Skip to content

Compatibility with relighting and non-perspective camera models (e.g., orthographic) #31

@HarryPeverell

Description

@HarryPeverell

Hi
Thank you for the excellent work and for releasing such a clean and practical acceleration framework — it’s been really inspiring to read and experiment with.

I have two questions regarding compatibility/extension:

Relighting / appearance models (e.g., GS³)
Have you considered whether FastGS can be applied to relighting-focused Gaussian representations, such as GS³: Efficient Relighting with Triple Gaussian Splatting (code: [https://github.com/gsrelight/gs-relight])?
Concretely, do you expect FastGS’s acceleration strategy (e.g., densification/pruning/optimization schedule) to transfer to pipelines that augment Gaussians with more complex reflectance/lighting parameters, or are there design assumptions in FastGS that are tightly coupled to the “standard” 3DGS SH-based appearance?

Other camera models (e.g., orthographic projection)
Is FastGS currently tied to a perspective/pinhole camera model, or could it support other camera models such as orthographic projection with reasonable effort? If it is feasible, migrating the rendering backend to something like gsplat (which offers multiple camera model options) seems like an attractive route. I also noticed the related discussion in [https://github.com//issues/21] about a gsplat-based renderer — is this on your roadmap, or would you recommend a particular integration approach?

Any pointers on the expected engineering effort (which modules would need changes, what assumptions might break, etc.) would be greatly appreciated. Thanks again for the great project!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions