LSTM-based Video Quality Prediction Accounting for Temporal Distortions in Videoconferencing Calls

Although other pooling mechanisms have been studied in order to take recency effects into account, in most studies Current state-of-the-art video quality models, such as VMAF, mean pooling achieved the best performance [6, 7]. VMAF is give excellent prediction results by comparing the degraded the only of those metrics that include a temporal component, video with its reference video. However, they do not consider which considers temporal masking effects. However, video temporal distortions (e.g., frame freezes or skips) that transmitted during VC calls can be affected by a number of occur during videoconferencing calls. In this paper, we temporal distortions that are perceived as frame freezes, frame present a data-driven approach for modeling such distortions skips, frame rate variations (e.g., video is played back faster automatically by training an LSTM with subjective after delayed packets arrive), or generally low frame rate.