SAL: Sign Agnostic Learning of Shapes from Raw Data

TL;DR

SAL is a novel deep learning method that learns implicit shape representations directly from raw, unsigned geometric data like point clouds, enabling improved surface reconstruction without requiring signed ground-truth functions.

Contribution

It introduces Sign Agnostic Learning (SAL), a new approach that bypasses the need for signed distance or occupancy functions, simplifying shape learning from raw data.

Findings

Achieved state-of-the-art surface reconstruction from un-oriented point clouds.

Enabled end-to-end human shape space learning from raw scans.

Reduced manual pre-processing in geometric deep learning applications.

Abstract

Recently, neural networks have been used as implicit representations for surface reconstruction, modelling, learning, and generation. So far, training neural networks to be implicit representations of surfaces required training data sampled from a ground-truth signed implicit functions such as signed distance or occupancy functions, which are notoriously hard to compute. In this paper we introduce Sign Agnostic Learning (SAL), a deep learning approach for learning implicit shape representations directly from raw, unsigned geometric data, such as point clouds and triangle soups. We have tested SAL on the challenging problem of surface reconstruction from an un-oriented point cloud, as well as end-to-end human shape space learning directly from raw scans dataset, and achieved state of the art reconstructions compared to current approaches. We believe SAL opens the door to many geometric deep learning applications with real-world data, alleviating the usual painstaking, often manual pre-process.