This project aims to classify images containing alphabets in sign languages using TensorFlow. The application features a user-friendly GUI built with Tkinter.
-
Original Dataset: The initial dataset used is available here.
- Train Accuracy: 95%
- Test Accuracy: 93%
- However, the model's performance in a live environment was suboptimal due to differences in image quality between the training and live data. Live images were notably of lower quality.
-
Custom Dataset: To address the bias in the original dataset, a custom dataset was generated using the dataset_generater_tool.py.
- Train Accuracy: 96%
- Test Accuracy: 93%
- Notably, this model exhibited significantly improved performance in live scenarios.
