reddyav1 / RoCoG-v2

RoCoG-v2 (Robot Control Gestures) is a dataset intended to support the study of synthetic-to-real and ground-to-air video domain adaptation.

Robot Control Gestures (RoCoG-v2)

This repository provides access to the RoCoG-v2 gesture recognition dataset introduced in the ICRA 2023 paper "Synthetic-to-Real Domain Adaptation for Action Recognition: A Dataset and Baseline Performances."

RoCoG-v2 (Robot Control Gestures) is a dataset intended to support the study of synthetic-to-real and ground-to-air video domain adaptation. It contains over 100K synthetically-generated videos of human avatars performing gestures from seven (7) classes. It also provides videos of real humans performing the same gestures from both ground and air perspectives.

Downloading the Dataset

All of the data for RoCoG-v2 can be found here. Each type of data is provided in a separate zip file.

You may download the data through the browser or using the following command:

wget https://www.cis.jhu.edu/~rocog/data/<FILENAME>

Replace <FILENAME> in the above command with a filename from the table below corresponding to the data type you wish to download.

Filename	Description
syn_ground.zip	Synthetic videos rendered from the ground perspective
syn_air.zip	Synthetic videos rendered from the air perspective (static hover)
syn_orbital.zip*	Synthetic videos rendered from the air perspective (orbiting the subject)
real_ground.zip	Real cropped videos collected from the ground perspective
real_air.zip	Real cropped videos collected from the air perspective
real_uncropped.zip	Uncropped versions of the real ground and air videos

* Our paper does not provide experimental results with this data type.

Dataset Details

Number of Videos across Splits

Data Type	View	Train	Test	Total
Synthetic	Ground	53,438	-	53,438
Synthetic	Air	53,558	-	53,558
Real	Ground	204	100	304
Real	Air	87	91	178

Dataset Directory Structure

.
├── annotations
├── real
│   ├── air
│   │   ├── Advance
│   │   ├── Attention
│   │   ├── FollowMe
│   │   ├── Halt
│   │   ├── MoveForward
│   │   ├── MoveInReverse
│   │   └── Rally
│   └── ground
│       ├── Advance
│       ├── Attention
│       ├── FollowMe
│       ├── Halt
│       ├── MoveForward
│       ├── MoveInReverse
│       └── Rally
└── syn
    ├── air
    │   ├── Advance
    │   ├── Attention
    │   ├── FollowMe
    │   ├── Halt
    │   ├── MoveForward
    │   ├── MoveInReverse
    │   └── Rally
    └── ground
        ├── Advance
        ├── Attention
        ├── FollowMe
        ├── Halt
        ├── MoveForward
        ├── MoveInReverse
        └── Rally

Citation

If you use this dataset in your work, please cite our ICRA 2023 paper:

bibtex

@inproceedings{2023rocogv2,
  title={Synthetic-to-Real Domain Adaptation for Action Recognition: A Dataset and Baseline Performances},
  author={Reddy, Arun V and Shah, Ketul and Paul, William and Mocharla, Rohita and Hoffman, Judy and Katyal, Kapil D and Manocha, Dinesh and de Melo, Celso M and Chellappa, Rama},
  booktitle={IEEE International Conference on Robotics and Automation (ICRA)},
  pages={},
  year={2023},
  organization={IEEE}
}

APA

Reddy, A. V., Shah, K., Paul, W., Mocharla, R., Hoffman, J., Katyal, K. D., Manocha, D., de Melo, C. M., & Chellappa, R. (2023). Synthetic-to-Real Domain Adaptation for Action Recognition: A Dataset and Baseline Performances. IEEE International Conference on Robotics and Automation (ICRA).

IEEE

A. V. Reddy et al., “Synthetic-to-Real Domain Adaptation for Action Recognition: A Dataset and Baseline Performances,” in IEEE International Conference on Robotics and Automation (ICRA), 2023.