Applying to Cubemap images

[rykawamura]

Thank you for the excellent work. Is it possible to apply PyCuVSALM to cubemap images extracted from 360 camera videos? I am considering using PyCuVSLAM to obtain pointclouds for 3D Gaussian splatting initialization.

[zwdoescode]

@rykawamura cuVSLAM is designed for various camera setups (mono, stereo, multi-camera) but implicitly assumes a standard, non-omnidirectional camera model. The input is expected to be traditional 2D images from cameras with a standard field of view (FOV).

If you want to use cubemap images as inputs, you probable need to do a lot of modifications in order to use it with PyCuVSLAM. Here is just some brainstorming:

1. Choose one or more faces of the cubemap that provide a relevant, non-featureless view for each frame. This would require custom preprocessing to select or combine faces effectively for PyCuVSLAM's input.
2. A good camera calibration
3. No significant distortion and motion blur in the provided images