Bootstrap with injected depths from Depthmaps

[servantftransperfect]

This is a first bootstrapping solution for depthmaps. It is very basic at the moment and should be improved over time !

This pull request introduces a new "depth" bootstrapping method to the Structure-from-Motion (SfM) pipeline, alongside the existing "classic" and "mesh" methods. It adds the necessary infrastructure to handle depth-based bootstrapping, including new C++ logic, Python interface updates, and supporting utilities for constructing SfM data from depth maps. The changes are grouped below by theme.

New bootstrapping method: "depth"

• Added a new parameter method (choices: "classic", "mesh", "depth") to the SfMBootStrapping node in the Python interface, allowing users to select the bootstrapping approach. The mesh file parameter is now only enabled if the "mesh" method is selected. [1] [2]
• Introduced an enum EBOOTSTRAPMETHOD and a string-to-enum conversion utility to handle the new method in the C++ pipeline entry point.

Depth-based bootstrapping implementation

• Added the bootstrapDepth function, which constructs initial camera poses and landmarks using depth information from tracks, and integrates it into the bootstrapping pipeline. [1] [2]
• Implemented buildSfmDataFromDepthMap in new files TracksDepths.cpp and TracksDepths.hpp, providing utilities to build minimal SfMData objects from depth maps for a given view. [1] [2]
• Updated CMake and includes to register and use the new depth-based files and logic. [1] [2] [3] [4]

Pair selection using depth information

• Added findBestPairFromTrackDepths to select the best initial image pair using depth information, with supporting documentation and logic. [1] [2]

Supporting API improvements

• Added a const overload for getViewSharedPtr in SfMData to facilitate safe access during depth-based operations.

[Copilot]

Pull Request Overview

Copilot reviewed 10 out of 10 changed files in this pull request and generated 2 comments.

---

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

[Copilot]

Pull Request Overview

Copilot reviewed 10 out of 10 changed files in this pull request and generated 4 comments.

---

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

[Copilot]

[nitpick] The parameter description 'Method choice.' is too vague. Consider providing more descriptive text like 'Bootstrapping method: classic (epipolar geometry), mesh (3D mesh constraints), or depth (depth map information)'.

Suggested change

	("method", po::value<std::string>(&methodString)->required(), "Method choice.")
	("method", po::value<std::string>(&methodString)->required(), "Bootstrapping method: classic (epipolar geometry), mesh (3D mesh constraints), or depth (depth map information).")

[Copilot]

The error message 'No known pose for mesh' is unclear. It should explain that mesh bootstrapping requires empty view filters or provide guidance on the expected configuration.

Suggested change

	ALICEVISION_LOG_ERROR("No known pose for mesh");
	ALICEVISION_LOG_ERROR("Mesh bootstrapping requires empty view filters. Please ensure that no view filters are set when using mesh bootstrapping.");

[Copilot]

Setting resection max error to infinity removes error constraints entirely. Consider using a reasonable upper bound or making this configurable to maintain stability and prevent degenerate solutions.

Suggested change

	resection.setResectionMaxError(std::numeric_limits<double>::infinity());
	// Use a reasonable finite reprojection error threshold (in pixels) instead of infinity.
	// Consider making this value configurable if needed.
	resection.setResectionMaxError(4.0);

[Copilot]

Using infinity for max error and 50000 iterations are magic numbers that should be defined as named constants or made configurable parameters for better maintainability.