Point Cloud Clustering Issues

[DuVogel87]

Here is a problem I am facing. I am trying to figure out why Splatica achieves better results from the same dataset than I currently can. I use GSplat for training and have used Metashape and Colmap for structure-from-motion (SFM).

The images I am sharing include a screenshot from Splatica and one from GSplat. They show that the "point cloud" in Splatica appears to be much more evenly distributed compared to what I get from my Colmap/Metashape input.

I took some screenshots of the final Gaussians inside Supersplat and enabled the "points" view. It seems that in my point cloud, a certain pattern is visible where many points cluster together, followed by empty spaces, and then clustering again.

I currently believe this issue is not related to the SFM process because within the Colmap and Metashape GUIs, the point cloud looks more evenly distributed. Instead, I suspect it has something to do with the training process, and I wonder if anyone has insights into what might be causing this and how I can achieve a more evenly distributed point cloud.

I have attached two images for reference. The first image is the Splatica and the second image the Gsplat model within Supersplat.