Fig. 2From: Deep learning-based strategies for the detection and tracking of drones using several camerasProposed lightweight YOLOv3 detector. Different than the default architecture, we prefer to use 812 × 812 × 3 input size to better detect the small drones. The circles with R denotes the route layer and the U denotes the upsampling layer. Note that, the previous layer filters, where the upsampled feature matrix is routed is underlined. Also, note that the layers where the positional dimensions of the feature matrix is divided in to two by proper strided max pooling are shownBack to article page