Hand Sign Classification — A to Y

A multi-class CNN image classifier built from scratch in PyTorch that recognizes 24 American Sign Language (ASL) hand sign alphabets (A–Y, excluding J which requires motion).

---

Demo

Live inference runs directly from your webcam using OpenCV.

---

Model Weights

⚠️ The .pth file (~100MB) is not tracked in this repo. Download it from Releases.

⬇ Download model weights (v1.0)

---

Project Structure

Hand_Sign-Classification/
│
├── Handsign_project.ipynb   # Full pipeline: training, evaluation, live inference
└── README.md

---

Pipeline Overview

Raw Images (24 classes)
        ↓
Preprocessing (Resize, Normalize)
        ↓
CNN Model (Conv → Pool → FC layers)
        ↓
Softmax Output (A–Y)
        ↓
Live Webcam Inference (OpenCV)

---

Classes

24 static hand signs — A through Y, skipping J (requires motion).

---

Setup & Usage

1. Clone the repo

git clone https://github.com/Deep-ii/Hand_Sign-Classification.git
cd Hand_Sign-Classification

2. Install dependencies

pip install torch torchvision opencv-python

3. Download the model

Download handsign_model.pth from Releases and place it in the project root.

4. Run inference

Open Handsign_project.ipynb and run the Live Inference section. It will open your webcam and start predicting in real time.

---

Tech Stack

Component	Tool
Framework	PyTorch
Architecture	Custom CNN
Live Inference	OpenCV
Notebook	Jupyter

---

What I Learned

• Designing a CNN architecture from scratch (Conv, Pool, FC layers)
• Debugging shape and device mismatches in PyTorch
• Saving and loading model weights with torch.save / torch.load
• Running live webcam inference with OpenCV

---

Author

Deep — GitHub