Skip to content
Scan a barcode
Scan
Paperback In-depth CNN - Part 1: Foundations, Handworked Calculations, and Image Classification Book

ISBN: 9334591048

ISBN13: 9789334591040

In-depth CNN - Part 1: Foundations, Handworked Calculations, and Image Classification

How does a convolutional neural network turn pixels into a prediction? What changes when it learns? And how can we tell whether its results are useful beyond the training data?

In-depth CNN - Part 1 takes you from the numerical representation of an image to working recognition systems. Through explicit calculations, worked examples and executable Python code, it connects the mathematics of convolutional neural networks with the decisions involved in building, training and evaluating them.

The journey begins with pixels, colour channels, tensors and normalisation. Small numerical examples make convolution, stride, padding and pooling visible. You follow values through a network, calculate prediction errors and examine how backpropagation updates its parameters. NumPy implementations expose the operations directly, while PyTorch connects those calculations to practical model development.

Each new task builds on that foundation:
- Understand images as numbers: inspect pixel values, tensor shapes, data types and dataset partitions.
- Calculate the core operations: work through convolution, pooling, receptive fields and parameter counts.
- Follow the learning process: connect logits, sigmoid outputs, cross-entropy, derivatives and gradient descent.
- Recognise handwritten digits: begin with binary MNIST, then extend the model to all ten classes.
- Investigate training choices: explore optimisation, initialisation, regularisation, learning-rate schedules and validation-based selection.
- Recognise clothing: train on Fashion-MNIST, inspect learned representations and use feature vectors for visual matching.
- Build an application workflow: connect retrieved clothing images to a demonstration inventory, customer requirements and a warehouse pick list.
- Apply transfer learning: adapt a pretrained CNN-based text recogniser to real US number plate images, evaluate it and save a reusable reader.

The book goes beyond reporting accuracy. Confusion matrices, per-class measurements, learning curves and individual predictions help explain where a model succeeds and where it struggles. An interactive digit-drawing interface reveals how changes in input preparation can affect recognition. The clothing application shows how model predictions, visual similarity and business rules contribute different information to a complete workflow.

The final experiment broadens the perspective from training networks from the beginning to adapting an existing recogniser. Using real images from the OpenALPR US benchmark, you examine annotations, keep related examples together when splitting data, fine-tune in stages and compare complete registration accuracy with character error rate.

Throughout, the emphasis is on understanding the experiment: what the data contains, what the model predicts and what the measurements establish. The boundaries between benchmark results and application behaviour help you interpret progress without overstating it.

Written for students, software developers, data scientists, and curious practitioners, this book suits readers who want to understand both the calculations and the code. Familiarity with Python, elementary algebra, matrices and basic differentiation will help. Previous experience building CNNs is not essential.

Read the calculations, run the programs, inspect the outputs and test your own changes. By the end, you will have a practical foundation for explaining a network's computations, adapting its design and judging its predictions through evidence.

Learn how CNNs work-and develop the reasoning to use them well.

Recommended

Format: Paperback

Condition: New

$89.99
Save $10.01!
List Price $100.00
Ships within 2-3 days
Save to List

Customer Reviews

0 rating
Copyright © 2026 Thriftbooks.com Terms of Use | Privacy Policy | Do Not Sell/Share My Personal Information | Cookie Notice | Cookie Preferences | Accessibility Statement
ThriftBooks ® and the ThriftBooks ® logo are registered trademarks of Thrift Books Global, LLC
GoDaddy Verified and Secured