common cause of georeferencing errors; always confirm the CRS against the GCP measurement method before importing.
• Not generating a check point — without check points, there is no independent accuracy check; always designate 2–3 GCPs as check points before optimisation.
• Processing drone images and close-range images in the same chunk without careful camera group management — Metashape may struggle to align cameras with very different focal lengths and GSD in a single chunk; use separate chunks, optimise separately, then merge.

Professional Practice

In professional practice, Metashape processing time is rarely the bottleneck — fieldwork, GCP marking and quality review are. A disciplined professional workflow treats GCP marking as requiring approximately 20–30 minutes per GCP (marking in 8–12 images, reviewing residuals, re-marking as needed) — for a 10-GCP project, that is 3–5 hours of careful work at the monitor.

Hardware investment matters significantly. The difference between a project that processes overnight on a workstation with a current-generation GPU and one that takes four days on a laptop without GPU processing is the difference between professional turnaround and an unbillable bottleneck. A workstation with 32GB RAM, NVMe SSD and a GPU with 12GB+ VRAM (RTX 3060 or better) is the minimum practical specification for professional heritage photogrammetry work. The approximate India cost for a suitable workstation build (2026) is ₹1.5–₹3 lakhs.

File management is a professional discipline that processing tutorials rarely address. A completed Metashape project for a major heritage site may include: the .psx project file, 50–200GB of raw images, multiple dense clouds at different quality settings, multiple chunk exports and a processing log. Establish a consistent folder structure at project inception; archive the final project with all source images in a format that can be re-opened without the original operator's knowledge. E57 export for the point cloud, GeoTIFF for orthophotos, and PDF accuracy report are the archival deliverables; the .psx file and raw images should be retained for the project archive period.