Horizontal-to-Vertical Video Conversion

At this blooming age of social media and mobile platform, mass consumers are migrating from horizontal video to vertical contents delivered on hand-held devices. Accordingly, revitalizing the exposure becomes vital urgent, which is hereby tackled by our automated horizontal-to-vertical (abbreviated as H2V) conversion framework. Essentially, {\it \textbf{H2V}} framework performs subject-preserving cropping instantiated in proposed Rank-SS module. incorporates object detection discover candidate subjects, we select primary subject-to-preserve leveraging location, appearance, salient cues a convolutional neural network. In addition converting videos vertically around selected subject, automatic shot multi-object tracking integrated into accommodate long complex videos. To develop systems, collect an \textbf{H2V-142K}} dataset containing 125 (132K frames) 9,500 cover images annotated with subject bounding boxes. On public datasets, method demonstrates promising results selection comparing related solutions. Furthermore, industrially deployed hosting millions daily active users exhibits favorable H2V performance. By making well approach publicly available, wish pave way for more research. Our collected H2V-142K available atH2V-142K website.