Fast Single Shot Instance Segmentation

摘要

In this work, we propose fast single shot instance segmentation framework (FSSI), which aims at jointly object detection, segmenting and distinguishing every individual instance (instance segmentation) in a flexible and fast way. In the pipeline of FSSI, the instance segmentation task is divided into three parallel sub-tasks: object detection, semantic segmentation, and direction prediction. The instance segmentation result is then generated from these three sub-tasks' results by the post-process in parallel. In order to accelerate the process, the SSD-like detection structure and two-path architecture which can generate more accurate segmentation prediction without heavy calculation burden are adopted. Our experiments on the PASCAL VOC and the MSCOCO datasets demonstrate the benefits of our approach, which accelerate the instance segmentation process with competitive result, compared to MaskRCNN. Code is public available (https://github.com/ lzx1413/FSSI).