This paper introduces SiamMask, a novel approach that combines visual object tracking and semi-supervised video object segmentation in real-time. SiamMask enhances the offline training of fully-convolutional Siamese networks by augmenting their loss function with a binary segmentation task. This method, which relies solely on a single bounding box initialization, produces class-agnostic object segmentation masks and rotated bounding boxes at 55 frames per second. Despite its simplicity and speed, SiamMask achieves state-of-the-art performance on the VOT-2018 benchmark for real-time object tracking and competitive results on the DAVIS-2016 and DAVIS-2017 benchmarks for semi-supervised video object segmentation. The project website is <http://www.robots.ox.ac.uk/~qwang/SiamMask>.This paper introduces SiamMask, a novel approach that combines visual object tracking and semi-supervised video object segmentation in real-time. SiamMask enhances the offline training of fully-convolutional Siamese networks by augmenting their loss function with a binary segmentation task. This method, which relies solely on a single bounding box initialization, produces class-agnostic object segmentation masks and rotated bounding boxes at 55 frames per second. Despite its simplicity and speed, SiamMask achieves state-of-the-art performance on the VOT-2018 benchmark for real-time object tracking and competitive results on the DAVIS-2016 and DAVIS-2017 benchmarks for semi-supervised video object segmentation. The project website is <http://www.robots.ox.ac.uk/~qwang/SiamMask>.