Deep Learning & CNN Image Classification Project

Overview

This repository documents my Deep Learning learning journey (Day 46 – Day 60) where I built and trained multiple neural network models using TensorFlow / Keras.

The final outcome of this phase is a Convolutional Neural Network (CNN) Image Classification system trained on the Fashion-MNIST dataset to recognize clothing items.

This project demonstrates the complete deep learning workflow used in real-world AI systems, including:

• Neural network fundamentals
• Model training and evaluation
• Handling overfitting
• Saving and loading models
• Building CNN architectures
• Image classification
• Model improvement
• Prediction demos

---

Technologies Used

• Python
• TensorFlow
• Keras
• NumPy
• Matplotlib
• Jupyter Notebook
• Git & GitHub

---

Dataset

The models were trained using the Fashion-MNIST dataset.

Fashion-MNIST contains 70,000 grayscale images of clothing items across 10 classes.

Image Details:

• Image Size: 28 × 28 pixels
• Format: Grayscale

Clothing Categories

Label	Clothing Item
0	T-shirt/top
1	Trouser
2	Pullover
3	Dress
4	Coat
5	Sandal
6	Shirt
7	Sneaker
8	Bag
9	Ankle boot

---

Sample Dataset Images

MNIST Digit Samples

Single Digit Example

Fashion Dataset Sample

---

CNN Prediction Example

Example output from the CNN model:

The trained model predicts the clothing category based on the image.

---

Project Structure

Deep Learning & CNN Image Classification Project

Overview

This repository documents my Deep Learning learning journey (Day 46 – Day 60) where I built and trained multiple neural network models using TensorFlow / Keras.

The final outcome of this phase is a Convolutional Neural Network (CNN) Image Classification system trained on the Fashion-MNIST dataset to recognize clothing items.

This project demonstrates the complete deep learning workflow used in real-world AI systems, including:

• Neural network fundamentals
• Model training and evaluation
• Handling overfitting
• Saving and loading models
• Building CNN architectures
• Image classification
• Model improvement
• Prediction demos

---

Technologies Used

• Python
• TensorFlow
• Keras
• NumPy
• Matplotlib
• Jupyter Notebook
• Git & GitHub

---

Dataset

The models were trained using the Fashion-MNIST dataset.

Fashion-MNIST contains 70,000 grayscale images of clothing items across 10 classes.

Image Details:

• Image Size: 28 × 28 pixels
• Format: Grayscale

Clothing Categories

Label	Clothing Item
0	T-shirt/top
1	Trouser
2	Pullover
3	Dress
4	Coat
5	Sandal
6	Shirt
7	Sneaker
8	Bag
9	Ankle boot

---

Sample Dataset Images

MNIST Digit Samples

Single Digit Example

Fashion Dataset Sample

---

CNN Prediction Example

Example output from the CNN model:

The trained model predicts the clothing category based on the image.

---

Project Structure

DEEP-LEARNING-NLP/

models/ fashion_cnn.keras fashion_cnn_improved.keras model_day52.keras model_day54_cnn.keras

notebooks/ day46_tensorflow_setup.ipynb day47_neural_network_basics.ipynb day48_first_ann.ipynb day49_loss_optimizer.ipynb day50_overfitting.ipynb day51_model_evaluation.ipynb day52_save_load_models.ipynb day53_intro_cnn.ipynb day54_image_classification_basics.ipynb day55_fashion_dataset.ipynb day56_fashion_cnn.ipynb day57_fashion_predictions.ipynb day58_model_analysis.ipynb day59_cnn_improvement.ipynb day60_cnn_demo.ipynb

output/ cnn_prediction_example.png fashion_sample.png mnist_sample_digits.png mnist_single_digit_example.png

---

Model Development Journey

Day 46 – TensorFlow Setup

• Installed TensorFlow
• Verified deep learning environment

Day 47 – Neural Network Basics

• Understanding neurons
• Activation functions
• Forward propagation

Day 48 – First Artificial Neural Network

• Built first ANN using TensorFlow
• Basic classification model

Day 49 – Loss Functions & Optimizers

Implemented:

• Cross entropy loss
• Adam optimizer

Day 50 – Handling Overfitting

Applied techniques to improve model generalization:

• Dropout
• Early stopping

Day 51 – Model Evaluation

Model performance measured using:

• Accuracy
• Loss

Day 52 – Saving & Loading Models

Implemented model persistence:

• Saving models using .keras
• Loading trained models for inference

Day 53 – Introduction to CNN

Learned CNN fundamentals:

• Convolution layers
• Filters
• Feature extraction
• Pooling layers

Day 54 – Image Classification Basics

Prepared image data for CNN:

• Image normalization
• Reshaping data
• Preparing tensors for CNN input

Day 55 – Dataset Exploration

Explored Fashion-MNIST:

• Visualized clothing images
• Inspected labels
• Understood dataset structure

Day 56 – First CNN Model

Built first convolutional neural network using:

• Convolution layers
• MaxPooling
• Dense layers

Day 57 – CNN Predictions

Implemented prediction pipeline:

• Generated predictions
• Visualized model outputs

Day 58 – Model Analysis

Analyzed model performance and prediction results.

Day 59 – CNN Improvement

Improved CNN architecture by adding:

• Additional convolution layers
• Dropout regularization

Day 60 – Final CNN Demo

Created full inference pipeline:

1. Load trained model
2. Predict clothing category
3. Display predicted label
4. Compare predicted vs actual label

---

Key Learnings

This project helped build strong understanding of:

• Deep learning model pipelines
• CNN architecture design
• Image preprocessing
• Model training and evaluation
• Model versioning
• Prediction pipelines

---

Future Work

Next steps in the learning journey:

• Computer Vision with OpenCV
• Transfer Learning
• Object Detection
• Real-world AI applications

---

Author

Abhihail Jacob

Learning journey toward becoming a Python AI Engineer.