Entropy Coding-Based Feature Encoding/Decoding Method and Device

ChangeBridge: Patent Grants - AI & Computing (G06N)

Published March 31st, 2026

Detected April 1st, 2026

Summary

USPTO granted Patent US12593045B2 to LG Electronics Inc. for an entropy coding-based feature encoding/decoding method and device. The patent relates to machine learning and video coding technology, with inventors Chulkeun Kim and Hyeong Moon Jang. The invention covers methods for determining context models based on channel grouping information of feature maps.

View original document View source feed page

What changed

USPTO granted Patent US12593045B2 to LG Electronics Inc. on March 31, 2026. The patent covers entropy coding methods for feature encoding/decoding, where context models are determined based on channel information including channel grouping information of feature maps. The technology applies to machine learning (G06N 20/00) and video coding applications (H04N 19 series).

Technology companies and manufacturers developing AI/ML compression technologies or video coding products should review the patent claims for potential licensing needs or freedom-to-operate considerations. The patent has 14 claims and covers methods for encoding and decoding current coding units using context models determined from channel grouping information.

Source document (simplified)

← USPTO Patent Grants

Entropy coding-based feature encoding/decoding method and device, recording medium having bitstream stored therein, and method for transmitting bitstream

Grant US12593045B2 Kind: B2 Mar 31, 2026

Assignee

LG Electronics Inc.

Inventors

Chulkeun Kim, Hyeong Moon Jang

Abstract

Herein provided are a feature encoding/decoding method, a recording medium storing a bitstream generated by the feature encoding method and a method of transmitting the bitstream. The feature decoding method may comprise obtaining channel information of a current coding unit, determining a context model for the current coding unit based on the channel information; and decoding the current coding unit based on the context model. The channel information may comprise channel grouping information of a feature map including the current coding unit, and the context model is determined based on the channel grouping information.