photo-import-setup

name	photo-import-setup
description	Set up a new Metashape project from scratch — import photos, load GPS reference, configure sensors (fisheye, rolling shutter, multi-camera), import EXR alpha masks, run image quality analysis, and disable bad frames. The first skill to use on any new capture. Works through the Metashape MCP server.
user-invocable	false

Photo Import & Project Setup

Overview

Every photogrammetry project starts with the same sequence: create project, import photos, load GPS, configure sensors, import masks, check image quality. This skill covers that entire setup phase, getting a project ready for alignment.

When to Use

• Starting a new project from raw photos
• Adding a new camera/capture to an existing project
• Re-importing after changing folder structure or GPS data
• Setting up sensors for the first time (fisheye, rolling shutter, etc.)

Prerequisites

• Photos on disk (EXR, JPEG, TIFF, PNG, DNG)
• GPS data (EXIF embedded or external CSV)
• Metashape MCP server running

Setup Pipeline

Step 1: Create or open project

create_project(path="E:\\Projects\\MyProject.psx")

Or for existing:

open_project(path="E:\\Projects\\MyProject.psx")

Step 2: Import photos

For a single folder:

add_photos(paths=["E:\\Captures\\Day1\\"])

For multiple folders (multi-day, multi-camera):

add_photos(paths=["E:\\Captures\\Day1\\", "E:\\Captures\\Day2\\"])

For glob patterns:

add_photos(paths=["E:\\Captures\\**\\*.exr"])

After import, verify:

list_chunks()

Check camera count matches expected photo count.

Step 3: Import GPS reference

If GPS is NOT in EXIF (common for vehicle-mounted captures):

import_reference(
    path="E:\\Captures\\gps_data.csv",
    columns="nxyz",        # n=label, x=longitude, y=latitude, z=altitude
    delimiter=","
)

Column format options:

• "nxyz" — label, lon, lat, alt (most common)
• "nxy" — label, lon, lat (no altitude)
• "nxyzabc" — label, lon, lat, alt, accuracy_x, accuracy_y, accuracy_z

After import, verify GPS loaded:

get_alignment_stats()

Check cameras_with_reference matches expected count.

Step 4: Configure sensors

This is the most critical setup step. Wrong sensor configuration is the #1 cause of total alignment failure.

Check current sensors:

list_sensors()

For fisheye lenses (vehicle-mounted, GoPro, 360 cameras):

set_sensor(
    sensor_index=0,
    sensor_type="fisheye",
    pixel_width=0.00488,      # Check camera spec sheet
    pixel_height=0.00488,
    focal_length=8.0           # Nominal focal length in mm
)

CRITICAL: If you have a fisheye lens and don't set sensor_type="fisheye", alignment WILL fail. This is the single most common setup error. The alignment-doctor agent exists largely because of this mistake.

For rolling shutter cameras (most consumer cameras, phones, GoPros): Rolling shutter compensation is set during match_photos, not sensor config. But knowing your camera uses electronic (rolling) shutter vs mechanical (global) shutter matters for choosing settings.

Multi-camera rigs: If you have multiple camera bodies, each needs its own sensor:

# Check if Metashape auto-detected separate sensors
list_sensors()
# If all cameras share one sensor but shouldn't:
# Separate cameras by body/lens in the Metashape GUI, then configure each sensor

Step 5: Import masks (for EXR with alpha)

For EXR files where the alpha channel masks the sky:

import_masks(
    method="alpha",
    path="{filename}"
)

The {filename} template tells Metashape to use each camera's own file as the mask source, reading the alpha channel.

Other mask sources:

Method	Use Case
"alpha"	EXR/PNG with alpha channel (sky mask)
"file"	Separate mask files (8-bit images, black=masked)
"model"	Generate from 3D model (requires existing model)
"background"	Auto-detect static background

After mask import, verify:

get_camera_metadata(label="your_first_camera_label")

Check that mask information is present.

Step 6: Analyze image quality

analyze_images()

This estimates sharpness quality for every camera. Review results:

select_cameras(max_quality=0.5)

Quality thresholds:

• >0.7: Good quality
• 0.5-0.7: Acceptable, may cause issues in sparse areas
• <0.5: Poor — likely blurry from motion, defocus, or obstruction

Disable low-quality cameras:

enable_cameras(labels=[...poor_quality_labels...], enable=False)

Or disable all below threshold in one step:

remove_cameras(quality_threshold=0.5)

WARNING: remove_cameras DELETES them. Use enable_cameras(enable=False) to keep them disabled but available for re-enabling later.

Step 7: Set reference accuracy

Tell Metashape how accurate your GPS is:

set_reference_settings(
    camera_accuracy_xy=3.0,     # Consumer GPS: 3-5m
    camera_accuracy_z=5.0       # Vertical always worse
)

GPS accuracy by type:

GPS Type	XY Accuracy	Z Accuracy
Consumer EXIF	5.0 m	10.0 m
Phone GPS	3.0 m	5.0 m
L1 GPS logger	1.0 m	2.0 m
RTK/PPK	0.02 m	0.04 m

Step 8: Set GPU config for alignment

set_gpu_config(cpu_enable=True)

CPU is ON for alignment (match_photos + align_cameras). It will be turned OFF before dense reconstruction.

Setup Complete — Ready for Alignment

At this point your project is ready. Next steps:

• For corridor captures: use the corridor-alignment-pipeline skill
• For aerial/close-range: proceed with match_photos + align_cameras directly
• If alignment fails: invoke the alignment-doctor agent

Decision Tree

Photos imported?
├── NO → Step 2: add_photos
├── YES → GPS loaded?
    ├── NO → Step 3: import_reference
    ├── YES → Sensors configured?
        ├── NO → Step 4: set_sensor (CHECK FISHEYE!)
        ├── YES → Masks needed?
            ├── YES (EXR) → Step 5: import_masks
            ├── NO → Quality checked?
                ├── NO → Step 6: analyze_images
                ├── YES → Ready for alignment

Common Setup Mistakes

Mistake	Symptom	Fix
Fisheye not set as fisheye	Total alignment failure (<20%)	set_sensor(sensor_type="fisheye")
GPS CSV wrong column order	Cameras placed in wrong hemisphere	Check columns parameter matches your CSV
GPS in wrong CRS	Cameras scattered randomly on map	Set chunk CRS to match GPS datum before import
Masks not imported	Sky reconstructed as tunnel	import_masks(method="alpha", path="{filename}")
Rolling shutter not enabled	Wavy/distorted alignment	Enable in match_photos settings
All cameras enabled despite blur	Poor alignment quality	analyze_images() then disable <0.5 quality

Road Corridor Specifics

For Decker Canyon-style vehicle-mounted fisheye captures:

• Camera: full-frame with fisheye lens → set_sensor(sensor_type="fisheye")
• Format: EXR with alpha → import_masks(method="alpha", path="{filename}")
• GPS: CSV from logger → import_reference(path="gps.csv", columns="nxyz")
• Quality: motion blur on turns → analyze_images(), disable <0.5
• Batch size: ~200 cameras per alignment batch
• Reference accuracy: consumer GPS → camera_accuracy_xy=3.0