In this section, we'll discuss one of the most popular detection algorithms, called YOLO. The name is an acronym for the popular motto you only live once, which reflects the one-stage nature of the algorithm. The authors have released three versions with incremental improvements of the algorithm. We'll only discuss the latest, v3 ( for more details, see YOLOv3: An Incremental Improvement, https://arxiv.org/abs/1804.02767).
The algorithm starts with the so-called backbone network called Darknet-53 (after the number of convolutional layers). It is trained to classify the ImageNet dataset, just as the networks in Chapter 3, Advanced Convolutional Networks. It is fully convolutional (no pooling layers) and uses residual ...
