Neural network-based content prediction methods for encoding or decoding video

Abstract

The proposed video encoding system performs in-image prediction in a mode using a neural network for a block with a specific set of block sizes. The signal of this mode is designed to be effective in scaling distortion under this constraint. Various transformations regarding the properties of a block and the neural network prediction of that block are incorporated to use a single neural network when predicting blocks of varying sizes, as well as corresponding signaling. The neural network-based prediction mode will consider both brightness and color blocks. The video encoding system includes encoders and decoders, encoding, decoding, and signal generation methods, and signal carrier information corresponding to the described encoding mode.