Incorrect tangential distortion parameter mapping in Metashape import

[f-dy]

Describe the bug The Metashape import functionality in nerfstudio incorrectly maps tangential distortion parameters p1 and p2, leading to incorrect camera distortion when importing Metashape projects.

Root Cause

Metashape and OpenCV use different conventions for tangential distortion parameters:

OpenCV convention (used by nerfstudio):

x_distorted = x + [2*p1*x*y + p2*(r² + 2*x²)]
y_distorted = y + [2*p2*x*y + p1*(r² + 2*y²)]

Metashape convention:

x_distorted = x + [2*p2*x*y + p1*(r² + 2*x²)]
y_distorted = y + [2*p1*x*y + p2*(r² + 2*y²)]

The parameters are swapped between the two systems:

• Metashape p1 = OpenCV p2
• Metashape p2 = OpenCV p1

Current Bug

In nerfstudio/process_data/metashape_utils.py lines 113-114:

s["p1"] = findparam(calib, "p1")  # Should be p2
s["p2"] = findparam(calib, "p2")  # Should be p1

This directly copies the parameters without accounting for the convention difference.

Fix

The parameters should be swapped during import:

s["p1"] = findparam(calib, "p2")  # Use Metashape's p2 for OpenCV's p1
s["p2"] = findparam(calib, "p1")  # Use Metashape's p1 for OpenCV's p2

Reference

This issue was identified and discussed in the EyefulTower project: https://github.com/facebookresearch/EyefulTower/issues/7#issuecomment-2662520442

@couperle

Impact

This bug affects the accuracy of camera calibration for any nerfstudio project imported from Metashape that uses tangential distortion correction, potentially leading to reduced reconstruction quality.