This repository contains a machine learning model capable of recognizing various sign language gestures based on the Sign Language MNIST dataset. The model utilizes a Convolutional Neural Network (CNN) architecture to extract relevant features from image data and classify them into corresponding sign language categories.
Sign Language MNIST: A dataset consisting of 27,455 grayscale images, each representing a single handwritten sign language character. The images are normalized to a size of 28x28 pixels.
- Training Data:
sign_mnist_train.csv - Validation Data:
sign_mnist_test.csv - Loss Function:
sparse_categorical_crossentropyis used to measure the difference between predicted and actual class labels. - Optimizer:
Adamoptimizer is employed to update model weights during training.
- Expand the dataset to include a wider range of sign language gestures and dialects.
- Explore techniques to improve the model's performance in challenging lighting conditions and with complex backgrounds.
- Investigate the use of generative adversarial networks (GANs) to augment the training dataset.
- Make a real-time sign detection model.