Hey! This is a very cool project!!! I am trying to leverage your work for my thesis, where I'm attempting to make Image Foundation Models (FMs) more spatially aware using Gaussian Splatting for the autonomous driving domain. Because of the 'long tail' in the data, I want to find the first K prototypes using clustering as a fixed reference set. What are the most important factors to consider here? Can I simply use DINO features, and are there any specific tricks you would recommend?
Hey! This is a very cool project!!! I am trying to leverage your work for my thesis, where I'm attempting to make Image Foundation Models (FMs) more spatially aware using Gaussian Splatting for the autonomous driving domain. Because of the 'long tail' in the data, I want to find the first K prototypes using clustering as a fixed reference set. What are the most important factors to consider here? Can I simply use DINO features, and are there any specific tricks you would recommend?