ElizabethBV/MSV_PG

---
license: cc-by-4.0
task_categories:
- video-classification
size_categories:
- 100B<n<1T
---
# Dataset Card for Dataset Name

Multi-Scale Violence and Public Gathering (MSV-PG) Dataset

This dataset classifies public events along two axes: the size of the crowd observed and the level of perceived violence in the crowd.
The videos were annotated to identify temporal segments corresponding to specific behavioral events. Click [here](https://www.frontiersin.org/journals/computer-science/articles/10.3389/fcomp.2024.1242690/full) for the details of the dataset.

## Dataset Details

### Classes

- 'natural' – Everyday scenes without significant events.
- 'lpg' – Large peaceful gatherings - Large groups gathered without violence.
- 'lvg' – Large violent gatherings - Widespread violence within a large crowd.
- 'violence' – Small-scale fighting events- Localized violent interactions among a smaller number of individuals.

## Dataset Structure

The dataset follows a hierarchical organization:

```text
MSV_PG Dataset/
│── train/
│ ├── natural/
│ │ ├── sample_000000001_1667_0001/
│ │ │ ├── frame_01.jpg
│ │ │ ├── frame_02.jpg
│ │ │ └── ...
│ │ ├── sample_000000002_0882_0028/
│ │ └── ...
│ │
│ ├── lpg/
│ ├── lvg/
│ └── violence/
│
│── val/
│ ├── natural/
│ ├── lpg/
│ ├── lvg/
│ └── violence/
│
│── train_sample_list.csv
│── valid_sample_list.csv
│── README.md
```

Each sample represents a short temporal segment of a video (e.g., ~2 seconds) and contains a fixed number of frames
extracted at a consistent frame rate. All frames within a sample belong to the same class label.

The CSV files act as index files and typically include:
```text
- Path to each sample folder
- Corresponding class label
```
They facilitate efficient loading of datasets during training and evaluation.

## Citation

Elzein A, Basaran E, Yang YD and Qaraqe M (2024) A novel multi-scale violence and public gathering dataset for crowd behavior classification. Front. Comput. Sci. 6:1242690. doi: 10.3389/fcomp.2024.1242690

**BibTeX:**

```text
@article{elzein2024novel,
title={A novel multi-scale violence and public gathering dataset for crowd behavior classification},
author={Elzein, Almiqdad and Basaran, Emrah and Yang, Yin David and Qaraqe, Marwa},
journal={Frontiers in Computer Science},
volume={6},
pages={1242690},
year={2024},
publisher={Frontiers Media SA},
doi= {10.3389/fcomp.2024.1242690}
}
```