FoveaBox is an accurate, flexible and completely anchor-free object detection system for object detection framework. Different from previous anchor-based methods, FoveaBox directly learns the object existing possibility and the bounding box coordinates without anchor reference. This is achieved by: (a) predicting category-sensitive semantic maps for the object existing possibility, and (b) producing category-agnostic bounding box for each position that potentially contains an object. FoveaBox avoids all computation and hyper-parameters related to anchor boxes, which are often sensitive to the final detection performance.

Object detection single-model results v.s. state-of-the-arts on COCO test-dev. We show results for FoveaBox models with 800 input scale. FoveaBox-align indicates utilizing feature alignment Qualitative comparison of RetinaNet (left) and FoveaBox (right). Our model shows improvements in classes with large aspect ratios. Better viewed in color.