EVCPP:Example-Driven Virtual Camera Pose Prediction for Cloud Performing Arts Scenes

Studying intelligent virtual shooting in cloud performing arts scenes is of great significance to the sustainable development of the cloud performing arts industry. Difficulties in summarizing the language of shots and the need to consider the attributes of cameras and actor movements are key concerns in stage filming. We propose EVCPP: Example-driven Virtual Camera Pose Prediction for cloud performing arts that uses existing shooting videos to guide the shooting of virtual scenes. By using the camera behavior information from reference videos as external guidance weights, along with the camera intrinsic parameters and actor state information as covariates, we combine them with the historical pose of the camera in the virtual scene to predict the future camera pose. Meanwhile, we propose a combined loss function for our task. Our method has achieved promising results in virtual 3D cloud performing arts scenes.