Skip to content
Deep Single-View 3D Object Reconstruction with Visual Hull Embedding
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
__pycache__ update Feb 24, 2019
readme update Feb 25, 2019
real_data/chair first commit Feb 24, 2019
syn_data/chair first commit Feb 24, 2019
.DS_Store update Feb 25, 2019
LICENSE update Feb 25, 2019 update readme Mar 3, 2019 first commit Feb 24, 2019 first commit Feb 24, 2019 update Feb 25, 2019 first commit Feb 24, 2019


This repository is the implementation of our AAAI 2019 paper:

Deep Single-View 3D Object Reconstruction with Visual Hull Embedding

Hanqing Wang, Jiaolong Yang, Wei Liang, Xin Tong

This work is implemented using TensorFlow.


In this paper, we present an approach which aims to preserve more shape details and improve the reconstruction quality. The key idea of our method is to leverage object mask and pose estimation from CNNs to assist the 3D shape learning by constructing a probabilistic single-view visual hull inside of the network.

Our method works by first predicting a coarse shape as well as the object pose and silhouette using CNNs, followed by a novel 3D refinement CNN which refines the coarse shapes using the constructed probabilistic visual hulls.



If you find our work helpful for your research, please cite our paper:

  author = {Wang, Hanqing and Yang, Jiaolong and Liang, Wei and Tong, Xin},
  title = {Deep Single-View 3D Object Reconstruction with Visual Hull Embedding},
  booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence (AAAI)},
  year = {2019}


Install python and the dependencies:

  • python 3.5
  • tensorflow 1.12.0
  • pillow

If your python environments are managed via Anaconda/Miniconda, you can install the dependencies using the following scrpit:

conda install tensorflow pillow

In our paper, we trained two network for rendered images and real images respectively. The checkpoint of the trained models are available here(426MB). Extract the files to the root directory.


Run python to run the examples. The outputs are reconstruction results before and after the refinement (Please refer to our paper for more details). The results are in obj format. You can use meshlab for visulization.


Part of this work was done when I was an intern at Microsoft Research Asia. It is impossible to finish without the help of my colleagues and my mentor.


PSVH is freely available for non-commercial use, and may be redistributed under these conditions. Please see the license for further details. For commercial license, please contact the authors.

You can’t perform that action at this time.