Question on StreetGS release plan and independent reproduction

[jcy-debug]

Hi, thank you very much for the great work on FastGS.

In the paper, FastGS is presented as an efficient and extensible framework for 3D Gaussian Splatting, and dynamic scenes are demonstrated using Deformable 3DGS as an example.

I also noticed that in the repository roadmap, street_gaussians (autonomous driving scenes) are listed under the upcoming multi-task expansion. At the moment, however, there does not seem to be a placeholder or runnable implementation available.

I was wondering whether there is an expected timeline for releasing street-scale / autonomous-driving related code. Alternatively, if I would like to reproduce such a method in the short term, would it be reasonable to implement it independently based on the existing FastGS backbone and the current Deformable 3DGS support?

From my understanding, this should be feasible in principle, but I would greatly appreciate your advice on whether there are any major pitfalls or missing components to be aware of.

Thank you very much for your time.

[Shiwei-Ren]

Hello! Thank you very much for your interest and recognition!!

1. Regarding the integration of street_gaussians with FastGS, we are still actively working on it, and we will release the corresponding mature and well-tested code as soon as it is ready.

2. Using Fast-D3DGS (Deformable 3DGS) directly should also be applicable to autonomous driving scenarios. In fact, Table 1 of DGGT (https://arxiv.org/pdf/2512.03004) reports results related to DeformableGS. However, since we have not conducted experiments in this setting ourselves, I am not entirely certain about some implementation details. I will consult with the authors about how they approached this and will follow up with a reply in the issue.

Thank you for your understanding!