"From Pixels to Patterns — a deep, code-first journey into Convolutional Neural Networks for real-world computer vision."

Live Site: https://mcklay.github.io/cnn-companion-book/
Author: Clay Mark Sarte
Built with MkDocs Material | Powered by PyTorch & TensorFlow

This repository contains a structured, book-style deep learning guide focused entirely on Convolutional Neural Networks (CNNs) for real-world computer vision. It is written in a modular, educational format tailored for developers, ML engineers, students, and builders who want to master CNNs through hands-on coding using both PyTorch and TensorFlow.

Whether you're training a CNN from scratch or deploying a fine-tuned model in production, this companion book bridges core theory, practical implementation, and deployment-readiness—all in one place.

• Part I – Foundations of Image Tensors and Preprocessing
  How CNNs interpret images, data formats ([H, W, C] vs [C, H, W]), RGB → Tensor pipelines, and input shape walkthroughs.

• Part II – Preprocessing and Input Pipelines
  Resize, Normalize, Augment, tf.image vs torchvision.transforms, image folders, batching, and preprocessing alignment with pretrained models.

• Part III – CNN Architectures and Concepts
  Conv2D, pooling, padding, strides, ReLU, BatchNorm, residual blocks, forward passes, and writing your own CNN models.

• Part IV – Training and Fine-Tuning
  Training loops, optimizers, freezing layers, adapting pretrained CNNs, dropout, early stopping, learning rate schedules, and generalization strategies.

• Part V – Inference, Evaluation, and Visual Debugging
  Train vs Eval mode, intermediate feature map visualizations, model evaluation strategies, and interpretability.

• Part VI – Deployment-Ready Insights
  Inference pipelines, input consistency, test-time augmentation, and a practical CNN model debugging checklist.

• Jupyter Notebooks + Markdown Chapters
• PyTorch & TensorFlow (side-by-side)
• Rich visualizations and debugging walkthroughs
• Optional Hugging Face deployment (coming soon)

# Clone the repo
git clone https://github.com/McKlay/cnn-companion-book.git
cd cnn-companion-book

# (Optional) Create a virtual environment and install dependencies
pip install -r requirements.txt

Most chapters are available as standalone Jupyter notebooks and markdown files. You can read them directly in the repo or render them using tools like Jupyter Lab, Colab, or MkDocs (if converting to a site).

Work in Progress This companion book is being built and refined chapter by chapter. Contributions, ideas, and feedback are welcome!

MIT License © Clay Mark Sarte Free to learn, fork, and remix with attribution.