Memory Profiling of H.266 Versatile Video Coding Standard

Abstract

This work presents a profiling of the memory requirements for the encoding flow of H.266 Versatile Video Coding. The goal is to analyze the impact of each encoding module and highlight the influence of the novel coding tools introduced in H.266NVC. This profiling is performed through two sets of experiments: (1) an overall memory requirements breakdown, displaying the distribution of the accesses per encoding module, and (2) a specific assessment of the prediction steps. As results, we observed that: prediction steps perform the most memory-intensive operations in the encoding process (47 % to 58 % of all accesses); H.266/VVC novelties on transforms operations increase their memory overhead (20%, on average, of the entire encoder). Moreover, at prediction-specific evaluation, we concluded that the interpolation filters and the distortion calculations have a significant impact in the memory requirements: an average of 39% and 32% of all prediction accesses, respectively. Further, H.266/VVC novel tools on the prediction steps, such as the Affine Motion Estimation and the Geometric-Based Partitioning, do not represent major overhead on H.266NVC high memory requirements.