An Efficient Neural Network Based Rate Control for Intra-Frame in Versatile Video Coding

Zeming Zhao, Xiaohai He, Shuhua Xiong, Xiaodong Bi, Honggang Chen

Abstract: As the prevalence of video continues to grow, the coding, transmission, and processing of video signals is becoming increasingly crucial for intelligent systems, particularly in vehicle driving systems. Moreover, with the rapid advancement of neural networks in the realm of image coding, they emerge as a compelling force shaping the future of video coding. To explore this potential, this paper proposes an Efficient Neural Network Based Rate Control (ENNRC) for intra-frame in Versatile Video Coding (VVC). In particular, a neural network based bits prediction model is developed to directly map video content features to predicted bits, serving as guidance for bit allocation at both the frame and Coding Tree Unit (CTU)-levels. Subsequently, we introduce the improved parameter updating algorithm at frame-level. Our experimental findings demonstrate that the proposed RC algorithm achieves a 7.23$\%$ BD-rate savings while offering a more accurate allocation compared to VVC's default RC algorithm.

External IDs:dblp:journals/tvt/ZhaoHXBC25