YOLOMask: Real-time Instance Segmentation With Integrating YOLOv5 and OrienMask

Abstract

In this paper, we propose a real-time framework for instance segmentation, which we call YOLOMask and which builds on the real-time project OrienMask. In YOLOMask, we integrate the YOLOv5 object detection framework with the OrienMask instance segmentation framework to form a new real-time instance segmentation framework and we integrate CBAM into YOLOMask, which can help the network to find regions of interest in images with large area coverage. Using this method, our YOLOMask can achieve 47.8/44.3 msak/box AP on Pascal 2012 SBD dataset evaluated at 84.3 fps with a V100 GPU. Compared to OrienMask, YOLOMask improves box AP by about 5.8% and mask AP by 4.5%, which is encouraging and competitive. Given its simplicity and efficiency, we hope that our YOLOMask can serve as a simple but strong baseline for a variety of instance-wise prediction tasks.