ply Output

[hecodeit]

Training completed successfully with python train.py --config config/3dgs/base_model.yaml. The generated .ply file in the point_cloud directory under outputs appears to be incorrect. However, outputs\...\test\ours_4000\renders looks normal. How can I output the correct .ply file?

test\ours_4000\renders:

ply on viewer:

[jianglh-WHU]

emm, could you please tell me how to upload the .ply file to the viewer and which viewer you have chosen?

[hecodeit]

@jianglh-WHU

Readme show:

...relevant SIBR visualizations forthcoming.

So I'm using Postshot as the viewer.

It's possible that the ply format output by Octree-AnyGS is not the standard Gaussian Splatting format. Can you point out the differences? Also, is there a way to output the standard Gaussian Splatting format?

[jianglh-WHU]

The Octree-AnyGS format is not compatible with Postshot; it is only supported by our viewer due to the MLP.

If you wish to convert it to the standard 3D-GS format, you will need to exclude the view input from each MLP and substitute the color MLP with the SH MLP. Upon completing the entire training process, you can then convert it to the explicit 3D-GS format.

[hecodeit]

I don't understand. The outputs after training is complete:

├── backup
├── cameras.json
├── cfg_args
├── config.yaml
├── events.out.tfevents.1728462264.AMA-E.223199.0
├── input.ply
├── outputs.log
├── per_view.json
├── point_cloud
├── results.json
└── test

So where is the MLP output?

[jianglh-WHU]

The MLP is in the point_cloud folder. However, I mean, you need to modify the input of the MLP in the code:

1. first exclude the view input from each MLP
2. replace the color MLP with the SH MLP to get color

[tongji-rkr]

We will update the ply conversion as soon as possible, you should choose the 3D-GS version.