How to Train an Image Classifier Using PyTorch

Explore the process of training an image classifier using PyTorch in this 42-minute EuroPython Conference talk. Learn the practical steps involved in creating a neural network for image classification, including code snippets and explanations of complex concepts. Discover how to utilize pretrained networks, implement data transforms and loaders, optimize learning rates, and evaluate model performance. Gain insights into common pitfalls to avoid and see real-world examples of city recognition results. Follow along as the speaker demonstrates the entire workflow, from dataset preparation to final model evaluation, providing valuable hands-on experience for both beginners and those with basic neural network knowledge. Access the full codebase and apply these techniques to your own image classification projects.

Intro
Deep Neural Networks
Why PyTorch
Neural Network
Pretrained Network
Internals
Updating the model
How to train a model
Evaluation mode
Transforms
Loaders
Learning Rate
Learning Rate Function
Learning Rate Scheduler
Data
Dataset
Viewing the images
Interesting tags
Other tags
Common cities
Training time
Model
Chicago
Chicago Results
Another Plan
Get what I did
Results
Train test split
More results
Chicago recognition
Philadelphia recognition
London recognition
Errors
Final remarks
Gitlab
Questions