Neural Stochastic Screened Poisson Reconstruction

SIGGRAPH Asia 2023
Silvia Sellán, University of Toronto
Alec Jacobson, University of Toronto and Adobe Research
Teaser image
We use a neural network to quantify the reconstruction uncertainty in Poisson Surface Reconstruction (center left), allowing us to efficiently select next sensor positions (center right) and update the reconstruction upon capturing data (right).


Reconstructing a surface from a point cloud is an underdetermined problem. We use a neural network to study and quantify this reconstruction uncertainty under a Poisson smoothness prior. Our algorithm addresses the main limitations of existing work and can be fully integrated into the 3D scanning pipeline, from obtaining an initial reconstruction to deciding on the next best sensor position and updating the reconstruction upon capturing more data.



  title = {Neural Stochastic Screened Poisson Reconstruction},
  author = {Silvia Sellán and Alec Jacobson},
  year = {2023},
  journal = {ACM Transactions on Graphics (Proc. SIGGRAPH Asia)}


This project is funded in part by NSERC Discovery (RGPIN2017-05235, RGPAS-2017-507938), New Frontiers of Research Fund (NFRFE-201), the Ontario Early Research Award program, the Canada Research Chairs Program, a Sloan Research Fellowship and the DSI Catalyst Grant program. The first author is funded in part by an NSERC Vanier Scholarship.

We thank Kirill Serkh, Kiriakos Kutulakos, David Lindell, Eitan Grinspun, David I.W. Levin, Oded Stein, Andrea Tagliasacchi, Otman Benchekroun, Lily Goli and Claas A. Voelcker for insightful conversations that inspired us in this work; Hsueh-Ti Derek Liu for his help rendering our results; as well as Rafael Rodrigues (Fig. 6, CC BY-NC-SA 4.0) and ShaggyDude (Fig. 13, CC BY 4.0) for releasing their 3D models for academic use. We would also like to thank Xuan Dam, John Hancock and all the University of Toronto Department of Computer Science research, administrative and maintenance staff.