SignLanguage-Dataset-Hub

Quick Start Guide

Welcome to the World Sign Language Dataset Hub! This guide will help you get started in 5 minutes.

🚀 What Would You Like to Do?

1️⃣ I want to download datasets

# Clone this repository
git clone https://github.com/rudra496/SignLanguage-Dataset-Hub.git
cd SignLanguage-Dataset-Hub

# List all available datasets
python scripts/download_datasets.py --list

# Download from Kaggle (requires Kaggle API)
pip install kaggle
kaggle datasets download -d datamunge/sign-language-mnist
kaggle datasets download -d grassknoted/asl-alphabet
kaggle datasets download -d ahmedkhan123/arabic-sign-language

# Download from Zenodo
wget https://zenodo.org/record/7067906/files/BdSL47.zip

# Download from Hugging Face
pip install datasets
python -c "from datasets import load_dataset; ds = load_dataset('banglagov/Ban-Sign-Sent-9K-V1')"

2️⃣ I want to use the demo sensor data

# Install requirements
pip install torch numpy

# Load demo data
import json

with open('data/bdsl/BdSL-Sensor-Glove/train/data_train.json') as f:
    data = json.load(f)

print(f"Loaded {len(data)} samples")

# Access a sample
sample = data[0]
print(f"Gesture: {sample['gesture_id']}")
print(f"Duration: {sample['duration_ms']}ms")
print(f"Flex sensors: {len(sample['channels']['flex_thumb'])} readings")

3️⃣ I want to build a sign language recognizer

import torch
import torch.nn as nn
import json
import numpy as np

# Simple LSTM classifier
class SignLanguageClassifier(nn.Module):
    def __init__(self, input_size=11, hidden_size=128, num_classes=36):
        super().__init__()
        self.lstm = nn.LSTM(input_size, hidden_size, batch_first=True, bidirectional=True)
        self.fc = nn.Linear(hidden_size * 2, num_classes)
    
    def forward(self, x):
        lstm_out, _ = self.lstm(x)
        return self.fc(lstm_out[:, -1, :])

# Load and prepare data
def load_data(filepath):
    with open(filepath) as f:
        data = json.load(f)
    
    X, y = [], []
    gesture_to_idx = {}
    
    for sample in data:
        # Combine all channels
        sensors = np.column_stack([
            sample['channels']['flex_thumb'],
            sample['channels']['flex_index'],
            sample['channels']['flex_middle'],
            sample['channels']['flex_ring'],
            sample['channels']['flex_pinky'],
            sample['channels']['accel_x'],
            sample['channels']['accel_y'],
            sample['channels']['accel_z'],
            sample['channels']['gyro_x'],
            sample['channels']['gyro_y'],
            sample['channels']['gyro_z']
        ])
        
        X.append(sensors)
        
        gid = sample['gesture_id']
        if gid not in gesture_to_idx:
            gesture_to_idx[gid] = len(gesture_to_idx)
        y.append(gesture_to_idx[gid])
    
    return X, y, gesture_to_idx

# Train
model = SignLanguageClassifier()
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=0.001)

X_train, y_train, label_map = load_data('data/bdsl/BdSL-Sensor-Glove/train/data_train.json')

print(f"Training on {len(X_train)} samples, {len(label_map)} classes")
print("Model ready for training!")

4️⃣ I want to visualize sign language data

import matplotlib.pyplot as plt
import json

with open('data/bdsl/BdSL-Sensor-Glove/train/data_train.json') as f:
    data = json.load(f)

sample = data[0]

# Plot flex sensors
plt.figure(figsize=(12, 6))
for name in ['thumb', 'index', 'middle', 'ring', 'pinky']:
    plt.plot(sample['channels'][f'flex_{name}'], label=name.title())

plt.xlabel('Time (samples)')
plt.ylabel('Flex Value')
plt.title(f"Gesture: {sample['gesture_id']}")
plt.legend()
plt.savefig('gesture_plot.png')
plt.show()

5️⃣ I want to convert data to standard formats

import json
import pandas as pd

# Load data
with open('data/bdsl/BdSL-Sensor-Glove/train/data_train.json') as f:
    data = json.load(f)

# Convert to CSV
rows = []
for sample in data:
    row = {
        'sample_id': sample['sample_id'],
        'gesture_id': sample['gesture_id'],
        'duration_ms': sample['duration_ms'],
        'participant_id': sample['participant_id']
    }
    # Add mean values
    for ch in ['thumb', 'index', 'middle', 'ring', 'pinky']:
        row[f'flex_{ch}_mean'] = sum(sample['channels'][f'flex_{ch}']) / len(sample['channels'][f'flex_{ch}'])
    rows.append(row)

df = pd.DataFrame(rows)
df.to_csv('processed_data.csv', index=False)
print(f"Saved {len(df)} rows to CSV")

📚 Learn More

Topic Link
All Datasets DATASETS.md
Contributing docs/CONTRIBUTING.md
Data Loader scripts/data_loader.py
Visualizer tools/visualize.py
Dataset Language Download
ASL-MNIST American kaggle datasets download -d datamunge/sign-language-mnist
ASL Alphabet American kaggle datasets download -d grassknoted/asl-alphabet
ArSL2018 Arabic kaggle datasets download -d ahmedkhan123/arabic-sign-language
BdSL47 Bangla wget https://zenodo.org/record/7067906/files/BdSL47.zip
ISL Alphabet Indian GitHub

💡 Tips

  1. Start with image datasets - Easier to work with
  2. Use MediaPipe for hand pose extraction
  3. Check licenses before using in commercial projects
  4. Cite original creators in your work
  5. Join our community for help and discussions

❓ Need Help?

Happy building! 🚀

This site is open source. Improve this page.