Training on D3DGS data

Hi, thanks for releasing this excellent work.

I tried to validate this code using [D3DGS](https://github.com/JonathonLuiten/Dynamic3DGaussians?tab=readme-ov-file) data, but it didn't work.

This is a dynamic dataset containing 30 or so cameras, where the cameras are in a semicircle inwards.

I am choosing to convert the D3DGS data to the format of the neu3d data (mp4 and poses_bounds.npy), and I assure you that I have performed the correct bitmap transformations, but the results still display all kinds of strange colors.

Do you have any suggestions?


Start by reproducing training with the converted D3DGS data in the neu3d format, using the generated mp4 and poses_bounds.npy files. Compare the bitmap transformations and resulting colors with the provided image; done means the converted dataset trains and renders without the reported color artifacts.

Written by the indexing model from the issue text.