Neural network based decision for picture resolution change in video coding

A neural network-based method for determining quantization switches and resampling classes addresses the complexity and efficiency challenges in video encoding, optimizing reference picture resampling decisions for improved compression performance.

WO2026027162A1PCT designated stage Publication Date: 2026-02-05INTERDIGITAL CE PATENT HOLDINGS SAS
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
PCT/EP2025/068835
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2024-07-31
Filing Date
2025-07-02
Publication Date
2026-02-05

AI Technical Summary

Technical Problem

Existing video encoding methods face challenges in efficiently determining when to apply reference picture resampling (RPR) for optimal picture resolution change, leading to increased encoding complexity and suboptimal compression efficiency.

Method used

A neural network-based approach is employed to determine quantization switches and resampling classes using features from a picture, allowing for real-time decision-making on whether to apply reference picture resampling based on quantization parameters, reducing encoding complexity and improving compression efficiency.

Benefits of technology

The neural network model effectively predicts optimal resampling ratios and decisions, enhancing video encoding performance by reducing encoding time and improving bitrate-quality tradeoffs without additional complexity.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure EP2025068835_05022026_PF_FP_ABST
    Figure EP2025068835_05022026_PF_FP_ABST
Patent Text Reader

Abstract

A method of video encoding is presented. At least one feature is obtained from a picture. For each resampling ratio, a quantization switch and a resampling class are determined using a neural network with the at least one feature as input. A resampling ratio and a decision on whether reference picture resampling is to be applied to encode the picture are determined based on the quantization switches, the resampling classes and a quantization parameter to be used for encoding the picture. The picture is encoded responsive to the determining.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Training method and apparatus for target detection model, device and storage medium

    US20220147822A1