Deep Geometric Prior for Surface Reconstruction

Abstract

The reconstruction of a discrete surface from a point cloud is a fundamental geometry processing problem that has been studied for decades, with many methods developed. We propose the use of a deep neural network as a geometric prior for surface reconstruction. Specifically, we overfit a neural network representing a local chart parameterization to part of an input point cloud using the Wasserstein distance as a measure of approximation. By jointly fitting many such networks to overlapping parts of the point cloud, while enforcing a consistency condition, we compute a manifold atlas. By sampling this atlas, we can produce a dense reconstruction of the surface approximating the input cloud. The entire procedure does not require any training data or explicit regularization, yet, we show that it is able to perform remarkably well: not introducing typical overfitting artifacts, and approximating sharp features closely at the same time. We experimentally show that this geometric prior produces good results for both man-made objects containing sharp features and smoother organic objects, as well as noisy inputs. We compare our method with a number of well-known reconstruction methods on a standard surface reconstruction benchmark.

Intra-operative Update of Boundary Conditions for Patient-Specific Surgical Simulation

“…In future works, we plan to improve the pre-processing stage, for instance by reconstructing the point cloud from the stereo-endoscope view [2]. Inaccurate surface matching can be addressed by either letting the network implicitly solve the surface correspondence problem as in [16], by providing salient points extracted from camera view to ZoomOut, or by improving surface estimation [25]. In particular, relying on DNN to directly solve for surface correspondences seems promising to further improve the time performances of the current implementation, where ZoomOut is responsible of the main computational overhead.…”

Deep No Learning Approach for Unsupervised Change Detection in Hyperspectral Images

“…This is remarkable as it demonstrates the power of untrained network. Following this work, several other works have followed similar approach demonstrating success of untrained network for different computer vision problems, including photo manipulation (Bau et al, 2020) and surface reconstruction (Williams et al, 2019). Another similar line of research is random projection network (Wójcik, 2018) that is proposed in the context of highdimensional data which implies a network architecture with an input layer that has a huge number of weights, making training infeasible.…”

BAE-NET: Branched Autoencoder for Shape Co-Segmentation

“…The relatively shallow fully-connected stack is high-capacity enough to model non-trivial parts, but shallow enough that the neurons are forced to learn a compact, efficient representation of the shape space, in terms of recurring parts carved out by successive simple units. Thus, the geometric prior is inherent in the architecture itself, similar in spirit to Deep Image Prior [55] and Deep Geometric Prior for Surface Reconstruction [61].…”