Abstract
Generating novel synthetic photographs of locations can be challenging, as some generative techniques may introduce visual artifacts or compositional inconsistencies when creating new camera perspectives. The described systems and methods can address this by using two or more source images as input for a generative video synthesis model. The model can produce a short video that interpolates or extrapolates between the viewpoints of the source images, creating a set of new, spatially coherent camera angles. This generated video may then be deconstructed into a pool of candidate frames. A heuristic, such as a feature similarity score, can be used to analyze and select one or more frames as synthetic photographs. This process can be used to generate new visual assets that are distinct from the original source material.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Leung, Mira, "Generation of Synthetic Photographs via Video Synthesis and Heuristic Frame Selection", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/9184