The idea of this chapter was to give you a taste of what CNNs are, what they are used for, and how to construct them in TensorFlow. However, it is useful to mention at this point that there are other types of convolution operations used nowadays for different purposes, and we will look at some of them in more detail in later chapters.

For now, we will just mention them by name and where they are used:

• Depthwise convolution: Used in MobileNets, they aim to make convolutions friendly for mobile platforms
• Dilated convolutions (Atrous convolution): They have an extra parameter called dilation rate that allows you to have a bigger field of view with the same computation cost (for instance, a 3x3 CONV could have the same field of view as a 5x5 CONV)
• Transposed convolutions (Deconvolutions): Used normally in CNN autoencoders and in semantic segmentation problems