unsupervised-learning-2023

Project Unsupervised Learning 2023 concentrates on solving a translation of manuscripts between image alphabets (EMNIST and KMNIST) utilizing unsupervised learning methods such as autoencoders (CAE and VAE), dimensionality reduction (PCA, UMAP) and clustering (K-Means, GMM). To simulate a real world problem, noise was added on generated manuscripts (salt and pepper, lines, rotation and scaling of letters).

App

GUI was created in streamlit:

Input:

Output:

How to run:

1. Download datasets and text download.py
2. (optional) Review spreadsheets of random samples to choose seeds generate_spreadsheet.py
3. (even more optional) Preview random (seeded) samples preview.py
4. Construct mapping from random (seeded) samples generate_mapping.py
5. Construct pages in both alphabets from mapping generate_dataset.py