| --- |
| license: mit |
| viewer: false |
| language: |
| - en |
| task_categories: |
| - keypoint-detection |
| pretty_name: Sign Language MediaPipe Keypoints |
| size_categories: |
| - 1K<n<10K |
| --- |
| # Sign Language and Gesture Recognition: MediaPipe Keypoints |
|
|
| ## Dataset Summary |
|
|
| This dataset contains pre-extracted MediaPipe keypoints for 29 distinct sign language gestures. It is specifically designed to train lightweight machine learning models—such as Dense Neural Networks (DNNs) or LSTMs—for real-time sign language translation without the computational overhead of processing raw images during training. |
|
|
| By providing the coordinate data directly, this dataset allows researchers and developers to bypass the MediaPipe extraction pipeline and jump straight into model architecture and training. |
|
|
| --- |
|
|
| ## Classes (29 Total) |
|
|
| The dataset includes both the standard alphabet and functional communication commands: |
|
|
| - **Alphabets (26 classes):** `A` through `Z` |
| - **Functional Commands (3 classes):** `space`, `speak`, `stop` |
|
|
| --- |
|
|
| ## Dataset Structure (Zip Archive) |
|
|
| To ensure fast and reliable downloading, the dataset is packaged as a single compressed archive: |
|
|
| ```text |
| media_pipe_keypoints_dataset.zip |
| ``` |
|
|
| Once extracted, the folder contains no predefined train/test split. It is structured into 29 distinct class folders, allowing researchers to implement custom validation strategies such as K-Fold Cross Validation or custom random splits. |
|
|
| ### Folder Structure |
|
|
| ```text |
| media_pipe_keypoints_dataset/ |
| ├── A/ |
| ├── B/ |
| ├── C/ |
| ├── ... |
| ├── Z/ |
| ├── space/ |
| ├── speak/ |
| └── stop/ |
| ``` |
|
|
| --- |
|
|
| ## How to Use This Dataset in Python |
|
|
| Since the dataset is zipped, here is a quick snippet to download and extract it directly in your Python code or Jupyter Notebook. |
|
|
| ```python |
| from huggingface_hub import hf_hub_download |
| import zipfile |
| import os |
| |
| # 1. Download the zip file from Hugging Face |
| zip_path = hf_hub_download( |
| repo_id="om192006/sign_language_keypoints", # Replace with your exact repo ID |
| filename="media_pipe_keypoints_dataset.zip", |
| repo_type="dataset" |
| ) |
| |
| # 2. Unzip the dataset into a local folder |
| extract_dir = "./sign_language_data" |
| os.makedirs(extract_dir, exist_ok=True) |
| |
| with zipfile.ZipFile(zip_path, 'r') as zip_ref: |
| zip_ref.extractall(extract_dir) |
| |
| print(f"Dataset extracted successfully to: {extract_dir}") |
| |
| # You can now iterate through the A-Z folders inside extract_dir |
| ``` |
|
|
| --- |
|
|
| ## Potential Use Cases |
|
|
| ### Real-time Gesture Translation |
| Training sequential or dense neural networks to classify gestures from live webcam feeds. |
|
|
| ### Accessibility Technology |
| Developing applications that bridge communication gaps for mute and hard-of-hearing communities. |
|
|
| ### Educational Tools |
| Building interactive systems to help users learn and practice sign language. |
|
|
| --- |
|
|
| ## Author |
|
|
| Created by **Om Pradip Chougule**. |
|
|
| ## Citation |
|
|
| If you use this dataset, please provide appropriate attribution by citing the repository in your research, projects, or publications. |
