cadenlpicard/dance_images

Dataset Card for Dance Style Classification Dataset

This dataset contains video frame images extracted from dance videos, annotated with their respective dance styles. It is designed to support machine learning tasks such as image classification and style recognition.

Dataset Details

Dataset Description

The dataset consists of images extracted from dance videos downloaded from Youtube, labeled with dance styles such as ballet, breakdance, contemporary, hip hop, jazz, salsa, and tap. Each image represents a frame from a video, and the dataset is intended for tasks like dance style recognition using deep learning models.

• Curated by: [Caden Picard]
• License: CC BY-SA 4.0

Dataset Sources

All videos were taken from youtube.com.

Uses

Direct Use

The dataset is intended for training and evaluating machine learning models for dance style classification. It is suitable for use in research and educational projects on computer vision and activity recognition.

Out-of-Scope Use

The dataset should not be used for any malicious activities, such as identifying or targeting individuals based on their movements or creating synthetic datasets for deceptive purposes.

Dataset Structure

The dataset contains the following:

• Images: JPG files extracted from video frames. Filenames follow the pattern <video_name>_<frame_index:03d>.jpg.
• Labels: Dance style annotations corresponding to each frame, provided in a CSV file (training_labels.csv).

Example of the CSV structure:

FrameFileName,DanceStyle
23kb58sd0,ballet
99fl4swiz0,jazz

Dataset Creation

Curation Rationale

The dataset was created to facilitate research in dance recognition and improve machine learning algorithms for identifying human activities in videos.

Source Data

Data Collection and Processing

Video clips were sourced from Youtube.com available platforms, and frames were extracted using ffmpeg. Each video was labeled with the associated dance style, and annotations were curated manually.

Who are the source data producers?

The videos were created by dance instructors and enthusiasts, with styles ranging from classical to contemporary.

Annotations

Annotation Process

Annotations were created by utilizing the search result and video name the dance style it represents. The labels were applied uniformly to all extracted frames from a video.

Personal and Sensitive Information

The dataset does not contain personal or sensitive information. Videos were sourced from publicly available repositories and do not include identifiable data.

Bias, Risks, and Limitations

The dataset may have bias towards specific dance styles or regions based on the availability of source videos. It does not fully represent global dance styles and may miss variations within a single style.

Recommendations

Users should be cautious when applying models trained on this dataset to real-world scenarios, as the limited diversity of styles may affect generalization.

Citation

BibTeX:

@dataset{caden_picard_dance_style,
  title={Dance Style Classification Dataset},
  author={Caden Picard},
  year={2024},
  howpublished={\url{https://huggingface.co/datasets/dance_style}}
}

APA:

Picard, C. (2024). Dance Style Classification Dataset. Retrieved from https://huggingface.co/datasets/dance_style

Glossary

• Dance Style: A category or genre of dance, such as ballet or hip hop.
• Frame: A single image extracted from a video.

More Information

For more details, please contact the dataset curator.

Dataset Card Authors

Caden Picard

Dataset Card Contact

Caden Picard