Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Bit Rate Matching Algorithm Optimization in JPEG-AI Verification Model

Panqi Jia, A. Burakhan Koyuncu, Jue Mao, Ze Cui, Yi Ma, Tiansheng Guo, Timofey Solovyev, Alexander Karabutov, Yin Zhao, Jing Wang, Elena Alshina, Andre Kaup

TL;DR

The paper tackles the slowdown of bit rate matching (BRM) in the JPEG-AI VM4.1 verification model by introducing three algorithmic optimizations. It leverages a relative bit-distance model selection, a log-linear relationship between $\beta_{test}$ and $bpp$ to reduce search iterations, and a validation approach that avoids redundant decoding, collectively achieving up to 6.3× speedups with BD-rate performance comparable to or better than the prior method. Experiments on a 50-image CTTC dataset show notable gains in runtime (4× at BOP, 6.3× at HOP) and modest improvements in BD-rate, highlighting practical benefits for continuous variable rate coding in NN-based image compression. The approach enhances the feasibility of bitrate-matching in JPEG-AI's verification framework, supporting robust standardization and deployment of JPEG-AI codecs. The mathematical relation $\log(bpp) = A \cdot \log(\beta_{test}) + B$ underpins the efficient $\beta_{test}$ search, while $D_r$-driven model selection minimizes unnecessary computations.

Abstract

The research on neural network (NN) based image compression has shown superior performance compared to classical compression frameworks. Unlike the hand-engineered transforms in the classical frameworks, NN-based models learn the non-linear transforms providing more compact bit representations, and achieve faster coding speed on parallel devices over their classical counterparts. Those properties evoked the attention of both scientific and industrial communities, resulting in the standardization activity JPEG-AI. The verification model for the standardization process of JPEG-AI is already in development and has surpassed the advanced VVC intra codec. To generate reconstructed images with the desired bits per pixel and assess the BD-rate performance of both the JPEG-AI verification model and VVC intra, bit rate matching is employed. However, the current state of the JPEG-AI verification model experiences significant slowdowns during bit rate matching, resulting in suboptimal performance due to an unsuitable model. The proposed methodology offers a gradual algorithmic optimization for matching bit rates, resulting in a fourfold acceleration and over 1% improvement in BD-rate at the base operation point. At the high operation point, the acceleration increases up to sixfold.

Bit Rate Matching Algorithm Optimization in JPEG-AI Verification Model

TL;DR

The paper tackles the slowdown of bit rate matching (BRM) in the JPEG-AI VM4.1 verification model by introducing three algorithmic optimizations. It leverages a relative bit-distance model selection, a log-linear relationship between and to reduce search iterations, and a validation approach that avoids redundant decoding, collectively achieving up to 6.3× speedups with BD-rate performance comparable to or better than the prior method. Experiments on a 50-image CTTC dataset show notable gains in runtime (4× at BOP, 6.3× at HOP) and modest improvements in BD-rate, highlighting practical benefits for continuous variable rate coding in NN-based image compression. The approach enhances the feasibility of bitrate-matching in JPEG-AI's verification framework, supporting robust standardization and deployment of JPEG-AI codecs. The mathematical relation underpins the efficient search, while -driven model selection minimizes unnecessary computations.

Abstract

The research on neural network (NN) based image compression has shown superior performance compared to classical compression frameworks. Unlike the hand-engineered transforms in the classical frameworks, NN-based models learn the non-linear transforms providing more compact bit representations, and achieve faster coding speed on parallel devices over their classical counterparts. Those properties evoked the attention of both scientific and industrial communities, resulting in the standardization activity JPEG-AI. The verification model for the standardization process of JPEG-AI is already in development and has surpassed the advanced VVC intra codec. To generate reconstructed images with the desired bits per pixel and assess the BD-rate performance of both the JPEG-AI verification model and VVC intra, bit rate matching is employed. However, the current state of the JPEG-AI verification model experiences significant slowdowns during bit rate matching, resulting in suboptimal performance due to an unsuitable model. The proposed methodology offers a gradual algorithmic optimization for matching bit rates, resulting in a fourfold acceleration and over 1% improvement in BD-rate at the base operation point. At the high operation point, the acceleration increases up to sixfold.
Paper Structure (11 sections, 5 figures, 1 table)

This paper contains 11 sections, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Bit Rate Matching in the JPEG-AI Verification Model
  3. Model Selection
  4. $\beta_{test}$ Searching for Single Model
  5. $\beta_{test}$ Validation
  6. Proposed Algorithm
  7. Relative Bit Distance Based Model Selection
  8. $\beta_{test}$ Searching Based on Linear Function
  9. $\beta_{test}$ Validation simplification
  10. Experimental Results
  11. Conclusion

Figures (5)

  • Figure 1: Two steps of the model selection in the prior art
  • Figure 2: $\beta_{test}$ validation process (a) in prior art and (b) in our simplified method
  • Figure 3: Variable rate curve of a single model with different $\delta_\beta$
  • Figure 4: Directly select model by comparing the $D_r$
  • Figure 5: Relationship between ${\rm log}(bpp)$ and ${\rm log}(\beta_{test})$