Classification challenges, such as ILSVRC, are great playgrounds for researchers, leading to the development of more advanced deep learning solutions. In their own way, each of the architectures we detailed in this chapter became instrumental in computer vision and are still applied to increasingly complex applications. As we will see in the following chapters, their technical contributions inspired other methods for a wide range of visual tasks.
Moreover, not only did we learn to reuse state-of-the-art solutions, but we also discovered how algorithms themselves can benefit from the knowledge acquired from previous tasks. With transfer learning, the performance of CNNs can be greatly improved for specific applications. This is especially true for tasks such as object detection, which will be the topic of our next chapter. Annotating datasets for object detection is more tedious than for image-level recognition, so methods usually have access to smaller training datasets. It is...
