Open apurvvarshney opened 6 days ago
Hi, thanks for your interest!
To get such a visualization, we treated the mean of each 3D Gaussian as a point and performed HDBSCAN clustering on the point cloud (the distance metric for clustering is their distance in the learned contrastive feature space).
The script we used for generating such visualizations can be found here. This script was used with an older version of the codebase and it needs some (minor) adaptations to work. But this script should describe the overall process we used for the visualization well. Feel free to reply if you find it working or not, or make a PR if you get it working!
Hi, can you point me in the right direction on how to generate this view from your model, I am able to train and render my scene using your model.
Thanks!