Retraining existing CNN models

Training a new image recognition model from scratch requires a lot of time and computational power. If we can take a pre-trained network and retrain it with our images, it may save us computational time. For this recipe, we will show how to use a pre-trained TensorFlow image recognition model and fine-tune it to work on a different set of images.

We will illustrate how to use transfer learning from a pre-trained network for CIFAR-10. The idea is to reuse the weights and structure of the prior model from the convolutional layers and retrain the fully connected layers at the top of the network. This method is called fine-tuning.

Getting ready

The CNN network we are going to employ uses a very popular architecture called Inception. The Inception CNN model was created by Google and has performed very well on many image recognition benchmarks. For details, see the paper referenced in the second bullet point of the See also section...