Bug Report: Issue with Point3D.bin Output from glomap

[mgaoling]

Thank you for your incredible software!

I am using glomap as a pre-processing step for some downstream tasks in colmap, as it is built upon colmap. However, I encountered an issue when attempting to run the following command with the output models from glomap:

colmap mapper --database_path xxx --image_path xxx --input_path xxx --output_path xxx

This command resulted in the following error message:

F0927 17:39:50.430783 20688 reconstruction.cc:2061] Check failed: point2D.HasPoint3D()
*** Check failure stack trace: ***
@ google::LogMessage::Fail()
@ google::LogMessage::SendToLog()
@ google::LogMessage::Flush()
@ google::LogMessageFatal::~LogMessageFatal()
@ colmap::Reconstruction::ResetTriObservations()
@ colmap::Reconstruction::DeletePoint3D()
@ colmap::Reconstruction::MergePoints3D()
@ _ZN6colmap23IncrementalTriangulator5MergeERKNSO_70ptionsEm.localalias
@ colmap::IncrementalTriangulator::MergeAllTracks()
@ colmap::IncrementalMapper::MergeTracks()
@ colmap::CompleteAndMergeTracks()
@ colmap::(anonymous namespace) ::IterativeGlobalRefinement)
@ colmap::IncrementalMapperController::Reconstruct)
@ colmap::IncrementalMapperController::Run()
@ colmap::Thread::RunFunc()
@ (unknown)
@ start_thread
@ clone
[1] 20687 abort (core dumped)

Upon investigating, I discovered that the results produced by glomap contain a small bug: for all 3D points triangulated by exactly two views, while these 3D points are included with pointers (to images) in the point3D.bin, they do not have valid back-pointers from those two images to the respective 3D points.

This inconsistency leads to the error message above. As a temporary workaround, I found that removing those invalid 3D points resolves the issue, allowing colmap to function correctly.

[lpanaf]

Hi @mgaoling, thank you very much for reporting the bug. Is it possible for you to open a pull request on this?