Patch-based Progressive 3D Point Set Upsampling

TL;DR

This paper introduces a neural network method for progressively upsampling 3D point clouds, significantly improving detail and fidelity for rendering and reconstruction tasks.

Contribution

It proposes a novel cascade of patch-based networks trained end-to-end for progressive 3D point set upsampling, with architectural innovations validated through ablation studies.

Findings

Outperforms state-of-the-art methods in detail preservation

Handles low-resolution inputs effectively

Reveals high-fidelity details in upsampled point sets

Abstract

We present a detail-driven deep neural network for point set upsampling. A high-resolution point set is essential for point-based rendering and surface reconstruction. Inspired by the recent success of neural image super-resolution techniques, we progressively train a cascade of patch-based upsampling networks on different levels of detail end-to-end. We propose a series of architectural design contributions that lead to a substantial performance boost. The effect of each technical contribution is demonstrated in an ablation study. Qualitative and quantitative experiments show that our method significantly outperforms the state-of-the-art learning-based and optimazation-based approaches, both in terms of handling low-resolution inputs and revealing high-fidelity details.