Technology, Software, AlgorithmJuly 11, 2023DynIBar Software Enables Creation of New Views from Existing Video FootageResearchers at Cornell University and Google Research have collaborated to develop a groundbreaking software called DynIBar, which has the capability to create entirely new views from existing video footage. This innovative algorithm, presented at the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition, synthesizes new views by utilizing pixel information from the original video, even compensating for moving objects and unstable camerawork. Unlike previous efforts that only yielded a few seconds of video and often resulted in blurry or glitchy moving subjects, DynIBar represents a significant advancement in the field. The software incorporates image-based rendering techniques to handle complex scenes and longer videos, making it possible to stabilize shaky footage, change viewpoints, and apply freeze-frame, zoom, and slow-motion effects without the need for shooting additional footage. Although the software is not yet integrated into commercial video editing tools, the code is freely available for further exploration. While the current processing time is several hours for just a short video clip, the researchers are actively working on improving efficiency and addressing challenges such as rendering new images when pixel information is lacking. With further development, DynIBar holds the potential to revolutionize both personal and professional video editing, opening up exciting possibilities for filmmakers and content creators alike.
Source: Cornell University