Problem
Research questions and friction points this paper is trying to address.
Next Best View Selection
3D Gaussian Splatting
Oracle Efficiency
Information Gain
View Selection
LiTe-GS: Oracle-Efficient Next Best View Selection for 3D Gaussian Splatting
🤖 AI Summary
This work addresses the prohibitive computational overhead of information gain evaluation caused by an increasing number of candidate views in 3D Gaussian Splatting. To accelerate next-best-view selection, we propose a randomized subset evaluation strategy that approximates full-set scoring via random subsets, reducing the complexity of Fisher information oracle calls to O(M log(1/ε)). This approach provides an explicit trade-off between efficiency and accuracy with theoretical guarantees. Experiments on the Blender and Mip-NeRF 360 datasets demonstrate that our method substantially reduces the number of evaluations while maintaining reconstruction quality comparable to baselines, thereby enabling efficient adaptive training and refinement.
📝 Abstract
Selecting informative camera views is critical for efficient training and adaptive refinement in 3D Gaussian Splatting, where each observation significantly influences model parameters. However, information-driven view-selection strategies can require repeated evaluations of expensive information-gain oracles as the number of candidate views increases. We propose LiTe-GS, an oracle-efficient method for next best view selection in 3D Gaussian Splatting. LiTe-GS reduces the number of information-oracle evaluations by performing randomized subset evaluation of candidate views rather than exhaustively scoring the full candidate pool. The resulting approach achieves expected $O(M\log(1/ε))$ oracle complexity with respect to the number of candidate views $M$, independent of the selection cardinality $K$, while providing an explicit trade-off between oracle efficiency and approximation quality through $ε$. We provide theoretical guarantees on oracle complexity and approximation performance under the proposed selection scheme. Experiments on Blender and Mip-NeRF 360 demonstrate that LiTe-GS maintains reconstruction quality comparable to Fisher-information-based baselines while substantially reducing the number of Fisher-oracle evaluations across different acquisition settings.
Innovation
Methods, ideas, or system contributions that make the work stand out.
3D Gaussian Splatting
Next Best View Selection
Oracle Efficiency
Randomized Subset Evaluation
Fisher Information
🔎 Similar Papers
No similar papers found.
