This project focused on gesture recognition using Convolutional Neural Networks (CNNs). I collected and cleaned a dataset of American Sign Language (ASL) gestures (A-I), comprising a total of 27 images. From there, I developed a custom CNN from scratch using PyTorch, achieving a classification accuracy of 94% on the ASL gesture dataset. To further enhance the model's performance, I implemented transfer learning using the pre-trained AlexNet to extract features from the ASL gesture images, which allowed me to train a smaller neural network on these features and achieve improved classification accuracy. This project highlighted the effectiveness of CNNs and transfer learning in gesture recognition tasks, demonstrating significant advancements in classification accuracy and model efficiency.
To learn more about this project, please refer to the file below.