Uncertainty improves sparse-view 3D Gaussian rendering quality and compactness
UGOD: Uncertainty-Guided Opacity and Dropout for Sparse-View 3D Gaussian Splatting
Computer Vision and Pattern Recognition
Summary
Creating 3D images from few pictures can lead to mistakes because some parts are not well-defined but still contribute to the final image. The authors show a new way to estimate how confident the system is about each part, then reduce the influence of uncertain parts so they don’t mess up the picture. They also use a dropout technique that teaches the system to rely less on unclear parts during training. This approach leads to clearer 3D images and fewer, more meaningful parts in the model.
What this means in practice
- •For computer graphics developers: Produce higher quality 3D reconstructions from limited viewpoints by reducing artifacts from uncertain model components in Gaussian splatting renderers.
- •For augmented reality engineers: Generate more reliable and compact 3D scene representations from sparse sensor data to improve real-time rendering and device performance.
Authors
Zhihao Guo, Peng Wang, Zidong Chen, Xiangyu Kong, Yan Lyu, Guanyu Gao, Chenghao Qian, Ziyang Wang, Xinqi Fan, Liangxiu Han
Abstract
Sparse-view 3D Gaussian Splatting is prone to overfitting because limited observations leave many Gaussian primitives weakly constrained, yet their contributions are still accumulated through alpha blending. Without uncertainty estimation, the renderer cannot distinguish unreliable primitives from well-constrained ones, allowing their erroneous contributions to corrupt novel-view synthesis. We introduce UGOD, an uncertainty-guided framework that estimates a view-dependent uncertainty score for each Gaussian and uses it to regulate its rendering contribution. A lightweight uncertainty head conditioned on Gaussian attributes and viewing direction predicts this score, which then drives a differentiable opacity-modulation mechanism that attenuates high-uncertainty primitives before compositing. During training, a detached soft-dropout branch applies an uncertainty-controlled continuous keep mask to discourage the model from relying on poorly constrained Gaussians and thereby reduce overfitting. Crucially, detaching the uncertainty score prevents gradients from this stochastic regulariser from biasing or collapsing the uncertainty prediction. Experiments on Mip-NeRF~360 and LLFF show that UGOD improves sparse-view novel-view synthesis while producing more compact Gaussian representations than the compared methods. These results demonstrate that Gaussian uncertainty provides an effective rendering-time control for sparse-view reconstruction.