A method and computing device are provided for performing virtual camera functions during playback of media content. In the method, the computing device selects a region of interest (RoI) during a playback of media content, and represents the RoI as an ensemble of disparate hypotheses. Then the computing device identifies the RoI in a frame sequence sampled according to the hypotheses, performs a virtual camera action on the identified RoI, and plays the RoI with a virtual camera.