Sign Language Detection

A simple sign-language (ISL) image classification project that trains and evaluates a model to recognize alphanumeric hand signs.

Project structure

• data/ - dataset organized into train/, valid/, and test/ with subfolders for each class (0-9, A-Z).
• model.py - model architecture definition.
• training.py - training script.
• result.py - evaluation / inference utilities and reporting.
• text.py - helper for text/labels processing.
• runs_simple/ - training outputs and model weights (e.g. best.pt, epoch_003.pt).
• requirements.txt - Python dependencies.
• label_map.json - mapping of class indices to labels.
• metrics.json, results/ - evaluation reports and visualizations.

Quick start

1. Create and activate a Python virtual environment (optional but recommended):

   source .venv/bin/activate

2. Install dependencies:

   pip install -r requirements.txt

3. Train the model:

   python training.py

4. Evaluate or run inference:

   python result.py

5. Run the model:

   python text.py spell --ckpt runs_simple/best.pt

6. To hear the audio(Linux):

   ffplay -nodisp -autoexit 'File name'

Notes

• Pretrained weights and experiment outputs are stored in runs_simple/. Use best.pt for inference when available.
• The dataset must follow the folder-per-class convention used in data/ for training and validation.
• Inspect label_map.json to map predicted indices to human-readable labels.

Tips

• To add more data, place images in the appropriate class subfolder under data/train/ (and similarly for data/valid/ and data/test/).
• Use small experiments first (fewer epochs, smaller batch) to check end-to-end pipeline before full training.

License

This repository does not include a license file. Add a LICENSE if you plan to publish or share the code.

Maintainer

Repository owner: Harsh1260

Acknowledgements

This project is a private/academic implementation for ISL classification using image datasets. See source files for implementation details.