Coherence-aware context aggregator for fast video object segmentation

•We re-analyze the generation and utilization mechanism of temporal context in the semi-supervised VOS task and introduce the efficient tracklet data flow, based on which we propose a novel VOS model and achieve real-time running speed. Experimental results on three benchmark datasets show that our proposed model achieves a better trade-off between efficiency and accuracy.•We design a coherence-aware module to estimate the coherence of the predicted target and update the temporal context in a more robust way.•We devise a spatio-temporal context aggregation module to aggregate the spatial context and temporal context at each level of the decoder to alleviate the aliasing effect, and learn a robust and discriminative feature representation for the target object.