Image Modeling with Keras
DataCamp
A four-hour course on building image classifiers with Keras, taught by Ariel Rokem, a senior data scientist at the University of Washington. It opens with images as data: how a picture becomes something a network can consume, one-hot encoding for the labels, and building, compiling, and fitting a first neural network on a clothing dataset — evaluated properly with cross-validation from the outset. The second chapter introduces the convolution itself, what a kernel does to an image, and how padding and strides change the output. The third stacks convolutional layers into deeper networks and deals with what that costs: counting the parameters as a network grows, and adding pooling layers to control them. The final chapter is about improving what you have built — plotting learning curves, reusing stored weights, adding dropout and batch normalization, and extracting a trained kernel to visualize what the network actually responds to. Four chapters, 13 videos, and 45 exercises.
What you'll learn
- Treating images as data, and one-hot encoding their labels
- Building, compiling, and fitting a neural network in Keras, and evaluating it with cross-validation
- What a convolution kernel does to an image, and how padding and strides change the result
- Calculating the output size of a convolutional layer
- Stacking convolutional layers into deeper networks, and counting the parameters that adds
- Pooling operations, written by hand and then with Keras layers
- Plotting learning curves, and reducing overfitting with dropout and batch normalization
- Extracting a kernel from a trained network and visualizing what it responds to
Frequently asked questions about Image Modeling with Keras
Who is Image Modeling with Keras for?
Python developers with some Keras experience who want to build, inspect, and improve convolutional networks for image classification.
Is Image Modeling with Keras free?
No — Image Modeling with Keras is a paid course.
What are the prerequisites for Image Modeling with Keras?
DataCamp's Introduction to Deep Learning with Keras, plus comfort with Python.
Do you need to code for Image Modeling with Keras?
Yes — Image Modeling with Keras involves hands-on coding.
Does Image Modeling with Keras offer a certificate?
Yes. DataCamp Statement of Accomplishment on completion (requires DataCamp Premium).
Why we suggest this course
A methodical build-up rather than a tour: images become data, then a plain network, then convolutions, then depth — with the cost of each step made explicit, including counting the parameters a deeper network adds and using pooling to keep them in check. The closing chapter is unusually focused on diagnosis: plotting learning curves, applying dropout and batch normalization to a network that is memorizing rather than generalizing, and pulling a kernel out of a trained model to see what it has learned to look for. One thing to know: it follows DataCamp's Introduction to Deep Learning with Keras, so it teaches image modeling rather than the framework. You can sample the opening chapter before subscribing; the full course and its Statement of Accomplishment are part of DataCamp Premium.