Error about training with gs_points

[cv-lab-x]

thanks for your great work, some errors when train with gs_type "gs_points"

Traceback (most recent call last): File "train.py", line 249, in <module> op = optimizationParamTypeCallbacks[args.gs_type](parser) KeyError: 'gs_points'

Looking forward to your reply, thanks! @waczjoan

[waczjoan]

Thank you! :)

Let me explain a little more broadly. In the specific context you're interested in, we utilize the following procedure:

• Initially, we have access to the original Gaussian splicing.
• To simplify the conversion of Gaussians into triplet "pseudo" faces, we opt for implementing flat Gaussian splatting. This ensures that the resulting Gaussians are as flat as possible, with the first scale component set to epsilon.
• It's important to note that only after training the flat splatting process can we parameterize the resulting flat Gaussians using gs_points.

In the ReadMe file, the available options are outlined as follows:

Available options:

• gs_flat - Basic gaussian splicing which one scale value is epsilion, thus the resulting gaussians are flat. Model used to parameterize the gs_points GaMeS model.
• gs_points - GaMeS model - used to parameterize the flat Gaussian splatting model.

It's crucial to understand that gs_points cannot be utilized for training purposes, but primarily serves for parameterization during rendering, as detailed in the "Scenario II: we don't have mesh; or we don't want to use it." section of the ReadMe.

This implies that you encountered an error with a purpose—because you shouldn't employ the gs_points flag for train.py . I've attempted to highlight the usage of gs_points, but if you have any suggestions or ideas to clarify its usage for other users, please feel free to share them, I would be happy!