Graph Transduction Learning of Object Proposals for Video Object Segmentation

We propose an unsupervised video object segmentation algorithm that detects recurring objects and learns cohort object proposals over space-time. Our core contribution is a graph transduction process that learns object proposals densely over space-time, exploiting both appearance models learned from rudimentary detections of sparse object-like regions, and their intrinsic structures. Our approach exploits the fact that rudimentary detections of recurring objects in video, despite appearance variation and sporadity of detection, collectively describe the primary object. By learning a holistic model given a small set of object-like regions, we propagate this prior knowledge of the recurring primary object to the rest of the video to generate a diverse set of object proposals in all frames, incorporating both spatial and temporal cues. This set of rich descriptions underpins a robust object segmentation method against the changes in appearance, shape and occlusion in natural videos.