VIDEO CODING METHOD AND DEVICE THEREFOR

Smart overview of the Invention

The patent application describes methods for video coding, specifically focusing on image decoding and encoding. It involves generating reference motion information for blocks within a current picture using pixel data or motion information from reconstructed images. Motion information for the current block is then obtained from these blocks, and the current block is reconstructed based on this information. A neural network may be used in generating the reference motion information.

Image Decoding Method

The image decoding method involves several key steps. Initially, reference motion information is generated for at least one block in the current picture using pixel data or motion information from reconstructed pictures. Motion information for a current block is then determined using this reference information along with syntax data from a bitstream. The current block is reconstructed based on its specific motion information, which may include various types of motion vector and prediction list utilization data.

Image Encoding Method

The encoding process mirrors the decoding method to some extent. Reference motion information is generated for blocks in the current picture, similar to the decoding process. This information helps in determining the motion data for a current block, which is then encoded into a bitstream. The encoding method ensures efficient storage and retrieval of video data by utilizing advanced techniques like neural networks for generating reference motion data.

Additional Features

Several additional processes are included in the methods. These include aggregating, scaling, or transforming reference motion information formats, and down-sampling reconstructed pictures to aid in generating reference motion data. The methods also consider time intervals between pictures to optimize the use of reference motion information and support the generation of additional picture motion data through interpolation or extrapolation.

Technical Aspects

The technical details encompass constructing merge candidate lists using reference motion information and dividing current blocks into subblocks for detailed motion compensation. The methods support both uni-directional and bi-directional motion vectors linked to reference picture lists, enhancing flexibility in video processing. This patent aims to address challenges in high-resolution image handling by proposing diverse coding unit configurations.