Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Complexity Reduction Study Based on RD Costs Approximation for VVC Intra Partitioning

M. E. A. Kherchouche, F. Galpin, T. Dumas, F. Schnitzler, D. Menard, L. Zhang

TL;DR

This work tackles the high complexity of VVC intra partitioning by proposing two size-independent approaches to approximate RD costs for CU splits: a regression-based neural network and a Deep Q-Network (DQN) reinforcement learning agent. Both methods rely on a fixed 115-feature CU vector composed of Neighbor, Parent, Block, and Spatial information to predict normalized RD costs for potential splits, enabling selective testing without exhaustive search. Experimental results on VTM-18.0 AI configurations show substantial complexity reductions (up to around $20\%$) with modest bd-rate penalties (roughly $0.1$–$0.2\%$), with the RL approach offering competitive trade-offs and robustness across depths. Ablation studies confirm the contribution of each feature group, and the authors integrate the models as SADL modules for practical use in encoder pipelines, highlighting the potential for online adaptation and broader CU-size coverage in future work.

Abstract

In this paper, a complexity study is conducted for Versatile Video Codec (VVC) intra partitioning to accelerate the exhaustive search involved in Rate-Distortion Optimization (RDO) process. To address this problem, two main machine learning techniques are proposed and compared. Unlike existing methods, the proposed approaches are size independent and incorporate the Rate-Distortion (RD) costs of neighboring blocks as input features. The first method is a regression based technique that predicts normalized RD costs of a given Coding Unit (CU). As partitioning possesses the Markov property, the associated decision-making problem can be modeled as a Markov Decision Process (MDP) and solved by Reinforcement Learning (RL). The second approach is a RL agent learned from trajectories of CU decision across two depths with Deep Q-Network (DQN) algorithm. Then a pre-determined thresholds are applied for both methods to select a suitable split for the current CU.

Complexity Reduction Study Based on RD Costs Approximation for VVC Intra Partitioning

TL;DR

This work tackles the high complexity of VVC intra partitioning by proposing two size-independent approaches to approximate RD costs for CU splits: a regression-based neural network and a Deep Q-Network (DQN) reinforcement learning agent. Both methods rely on a fixed 115-feature CU vector composed of Neighbor, Parent, Block, and Spatial information to predict normalized RD costs for potential splits, enabling selective testing without exhaustive search. Experimental results on VTM-18.0 AI configurations show substantial complexity reductions (up to around ) with modest bd-rate penalties (roughly –), with the RL approach offering competitive trade-offs and robustness across depths. Ablation studies confirm the contribution of each feature group, and the authors integrate the models as SADL modules for practical use in encoder pipelines, highlighting the potential for online adaptation and broader CU-size coverage in future work.

Abstract

In this paper, a complexity study is conducted for Versatile Video Codec (VVC) intra partitioning to accelerate the exhaustive search involved in Rate-Distortion Optimization (RDO) process. To address this problem, two main machine learning techniques are proposed and compared. Unlike existing methods, the proposed approaches are size independent and incorporate the Rate-Distortion (RD) costs of neighboring blocks as input features. The first method is a regression based technique that predicts normalized RD costs of a given Coding Unit (CU). As partitioning possesses the Markov property, the associated decision-making problem can be modeled as a Markov Decision Process (MDP) and solved by Reinforcement Learning (RL). The second approach is a RL agent learned from trajectories of CU decision across two depths with Deep Q-Network (DQN) algorithm. Then a pre-determined thresholds are applied for both methods to select a suitable split for the current CU.

Paper Structure

This paper contains 14 sections, 6 equations, 3 figures, 2 tables, 1 algorithm.

Table of Contents

  1. VTM Fixed Configuration
  2. CU vector representation and RD costs approximation
  3. Neighbor Information (NI)
  4. Parent Information (PI)
  5. Block Information (BI)
  6. Spatial Information (SI)
  7. RD costs approximation
  8. Regression based algorithm overview
  9. RL based algorithm overview
  10. DQN training
  11. Experimental results
  12. Regression NNs versus DQN algorithm
  13. Ablation study
  14. Conclusion

Figures (3)

  • Figure 1: Left: Benchmark VTM-18.0 in All Intra configuration with four QTDepths. Right: Rate-distortion costs of a $32\times32$ CU for each splitting mode at different quantization parameter values. NS RD cost is subtracted from all RD costs.
  • Figure 2: Workflow diagram of the proposed methods
  • Figure 3: BD-rate versus complexity speed-up ratio comparison in All Intra (AI) configuration with four QTDepth configurations.