Analysis of Scale Sensitivity of Ship Detection in an Anchor-Free Deep Learning Framework

Ship detection is an important task in sea surveillance. In the past decade, deep learning-based methods have been proposed for ship from images and videos. Convolutional features are observed to be very effective representing objects. However, scales of convolution often lead different capacities feature representation. It unclear how scale influences performance learning detection. To this end, paper studies sensitivity anchor-free framework. Specifically, we employ classical CenterNet as base analyze influence size, depth, fusion strategy on multi-scale target Experiments show that, small targets, obtained top-down path can improve more significantly than that bottom-up fusion; contrary, achieves better larger targets.