FROST

The goal of FROST: FROxel-based Semantic Processing Techniques, is to demonstrate the advantages of a light field native data representation. Unlike approaches such as MPEG-MIV and JPEG-PLENO, which rely on traditional image processing techniques like atlases and warping, FROST utilizes the native resolution of the light field camera in all dimensions (spatial and temporal) by working at a ray level. Although this is similar to approaches like Neural Radiance Fields (NeRFs), FROST is unique in that it does not require per-scene training, as it remains a traditional (non-neural) processing technique. As such, it aims to bridge the gap between conventional representation and the emerging neural representation by leveraging knowledge about the sampling patterns of light field arrays.