search

Pokerlishao / LoopGaussian

https://pokerlishao.github.io/LoopGaussian/
MIT License
35 stars 1 forks source link

Cluster #5

Open Tzy010822 opened 3 months ago

Tzy010822 commented 3 months ago

I saw other implementations of kmeans in the code. Why did you choose to use VCCS for clustering?

Pokerlishao commented 3 months ago

Thank you for your great question.

In fact, the core reason is the memory usage. Especially when the number of 3D Gaussian points is huge, the memory consumption of the K-means method is unacceptable. In addition, VCCS is designed for point cloud data, making it more effective at preserving the geometric features of objects by focusing on voxel connectivity. In our experiment, it can indeed meet our requirements.

Tzy010822 commented 3 months ago

I noticed that Cluster is integrated into the binary files in the lib folder. Can you please open source the process of how to get these binary files?

Pokerlishao commented 2 months ago

Thank you for bringing this up! In this version of the code, the clustering function found in the lib directory mainly references the Repository. And I wrapped it to fit the specific requirements of our task. If you’re interested in exploring how the clustering process is handled, I would recommend checking out the Repository, where you can find the original source code.