Precise Encoding Complexity Control for Versatile Video Coding
Yan Huang,Jun Xu,Chen Zhu,Li Song,Wenjun Zhang
DOI: https://doi.org/10.1109/tbc.2022.3187813
IF: 4.5
2023-01-01
IEEE Transactions on Broadcasting
Abstract:Complexity reduction is a commonly used method to deal with complicated video coding standards, such as High Efficiency Video Coding (HEVC) and Versatile Video Coding (VVC). But the unstable performance under different video contents and Quantization Parameters (QPs) makes it difficult to precisely specify the target encoding time of every single sequence, which limits the practical use of the encoder. Inspired by rate control, in this paper, encoding time budget is regarded as a resource. We incorporate the hierarchical Group of Picture (GOP) encoding structure, the non-accelerated proportion, and block content variation within the frame, and design a top-down allocation and bottom-up feedback scheme, to achieve precise control of encoding complexity by controlling the maximum depth of QuadTree with nested Multi-type Tree (QTMT). In the scheme, Temporal ID (Tid)-based, Sum of Absolute Transformed Difference (SATD)-based methods and Linear (L) Model are designed to facilitate weighted allocation and feedback. The relationship between Planar Cost and encoding time is exploited as a Time-Cost (T-C) model, which guides the selection of Largest Coding Unit (LCU) encoding strategies in I-frames. For B-frames, we investigate a switching suppression assisted status-based method, to efficiently decide the encoding strategy of each LCU. Through the collaboration of the proposed technologies in the scheme, given any encoding time target achievable, encoding strategies will be automatically switched to help the actual encoding time gradually approach the target. To the best of our knowledge, this work is the first work on VVC-based encoding complexity control. The proposed scheme supports directly specifying the target encoding time or target Frame Per Second (FPS), and accurately realizing it within one pass. According to experimental results, under the target encoding time ratio of 80%, 60% and 40%, the average encoding time error is kept under 0.24%, 0.03% and 0.02% by our method, outperforming state-of-the-art works base on HEVC. These results prove the advantage of the proposed scheme.