Umberto Michelucci - Advanced Applied Deep Learning: Convolutional Neural Networks and Object Detection

Any source code or other supplementary material referenced by the author in this book is available to readers on GitHub via the books product page, located at www.apress.com/978-1-4842-4975-8 . For more detailed information, please visit http://www.apress.com/source-code .

This is the second book I have written, and it covers advanced topics in deep learning. It will require some knowledge to be understood. Its not for beginners. If you are one, I suggest you check out my first book published by Apress ( Applied Deep Learning: A Case-Based Approach, ISBN 978-1-4842-3790-8). To understand this book, you should have some intermediate to advanced experience in Python and some intermediate to advanced deep learning experience (and experience with neural networks in general). This book assumes you know things like regularization, hyperparameter tuning, mini-batch gradient descent, which optimizers are more efficient (does the name Adam tell you something?), and so on. I also use heavily Keras (from TensorFlow), so I suggest you get some experience with that too. It will help you work through the examples of the book.

I tried to tackle advanced topics, like transfer learning or multi-loss function networks , with a practical approach. That means I explain the concepts and then show you how to implement those things in Keras. I invested quite some time in preparing code for you and its available on the GitHub repository for the book, so get the code and use it while youre reading the book. The code for the advanced topics we deal with, is too long to discuss completely, so I only dissect the most important parts. In GitHub, you have all of it.

This book touches on several research fields, but was not written for very experienced researchers. It has been written for practitioners who want to start doing research; therefore, its goal is to bridge the gap between the beginner and the researcher. Very advanced topics, like in object detection, are not explained in much technical detail, since otherwise the book would turn into a collection of research papers.

Keep in mind that many of the things I describe in the book, like the YOLO object detection algorithm, are only a few years old. For advanced topics, the only way to understand an algorithm is to read the original paper. You should get used to doing that without any book. Here, I try to give you the tools and explain the language that you need to read the research papers. From there, you are on your own.

If you want to proceed further in your deep learning adventure, you should get used to reading research papers. They are not easy to read and will require time. But this book should give you many tools and tips for a head start. Reading this book and understanding all of it will put you at the start of your research career. From there on, start reading research papers. Try to repeat what they did if possible (mostly is not, given the infrastructure needed for deep learning , but you can always try). Understanding algorithms and research papers will give you enough knowledge to evaluate libraries and see what others have done, if you are searching for a way to use a specific algorithm in your projects.

I hope you enjoy the book, that you learn something from it, and that it helps you, but more important than anythingI hope you have fun!

Umberto Michelucci, Zrich, 3 rd of July 2019