The current application has a significant drawback as it can only handle forward-facing scenes, similar to the NeRF-method, and may fail with camera motions having rotation perturbations exceeding ±20◦. As a consequence, spherical datasets cannot be processed by the system, and this limitation may adversely affect non-linear passes.
CanGaussian Splatting be used with spherical datasets?
According to Chang et al.
CanGaussian Splatting be used with spherical datasets?