Slides and code for my presentation - meet-up details here:
https://www.meetup.com/Cape-Town-deep-learning/events/240485642/
The notebook covers how to:
- Use a pre-trained model (VGG16) to predict imagenet classes for new images
- How to fine-tune a pre-trained model on another dataset to predict new clases
- How to get CNN activations to use as numerical representations of images
- Using numerical representations of images to find similar images for an input image
cd to the directory and run
jupyter notebook
keras, theano, pandas, numpy, etc
(conda distribution should include all pre-requisites, otherwise pip install any missing packages)
Install the Anaconda python distribution available here: https://www.continuum.io/Downloads
- Alex Conway - NumberBoost
- Jeremy Howard & Rachel Thomas @ Fast.ai (http://fast.ai)