LiTe-GS speeds up camera view selection in 3D modeling tasks
LiTe-GS: Oracle-Efficient Next Best View Selection for 3D Gaussian Splatting
Computer Vision and Pattern RecognitionRobotics
Summary
When creating 3D models from photos, choosing the best new camera views helps improve the model faster. However, this selection process can be very slow because it involves checking lots of possible views using costly calculations. The authors propose LiTe-GS, a method that efficiently picks the next best view by looking at random subsets instead of all possibilities. This approach keeps the model quality high while greatly cutting down the number of expensive computations needed.
What this means in practice
- •For 3d graphics developers: Improve 3D reconstruction workflows by selecting informative camera angles faster without losing model quality.
- •For robotics perception teams: Enable robots to efficiently choose next sensing viewpoints for building 3D maps with fewer computational resources.
Authors
Vivek Pandey, Amirhossein Mollaei Khass, Nader Motee
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.