README.md

---
license: mit
tags:
  - ASL
  - American Sign Language
  - Gesture Recognition
  - Video Dataset
---

# American Sign Language (ASL) Dataset

**Description:**  
This dataset contains **108,618 videos** representing **2,208 ASL words**, with each word having a minimum of **30 videos**. The videos were **scraped, collected from multiple sources, and preprocessed** to ensure consistency, quality, and usability for machine learning and gesture recognition tasks. Each video is **≤10 MB**, optimized for storage and model training.  
The dataset can be used for ASL gesture recognition, video-based ML tasks, and model training for research and educational purposes.

## Folder Structure

E:\Dataset\American-Sign-Language-Dataset
├── part1\       # First subset of videos (≤10,000 files)
├── part2\       # Second subset of videos (≤10,000 files)
├── part3\       # Third subset of videos (≤10,000 files)
├── README.md    # This file
└── dataset.csv  # CSV file mapping words to video paths

- Each subfolder contains **≤10,000 video files** to comply with Hugging Face repository limits.  
- Videos are organized into multiple parts to prevent exceeding per-folder file limits.  
- `dataset.csv` maps each ASL word to its respective video path.  
- `README.md` provides detailed dataset information and usage instructions.

## CSV File (`dataset.csv`)

The CSV contains the following columns:

| Column       | Description                                 |
|-------------|---------------------------------------------|
| `word`      | The ASL word represented in the video       |
| `video_path`| Relative path to the video file in the repo |

**Example:**

| word   | video_path                |
|--------|---------------------------|
| HELLO  | part1/12345-HELLO.mp4    |
| THANKS | part2/67890-THANKS.mp4   |
| YES    | part3/54321-YES.mp4      |

- All paths are **relative to the root dataset folder**.

## Dataset Statistics

- **Total videos:** 108,618  
- **Total words:** 2,208  
- **Minimum videos per word:** 30  
- **Maximum files per folder:** ≤10,000 (split into 3 parts)

## Usage Example

import pandas as pd

# Load CSV
df = pd.read_csv("dataset.csv")

# Example: filter videos for the word "HELLO"
hello_videos = df[df['word'] == 'HELLO']['video_path'].tolist()
print(hello_videos)

## Key Usage Notes

- Recommended to split dataset into subfolders due to repository file limits.  
- Dataset combines multiple sources and has been carefully cleaned to remove duplicates and corrupted videos.  
- Ensure that all video paths in `dataset.csv` match the actual subfolder structure when using programmatically.  
- Large uploads were handled using Hugging Face's `upload-large-folder` utility.

## Citation

If using this dataset in research:

American Sign Language Dataset, collected and preprocessed by Zahid Yasin Mittha, 2025.

## License

This dataset is released under the **MIT License**.  
You are free to use, modify, and distribute it for research, educational, or commercial purposes, with proper attribution.