Sign Language Recognition with CNN

A PyTorch-based convolutional neural network for real-time American Sign Language (ASL) recognition using webcam input.

Overview

This project implements a CNN model that recognizes 29 different ASL signs: 26 letters (A-Z) plus "space", "delete", and "nothing" gestures. The model processes 64x64 RGB images and provides real-time predictions through webcam feed.

Requirements

Install dependencies using:

install -r requirements.txt

Model Architecture

The CNN consists of:

• 3 convolutional layers (32, 64, 128 filters)
• MaxPooling after each conv layer
• 2 fully connected layers (512, 29 outputs)
• ReLU activations
• Input: 64x64 RGB images
• Output: 29 classes

Usage

1. Data Preparation

If your test images aren't organized by class, run:

cd src/utils
python organize_test.py

Check for corrupted images:

python detect_bad_files.py

2. Training

Train the model:

cd src
python train.py

Training parameters:

• Batch size: 64
• Learning rate: 0.001
• Optimizer: Adam
• Loss: CrossEntropyLoss
• Epochs: 10
• Image size: 64x64

3. Testing

Evaluate model performance:

python test.py

This outputs the accuracy percentage on the test set.

4. Real-time Prediction

Run webcam prediction:

python predict_webcam.py

Controls:

• Place your hand in the blue rectangle (100,100 to 300,300)
• Press 'q' to quit
• Close the window to exit

Model Performance

The model saves checkpoints after each epoch as models/model_epoch_X.pth. The final model (model_epoch_10.pth) is used for inference.

Data Preprocessing

Images are preprocessed with:

• Resize to 64x64 pixels
• Convert to tensor
• Normalize with mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5]

Hardware Requirements

• GPU: CUDA-compatible GPU recommended for training
• CPU: Falls back to CPU if CUDA unavailable
• Camera: Webcam for real-time prediction