Pareto-optimized coding configurations for VVenC, a fast and efficient VVC encoder

The Versatile Video Coding (VVC) standard was finalized in July 2020. It has since been studied in the literature with regards to both compression efficiency and complexity. Most of the experiments regarding VVC were performed based on the VTM reference software using its full configuration, usually compared to the test model of its predecessor, High Efficiency Video Coding (HEVC), HM. In this paper, we present a comprehensive and in-depth analysis of the VVC encoding complexity by using the fast and efficient open VVC encoder implementation VVenC. As a practical non-reference encoder, it offers alternative tradeoffs between compression efficiency and complexity. To find those, we approximate the Pareto set of the configuration space in an iterative process, encompassing search space extending encoding tools as well as search space pruning fast algorithms. While the VVenC Pareto set is derived based on HD and UHD video encoding results, it is validated and improved for alternative input material, namely for low-resolution and screen content video. Based on this optimization, five presets are defined constituting approximate Pareto-optimal encoding points for various content types as well as for multi-threaded operation. Without multi-threading, the presets span working points from faster at 4.7% HM runtime at 11% BD-rate gain through medium providing 38% BD-rate gain over HM at a quarter of the HM runtime, up to slower providing all of VTM's gain at less than half of its runtime.