Home

Awesome

Multi-View Human Action (MVHA) Dataset

Official repository of "Rotationally-Temporally Consistent Novel View Synthesis of Human Performance Video" (ECCV 2020 Spotlight) <br/> [Paper (ECCV)]

<br/>

Youngjoong Kwon, Stefano Petrangeli, Dahun Kim, Haoliang Wang, Eunbyung Park, Viswanathan Swaminathan, Henry Fuchs.

Dataset Overview

<img src="./images/supple_dataset_overview (1).png" width="800"> <br/>

We provide multi-view human action (MVHA) dataset consists of RGB and binary mask (foreground/background) which are rendered from multiple views. For now, we provide images that are generated from 18 views along the azimuth with 20-degree interval.

Download Instructions

a. Before downloading the dataset, first use the following command:

pip install gdown
conda install zip unzip

b. If you want to download 256x256 dataset, use the following command:

gdown https://drive.google.com/uc?id=1FmxdgfQmjsGySNyWLZ_WaCN9pQB0Gc7o

After the download, you will see a zip file, e.g. xxx.zip unzip the zip file using the following command:

unzip xxx.zip

After the unzip process, you will see the rgb image folder and mask image folder(foreground/background --> foreground is denoted as 1 and background as 0)

c. If you want to download 512x512 dataset, use the following command:

gdown https://drive.google.com/uc?id=1V4UbHXAAmdKM9v63DvP-_kmp1NlV7skw

After the download, you will see the zip file, ex) xxx.zip unzip the zip file using the following command:

unzip xxx.zip

Every image is 4 channel which is: RGB+mask (foreground/background -> foreground is denoted as 1 and background as 0)

Citation

If you use this dataset in your research, please cite following two papers.

Rotationally-Temporally Consistent Novel View Synthesis of Human Performance Video (ECCV 2020) <br/>

@inproceedings{kwon2020rotationally,
  title={Rotationally-Temporally Consistent Novel View Synthesis of Human Performance Video},
  author={Kwon, YoungJoong and Petrangeli, Stefano and Kim, Dahun and Wang, Haoliang and Park, Eunbyung and Swaminathan, Viswanathan and Fuchs, Henry},
  booktitle={Proceedings of the European Conference on Computer Vision (ECCV)},
  year={2020}
}
<br/> Rotationally-Consistent Novel View Synthesis for Humans (ACM MM 2020) <br/> 
@inproceedings{kwon2020rotationally,
  title={Rotationally-Consistent Novel View Synthesis for Humans},
  author={Kwon, Youngjoong and Petrangeli, Stefano and Kim, Dahun and Wang, Haoliang and Fuchs, Henry and Swaminathan, Viswanathan},
  booktitle={Proceedings of the 28th ACM International Conference on Multimedia},
  pages={2308--2316},
  year={2020}
}

License

The Multi-View Human Action (MVHA) dataset is released under the CC-BY-NC license (https://creativecommons.org/licenses/by-nc/4.0/)

Contact

If you have any questions regarding the repo, please contact Youngjoong Kwon (youngjoong@cs.unc.edu) or create an issue.