Paul-Edouard Sarlin
Paul-Edouard Sarlin
No luck so far.
Will fix this issue in a later PR, along with: - add stubs - switch to GHA VCPKG caching
Thanks for providing a self-contained reproduction script. Does increasing the number of iterations help? Alternatively, the numerical Jacobians are maybe not sufficiently accurate for heavy distortions. https://github.com/colmap/colmap/blob/d70edd784808d29c855fff0b4ffc2bdc3ea290f1/src/colmap/sensor/models.h#L657-L659
The documentation only reflects the state of HEAD. This option was introduced recently, in 3.11. Ideally we should find a way to generate version-specific docs. Using the GPU also requires...
Thank you for reporting this. This should be fixed by https://github.com/colmap/colmap/pull/2704, which got merged after 3.10 was released. Could you try the code at HEAD and confirm that the issue...
Please also update the instructions at https://github.com/colmap/colmap/blob/main/doc/pycolmap/index.rst
We have a working prototype of SfM aided by monocular/sensor depth. We might release it in the coming months. Could you point us to a dataset (with ground truth poses...
You can check out https://github.com/cvg/mpsfm We've only used it with monocular depth but in principle sensor depth work fine too.
Either use images from a scene with more texture, or calibrate your image manually with https://github.com/ethz-asl/kalibr for example.
`classh` was introduced in pybind11 v3.0.0, so you are likely using an older version. The pyproject.toml is the source of truth for the version of pybind11 required for a given...