Active View Selection for 3D Gaussian Splatting

Published:Dec 28, 2025 04:19
1 min read
ArXiv

Analysis

This paper addresses the problem of efficiently training 3D Gaussian Splatting models for semantic understanding and dynamic scene modeling. It tackles the data redundancy issue inherent in these tasks by proposing an active learning algorithm. This is significant because it offers a principled approach to view selection, potentially improving model performance and reducing training costs compared to naive methods.

Reference

The paper proposes an active learning algorithm with Fisher Information that quantifies the informativeness of candidate views with respect to both semantic Gaussian parameters and deformation networks.