Sphere-Guided Training of Neural Implicit Surfaces

CVPR 2023

Andreea Dogaru1,2     Andrei-Timotei Ardelean1,2     Savva Ignatyev1
Egor Zakharov1     Evgeny Burnaev1,3

1Skolkovo Institute of Science and Technology     2Friedrich-Alexander-Universität Erlangen-Nürnberg
3Artificial Intelligence Research Institute


In recent years, neural distance functions trained via volumetric ray marching have been widely adopted for multi-view 3D reconstruction. These methods, however, apply the ray marching procedure for the entire scene volume, leading to reduced sampling efficiency and, as a result, lower reconstruction quality in the areas of high-frequency details. In this work, we address this problem via joint training of the implicit function and our new coarse sphere-based surface reconstruction. We use the coarse representation to efficiently exclude the empty volume of the scene from the volumetric ray marching procedure without additional forward passes of the neural surface network, which leads to an increased fidelity of the reconstructions compared to the base systems. We evaluate our approach by incorporating it into the training procedures of several implicit surface modeling methods and observe uniform improvements across both synthetic and real-world datasets.

Spheres Optimization

We jointly optimize a sphere cloud toward the current surface estimated by the implicit function. This coarse, explicit representation is used to bound the volume in which the implicit function is evaluated, guiding the ray-sampling and ray-marching procedures used during training. The sphere radius decreases over time, producing a tighter volume around the object's surface as the implicit surface becomes more accurate.


Video comparison



Interactive visualization




Base method    


        author    = {Dogaru, Andreea and Ardelean, Andrei-Timotei and Ignatyev, Savva and Zakharov, Egor and Burnaev, Evgeny},
        title     = {Sphere-Guided Training of Neural Implicit Surfaces},
        booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
        month     = {June},
        year      = {2023},
        pages     = {20844-20853}