Unsupervised-Image-Clustering-Classification-with-Deep-Learning

This GitHub repository is dedicated to a deep learning project focused on classifying the FashionMNIST dataset using unsupervised Learning.

Description :-

The "FashionMNIST" dataset comprises 60,000 grayscale images of clothing items. The mystery labels provide an intriguing challenge to classify these items and the aim of this project is to decode their meaning through deep learning and data analysis.

Project Overview:-

Default Network

• Implemented and trained a default CNN architecture for classification.
• Analyzed runtime performance, hyper-parameters, and design choices.
• Visualized training loss and classification accuracy over training epochs.

Custom CNN Architecture

• Created a custom CNN architecture to classify the dataset.
• Documented the model's architecture, design choices, and training insights.
• Visualized training loss and classification accuracy for the custom model.

Results Analysis

• Described the runtime performance, hyper-parameter comparisons, and model designs.
• Produced plots illustrating training loss and classification accuracy for both models.

Using Your Own Encoding

• Visualized the encoding using PCA, LDA and applied clustering algorithms (DBSCAN and K-means). Used t-SNE for visualization.
• Make educated guesses about the mystery labels based on clustering results.

Software Stack:

The analysis and classification tasks are carried out using Python. Key libraries used include Pandas for data manipulation, Seaborn for data visualization, Scikit-learn, Keras ,and tensorflow for CNN, clustering and Encoding.

Data Source:

The FashionMNIST dataset used in this project can be accessed from https://github.com/zalandoresearch/fashion-mnist