AI-based Stereoview to Multiview Generation by Using Deformable Convolution

Three dimension (3D) movies are the main trend in the film industry. In the current stereoview, the audiences require wearing 3D glasses to perceive 3D visualization. The 3D movies with stereoview with only left and right views, which cannot be directly displayed in the naked-eyes 3D displays. To directly support naked-eyes 3D displays, which require multiple views, we propose a deep learning based stereo to multiview conversion system by using the deformable convolution to synthesize additional virtual views. For immersive 3D multimedia services, we hope we can improve the quality of user 3D experiences without wearing 3D glasses without the needs of depth estimation and depth image based rendering functions.