Deep learning for image classification

This workshop on deep learning techniques for image classification explores the basics of convolutional neural networks (CNNs), including the convolution operation, nonlinear activation functions, and downsampling via max-pooling. Also, it will be studied how transfer learning is used to solve new classification tasks from a pre-trained model, and how to use Grad-CAM to improve model explainability.

Participants will also perform practical Python exercises on Google Colab, featuring implementations of a basic CNN from scratch, named LeNet-5, used for handwritten character recognition. Also, this trained model will be transferred to solve another object recognition task from the Fashion-MNIST dataset. Finally, examples of image classification by the ResNet-50 model will focus on Grad-CAM explainability.

Workshop slides

• PDF: Deep learning for image classification

Google Colab notebooks

• Convolutional layer elements
• Build and train the LeNet-5 model
• Test the LeNet-5 on handwritten numbers
• Transfer learning based on LeNet-5
• Grad-CAM visualization

Bibliography

Contact information

• Email: wgomez at cinvestav.mx