The intent of this chapter was to give you a more hands-on perspective of building deep learning models to solve a real-world problem, and also to understand the effectiveness of transfer learning. We covered aspects of the need for transfer learning, especially when solving problems with the constraints of limited data. We built several CNN models from scratch and also saw the benefits of a proper image augmentation strategy. We also looked at how to leverage pretrained models for transfer learning and covered the various ways to use them, including as feature extractors, as well as fine-tuning. We saw the detailed architecture of the VGG-16 model and how to leverage the model as an efficient image feature extractor. Strategies pertaining to transfer learning, including feature extraction and fine-tuning, along with image augmentation, were used to build effective deep learning image classifiers.

Last but not least, we evaluated all our models on our test dataset and also gained some perspective of how convolutional neural networks visualize images internally when building feature maps. In subsequent chapters, we will look at more complex real-world case studies that require transfer learning. Stay tuned!