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.
@article{arxiv.1811.11286,
title = {Patch-based Progressive 3D Point Set Upsampling},
author = {Wang Yifan and Shihao Wu and Hui Huang and Daniel Cohen-Or and Olga Sorkine-Hornung},
journal= {arXiv preprint arXiv:1811.11286},
year = {2019}
}
Comments
accepted to cvpr2019, code available at https://github.com/yifita/P3U