Required steps for rendering a point cloud

[ur10]

Hi I am trying to render point clouds but I have some confusion. Here are the steps that I have followed.

• Used the standard Playroom dataset with the required directory structure for colmap generation.
• Ran python convert.py --camera PINHOLE -s tandt_db/db/playroom to generate sfm based sparse data.
• Then I run the training script - python train.py -s /home/ur10/gaussian-splatting/tandt_db/db/playroom.
• Finally I run the render script - python render.py -m /home/ur10/models/playroom -s /home/ur10/gaussian-splatting/tandt_db/db/playroom/ --data_device cuda

I have the following questions -

• I can locate the rendered images however I am not sure where the rendered point clouds are located/how do I render them.
• Moreover I believe running the training script every time is an overkill but is it valid to generate a random point cloud with the scale of my scene and place it in the required directory. As currently I get the following error when I do this - "ValueError: no field of name opacity"

[jaco001]

Output models are in gaussian-splatting software folder in the output. Each folder have unique number, but I suggest you to change it for nice_model_name after training. At defoult in this folder you get point_cloud\iteration_7000 and point_cloud\iteration_30000 folders contains point_cloud_.ply

For 'rendering' I sugest you use SIBR_viewer . For my windows it is like
SIBR_gaussianViewer_app.exe -m "C:\tools\gaussian-splatting\output\szpilki_29012025_b1" --iteration 30000

[ur10]

Thanks for clarifying this, just a quick follow-up, I wanted to know does a point cloud file get generated during the rendering phase as well? If so where is it generated, I can only locate the newly rendered images after I run the last command.

[jaco001]

From main page ;)
render.py -m path to pre-trained model -s path to COLMAP dataset
metrics.py -m path to pre-trained model

path to pre-trained model --> you must already have a model...

[ur10]

Okay, I get that, and maybe I am just being dumb here. But what is the 'path' of the saved point cloud, I just went through the entire render.py script but I don't see anywhere a point cloud being generated while rendering.

[jaco001]

Look at this step by step

E:\splat\gaussian-splatting>python render.py -m "E:\splat\gaussian-splatting\output\beczka" -s "E:
splat\gaussian-splatting\datasets\beczka"
Looking for config file in E:\splat\gaussian-splatting\output\beczka\cfg_args
Config file found: E:\splat\gaussian-splatting\output\beczka\cfg_args
Rendering E:\splat\gaussian-splatting\output\beczka
Loading trained model at iteration 30000 [30/01 19:52:53]
Reading camera 56/56 [30/01 19:52:53]
Loading Training Cameras [30/01 19:52:53]
[ INFO ] Encountered quite large input images (>1.6K pixels width), rescaling to 1.6K.
If this is not desired, please explicitly specify '--resolution/-r' as 1 [30/01 19:52:53]
Loading Test Cameras [30/01 19:53:05]
Rendering progress: 100%|██████████████████████████████████████████████████████████████| 56/56 [01:08<00:00, 1.22s/it]
Rendering progress: 0it [00:00, ?it/s]

This example from my works...
Model 'physicaly' is in e:\splat\gaussian-splatting\output\beczka\point_cloud\iteration_30000\point_cloud.ply
But you provide path to the 'top' like "E:\splat\gaussian-splatting\output\beczka"
There is only rendered pic for GT (ground truth) and 'same' pictures from model
All this going to
"E:\splat\gaussian-splatting\output\beczka\train"
and
"E:\splat\gaussian-splatting\output\beczka\test"

[ur10]

Oh, okay. I get it now. Thanks for your patient explanation.