.png){kind=icon}
.png){kind=icon}
Return
ResNet-Based Fast CU Partition Decision Algorithm for VVC
.png){kind=icon}
OA
AI
DOI：10.1109/ACCESS.2022.3208135.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
H.266/VVC introduces new technologies based on previous video coding standards, including the QTMT partition structure and support for 65 angular intra-prediction modes. The introduced new technologies significantly improve the coding efficiency, but also lead to a great increase in coding complexity. This paper proposes a ResNet-based fast CU partition decision algorithm to reduce the coding complexity of VVC. First, we count and analyze the proportion of CU split modes to explore the appropriate CNN models. Then, the ResNet-based CNN models are designed to predict CU split mode, and the first convolutional layer of the CNN models combines symmetric and asymmetric convolutional kernels to extract features efficiently. We also introduce the RD cost into the loss function to improve the prediction accuracy of CNN models. Finally, two threshold schemes are used to achieve a compromise between coding complexity and coding performance. The experimental results show that the Fast scheme saves 55.12% of the encoding time with 1.83% BDBR increase and the Moderate scheme saves 47.03% of the encoding time with only 1.27% BDBR increase compared with VTM7.0.
Keywords:
Encoding
Complexity theory
Predictive models
Copper
Feature extraction
Video sequences
Costs
Kernel
Convolutional neural networks
VVC
QTMT partition structure
ResNet
asymmetric convolutional kernel
Journal
IF:
3.6
Papers:
9.8W
Citations:
29.4W