landmarks can now be defined with respect to a camera geometric frame instead of the world geometric frame

[servantftransperfect]

A new parameter to landmarks is added "referenceViewId".

For a landmark, if this parameter is not UndefinedIndexT, this is considered as the reference view index for this landmark.

In this case, the landmark coordinates will be considered to be relative to the reference view geometric frame.

[fabiencastan]

This needs a unit test to validate the behavior on simple generated data.

[Copilot]

Pull Request Overview

This PR adds support for landmarks to be defined relative to a camera's geometric frame instead of the world frame. The new referenceViewIndex parameter enables landmarks to store their coordinates relative to a specific camera view, which requires coordinate transformations during bundle adjustment.

• New referenceViewIndex field added to landmarks for specifying reference camera frame
• Updated serialization/deserialization to handle the new reference view parameter
• Added projection cost function for relative landmarks with coordinate transformation logic

Reviewed Changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
Landmark.hpp	Adds referenceViewIndex field and updates equality operator
jsonIO.cpp	Adds JSON serialization support for referenceViewIndex
AlembicImporter.cpp	Imports referenceViewIndex from Alembic format
AlembicExporter.cpp	Exports referenceViewIndex to Alembic format
projection.hpp	New ProjectionRelativeErrorFunctor for relative landmark projections
BundleAdjustmentCeres.cpp	Integrates relative projection cost function into bundle adjustment
bundleAdjustment_test.cpp	Adds test case for relative landmark bundle adjustment
NViewDataSet.hpp	Adds useRelative configuration flag for test datasets

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[Copilot]

The warning message incorrectly states 'Describer type will be ignored' when the issue is with reference view indices. It should say 'Reference view indices will be ignored' to accurately reflect what is being discarded.

Suggested change

	ALICEVISION_LOG_WARNING("[Alembic Importer] Describer type will be ignored. describerType vector size: "
	ALICEVISION_LOG_WARNING("[Alembic Importer] Reference view indices will be ignored. referenceViewIndices vector size: "

[Copilot]

The coordinate transformation logic is incorrect. The proper transformation from camera frame to world frame should first apply the inverse rotation, then subtract the translation. Currently, translation is subtracted before rotation, which will produce incorrect world coordinates.

Suggested change

	relpoint[0] = relpoint[0] - refcam_t[0];
	relpoint[1] = relpoint[1] - refcam_t[1];
	relpoint[2] = relpoint[2] - refcam_t[2];
	T invrefcam_R[3];
	invrefcam_R[0] = -refcam_R[0];
	invrefcam_R[1] = -refcam_R[1];
	invrefcam_R[2] = -refcam_R[2];
	T absolutePoint[3];
	ceres::AngleAxisRotatePoint(invrefcam_R, relpoint, absolutePoint);
	T invrefcam_R[3];
	invrefcam_R[0] = -refcam_R[0];
	invrefcam_R[1] = -refcam_R[1];
	invrefcam_R[2] = -refcam_R[2];
	T rotatedPoint[3];
	ceres::AngleAxisRotatePoint(invrefcam_R, relpoint, rotatedPoint);
	T absolutePoint[3];
	absolutePoint[0] = rotatedPoint[0] - refcam_t[0];
	absolutePoint[1] = rotatedPoint[1] - refcam_t[1];
	absolutePoint[2] = rotatedPoint[2] - refcam_t[2];

[Copilot]

The conditional logic creates duplicate residual blocks - one for relative landmarks and another for absolute landmarks. This should be an if-else structure to avoid adding two residual blocks for the same observation when referencePoseBlockPtr is not null.