Improved Adversarial Systems for 3D Object Generation and Reconstruction

Edward Smith, David Meger
7/29/2017
cs.CV

Official code: EdwardSmith1884/3D-IWGAN (206 stars, MIT, last commit 2020-11-13).

Abstract

This paper describes a new approach for training generative adversarial networks (GAN) to understand the detailed 3D shape of objects. While GANs have been used in this domain previously, they are notoriously hard to train, especially for the complex joint data distribution over 3D objects of many categories and orientations. Our method extends previous work by employing the Wasserstein distance normalized with gradient penalization as a training objective. This enables improved generation from the joint object shape distribution. Our system can also reconstruct 3D shape from 2D images and perform shape completion from occluded 2.5D range scans. We achieve notable quantitative improvements in comparison to existing baselines

View on arXivView PDF

Code Implementations(1)

EdwardSmith1884/3D-IWGANOfficial100%
MIT

A repository for the paper "Improved Adversarial Systems for 3D Object Generation and Reconstruction".

20653Shell, MATLABJul 28, 20175 years agoMIT
3d3d-iwgangangenerationpaper+1 more

Similar Papers

Progressive Growing of GANs for Improved Quality, Stability, and Variation

87% match1 repo6.2k2017

Improved Techniques for Training GANs

85% match1 repo2.3k2016

Generative Adversarial Networks and Other Generative Models

85% match2022

Training Generative Adversarial Networks with Limited Data

84% match1 repo1.8k2020

3D Object Reconstruction from a Single Depth View with Adversarial Learning

83% match1 repo1282017

HoloGAN: Unsupervised learning of 3D representations from natural images

83% match1 repo3332019

Cite this paper

@article{smith2017improved,
  title  = {Improved Adversarial Systems for 3D Object Generation and Reconstruction},
  author = {Edward Smith and David Meger},
  year   = {2017},
  eprint = {1707.09557},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1707.09557}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1707.09557.svg)](https://codeofpaper.com/paper/1707.09557)

Discussion