Constant-slope rate allocation for distributed real-world encoding

Parallel encoding systems face the problem of distributed coding decisions, an example of which is rate allocation between different encoding chunks. Distributed encoding is traditionally performed by assigning a fixed bitrate or quality setting to each chunk, after which the chunk uses rate control to achieve that particular target. In this paper, we discuss the use of constant-slope encoding, resulting in chunks which all operate at the same quality-rate trade-off. The (Lagrangian) optimization can be performed independent of the chosen codec or distortion/quality metric. In this paper, we report results for mobile resolutions for VP9 encodes, on both open-source and Netflix catalog original and licensed content titles. BD-rate gains are obtained of 6.8 % on average, and a more constant quality across titles is observed.