Patent attributes
A method for quality assessment of a video that includes M video frames. The method includes repeating a first iterative process M times and extracting a score distribution for a plurality of scores. An mth iteration of the first iterative process, where m ∈ [1,M], includes generating an mth frame-level feature set, generating a first recurrent output of a plurality of recurrent outputs based on a zeroth recurrent output of the plurality of recurrent outputs, and generating an mth recurrent output of the plurality of recurrent outputs based on an (m−1)th recurrent output of the plurality of recurrent outputs. The first recurrent output is generated by feeding a first frame-level feature set to a recurrent neural network. The mth recurrent output is generated by feeding the mth frame-level feature set to the recurrent neural network. The score distribution is extracted from an Mth recurrent output of the plurality of recurrent outputs.
