Efficient Content-Driven Encoding Towards a Target Video Quality

Video streaming workflows aim to maximize video quality while still maintaining smooth video streaming performance. A traditional fixed bitrate ladder consists of predetermined bitrate-resolution pairs which are optimized across a wide variety of content. Consequently, these pairs are rarely optimized for a given piece of content. Some encoding tools address this by encoding each piece of video content with many codec parameters and then evaluating the results using a video quality metric. However, this process requires significant computation which increases cost and encoding time. In this paper, we propose a novel content-driven workflow that predicts optimal encoding parameters to achieve a target perceptual video quality. We do so by designing a deep learning model that, based on the video input, predicts a VMAF rate-distortion curve. Our results indicate that such a content-driven approach is an efficient way to reduce the number of encoding attempts, minimize necessary cloud computing resources, encode most efficiently, and maximize perceptual video quality.

Published

2024-10-21

Content type

Original Research

Keywords

bitrate ladder, live streaming, rate-quality curves, video coding, video compression, video quality, vmaf

DOI

10.5594/MOO/3048

ISBN

978-1-61482-965-2