This repository / notebook is a structured, hands-on exploration of Convolutional Neural Networks (CNNs), focusing on understanding how computer vision models actually learn, not just how to run them.
The content follows a progressive learning path, where each section builds directly on the previous one.
The exercises include:
-
Convolutional Classifier
Building a first CNN model from scratch using Keras. -
Convolution & ReLU
Understanding how convolutional layers extract features and why non-linearity matters. -
Max Pooling
Learning how spatial dimensionality is reduced while preserving important features. -
Sliding Window (Stride & Padding)
Exploring how stride and padding affect feature maps and receptive fields. -
Custom Convnets
Designing custom CNN architectures and experimenting with depth and filters. -
Data Augmentation
Improving model generalization by generating diverse training samples.
The main goal of this project is conceptual clarity.
Instead of treating CNNs as black boxes, this notebook emphasizes:
- How architectural choices influence feature extraction
- How spatial information flows through a network
- Why certain design patterns are commonly used in computer vision models
Accuracy matters — but understanding matters more.
- Python
- TensorFlow / Keras
- Convolutional Neural Networks (CNN)
- Image Data Pipelines & Augmentation
- Learners starting their journey in Computer Vision
- Anyone who wants an intuitive understanding of CNNs
- Practitioners revisiting fundamentals before building advanced models
🎯 MY GOAL:Aiming to lead the evolution of Automotive AI Systems through innovation and data-driven solutions.
This project is intended for learning and experimentation.
Code is written for readability and exploration rather than heavy optimization.
"Building the future of artifitial intelligence, one line of code at a time."