Revisiting Pre-analysis Information Based Rate Control in x265

TL;DR

This paper introduces an improved rate control scheme for x265 that leverages pre-analysis data and combines CU-tree with a lambda-domain model, resulting in significant BD-rate gains and more accurate bitrate management.

Contribution

The paper presents a novel cost-guided rate control method that better utilizes pre-analysis information and integrates CU-tree with a lambda-domain model for enhanced performance.

Findings

Achieves 10.3% BD-rate reduction

Maintains only 0.22‰ bitrate error

Outperforms original x265 in R-D performance

Abstract

Due to the excellent compression and high real-time performance, x265 is widely used in practical applications. Combined with CU-tree based pre-analysis, x265 rate control can obtain high rate-distortion (R-D) performance. However, the pre-analysis information is not fully utilized, and the accuracy of rate control is not satisfactory in x265 because of an empirical linear model. In this paper, we propose an improved cost-guided rate control scheme for x265. Firstly, the pre-analysis information is further used to refine the bit allocation. Secondly, CU-tree is combined with the lambda-domain model for more accurate rate control and higher R-D performance. Experimental results show that compared with the original x265, our method can achieve 10.3\% BD-rate gain with only 0.22\textperthousand bitrate error.