I. Schiopu and A. Munteanu, "Deep-Learning based Macro-Pixel Synthesis and Lossless Coding of Light Field Images," APSIPA Transactions on Signal and Information Processing (TSIP). Manuscript submitted March 20, 2019; revised May 15, 2019; accepted June 13, 2019.

Two models are proposed: MPS-CNN for synthesis and PSMP-CNN for coding. Each model was trained for each color channel R, G, B, and for each configuration 2x2, 3x3, 4x4, and 5x5. The repository contains 24 models trained based on 10 LF images from the JPEG Pleno Database jpeg.org/plenodb/lf/epfl.