How to train an image classifier using PyTorch

Building an image classifier that can recognise cities

Rogier van der Geer

Deep Learning Fun and Humor Image Processing Machine-Learning Scientific Libraries (Numpy/Pandas/SciKit/...)

Neural networks are everywhere nowadays. But while it seems everyone is using them, training your first neural network can be quite a hurdle to overcome.

In this talk I will take you by the hand, and following an example image classifier I trained, I will take you through the steps of making an image classifier in PyTorch. I will show you code snippets and explain the more intricate parts. Also, I will tell you about my experience, and about what mistakes to prevent. After this all you need to start training your first classifier is a data set!

Of course I will provide a link to the full codebase at the end. The talk will focus on the practical aspect of training a neural network, and will only touch the theoretical side very briefly. Some basic prior knowledge of neural networks is beneficial, but not required, to follow this talk.

Type: Talk (45 mins); Python level: Intermediate; Domain level: Beginner

Rogier van der Geer


Before joining GoDataDriven, Rogier obtained a PhD in particle physics. Rogier gained hands-on experience with handling enormous quantities of data and processing, or 'charming', them into a manageable format before performing complicated analyses. After his PhD he exchanged physical science for data science at GoDataDriven, where he is now putting his skills to use on more business-driven problems. He likes applying data science to anything; be it his daily commute, improving his photography skills or the contents of his lunch box.