Real-time Object Detection with FPGA Using CenterNet

the paper proposes methodology for transferring architecture of modern neural network CenterNet to FPGA. CenterNet is a OneStage object detector that is used to detect and locate objects in images. Although this neural network has simple decoder, it shows good performance in terms of accuracy. Very high operation speed of the neural network hardware is achieved due to the choice of suitable encoder and efficient hardware implementation of both the decoder and the last layer with object filtering, and also transition to fixed-point arithmetic. At this, quality of the obtained predictions remains high.