Skip to content

Synthetically Generated Surveillance Perspective Human Action Recognition Dataset: 6901 Videos from 10 action classes, made by a 3D Simulation, all cropped spatio-temporally and filmed from a surveillance-camera like position.

License

Notifications You must be signed in to change notification settings

AlexanderMelde/S-SPHAR-Dataset

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

26 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

S-SPHAR Synthetic Surveillance Perspective Human Action Recognition Logo

S-SPHAR: Synthetic Surveillance Perspective Human Action Recognition Dataset

S-SPHAR is a synthetically generated video dataset for human action recognition. Its main purpose is to support research in the application area of analyzing activities on public places.

In this domain, most cameras will share a similar mounting angle and perspective, which we will call the surveillance perspective from now on. In S-SPHAR, all videos are shot from this or a similar perspective.

The videos have been generated using a self-made simulation built using the Unity game engine and characters and animations from Mixamo.

This Repository contains all videos of the S-SPHAR dataset.


Looking for the non-synthetic SPHAR-Dataset? Click here


Dataset Overview

Dataset Version # Videos # Classes Videos per Class Dataset Size Original Video Length Original Video Size Year
1 260 9 (HMDB) 8 - 92 40 MB 02:36 min 860 MB 2020
2 696 10 (SPHAR) 3 - 236 168 MB 09:04 min 4.01 GB 2020
3 6901 10 (SPHAR) 42 - 2328 1.03 GB 48:22 min 12.9 GB 2020

Original Videos

Due to their size, the original videos were not uploaded to GitHub. Instead, they can be found at YouTube under the following links. Read their descriptions for more information:

preview image of S-SPHAR videos

Those Videos are combinations of the following video views:

  • Original camera image
  • Segmentation masks for labeled action classes
  • Overlay of segmentation masks over original camera image
  • Bounding boxes around action instance tubes

The videos are all 4K resolution, so you can split it back into four FHD videos using simple tools like OpenCV or other video editors.

Download

The easiest way to just get the cropped dataset videos is by downloading one of our releases:

You can also clone or fork this repository using:

git clone git@github.com:AlexanderMelde/S-SPHAR-Dataset.git

If you want to download the original videos, use the links above and a YouTube downloader like youtube-dl.

License

The S-SPHAR Dataset is released under the GNU GPL v3 license (contact me for further licensing options).

See the LICENSE file for more details.

Citation

If you want to cite and reference the S-SPHAR Dataset, please link to this GitHub page. You can use the following BibTex entry:

@article{s-sphar-dataset,
  title={S-SPHAR: Synthetic Surveillance Perspective Human Action Recognition Dataset},
  author={Alexander Melde},
  year={2020},
  publisher = {GitHub},
  journal = {GitHub repository},
  url = {https://github.com/AlexanderMelde/S-SPHAR-Dataset},
  version = {\UrlFont\href{https://github.com/AlexanderMelde/S-SPHAR-Dataset/commit/a6131c7}{a6131c7}},
  urldate={2020-09-29}
}

(replace urldate with your own day of retrieval)

About

Synthetically Generated Surveillance Perspective Human Action Recognition Dataset: 6901 Videos from 10 action classes, made by a 3D Simulation, all cropped spatio-temporally and filmed from a surveillance-camera like position.

Topics

Resources

License

Stars

Watchers

Forks