Abstract: Current prevailing Video Object Segmentation (VOS) methods follow the pipeline of extraction-then-matching, which first extracts features on current and reference frames independently, and ...

Abstract: Convolutional neural networks (CNNs) are frequently used to analyze remote sensing images and achieve impressive progress. Limited by the receptive field size of CNNs, small objects tended ...