/3D-RecGAN-extended

3D-RecGAN++ in Tensorflow (preprint arXiv:1802.00411)

Primary LanguagePython

3D Object Dense Reconstruction from a Single Depth View

Bo Yang, Stefano Rosa, Andrew Markham, Niki Trigoni, Hongkai Wen. arXiv preprint arXiv:1802.00411, 2018.

(1) Architecture

Arch_Image

(2) Sample Results

Teaser_Image

(3) Data

Part 1: {ShapeNetCore.v2: bench, chair, couch, table}, 15G

https://drive.google.com/open?id=12FeihIJ6YL-RiLFQL5OdyVbwbasQGRii

Part 2: {ShapeNetCore.v2: airplane, car, monitor, faucet, guitar, gun}, 3G

https://drive.google.com/open?id=17_GIR5bUj_-g1GRmlyWnbjC1j7adpe5y

Real Dataset: {Kinect: bench, chair, couch, table}

https://drive.google.com/open?id=1oAaYp36GgzvzEdVg8s2j90JnFVok9juq

(4) Released Model

Trained on {bench, chair, couch, table}, 2G

https://drive.google.com/open?id=1IzwZLgRhzd6GVofzdjBZTblxMPH7NuxP

(5) Requirements

python 2.7.6

tensorflow 1.2.0

numpy 1.13.3

scipy 0.19.0

matplotlib 2.0.2

skimage 0.13.0

(6) Run

Training

python main_3D-RecGAN++.py

Test Demo (Download released model first)

python demo_3D-RecGAN++.py

(7) Citation

If you use the paper, code or data for your research, please cite:

@inProceedings{Yang18,
  title={3D Object Dense Reconstruction from a Single Depth View},
  author = {Bo Yang
  and Stefano Rosa
  and Andrew Markham
  and Niki Trigoni
  and Hongkai Wen},
  booktitle={arXiv preprint arXiv:1802.00411},
  year={2018}
}