Summary

This chapter has been all about CNN architectures. First, we briefly discussed the history and evolution of CNNs. We then explored in detail one of the earliest CNN models – LeNet. Using PyTorch, we built the model from scratch and trained and tested it on an image classification dataset. We then explored LeNet's successor – AlexNet. Instead of building it from scratch, we used PyTorch's pre-trained model repository to load a pre-trained AlexNet model. We then fine-tuned the loaded model on a different dataset and evaluated its performance.

Next, we looked at the VGG model, which is a deeper and a more advanced successor to AlexNet. We loaded a pre-trained VGG model using PyTorch and used it to make predictions on a different image classification dataset. We then successively discussed the GoogLeNet and Inception v3 models that are composed of several inception modules. Using PyTorch, we wrote the implementation of an inception module and the whole...