Methods, apparatus and storage media for video processing

By utilizing decoding devices and feature reuse modes in video processing to share cached feature information and generate high-resolution video frames, the problem of redundant computation in super-resolution models is solved, and efficient video processing is achieved.

CN121397272BActive Publication Date: 2026-05-26VASTAI TECH (SHANGHAI) INC
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
VASTAI TECH (SHANGHAI) INC
Filing Date
2025-12-24
Publication Date
2026-05-26

AI Technical Summary

Technical Problem

Existing super-resolution models fail to effectively utilize the temporal stability of mid-to-deep features in video scenarios, leading to increased redundant computation and intermediate activation/memory access overhead. Furthermore, existing cross-frame propagation methods fail to integrate with the rich residual/long-skip structures in super-resolution models, thus failing to effectively reduce redundant computation.

Method used

By decoding video frames using a decoding device and determining the feature reuse mode, high-resolution video frames are generated using cached feature information associated with the previous frame. This avoids performing some processing stages of the super-resolution model, enabling feature sharing between different video frames and reducing redundant calculations.

Benefits of technology

It effectively reduces the computational cost of super-resolution models, improves video image quality and processing efficiency, reduces redundant calculations, and enhances processing efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121397272B_ABST
    Figure CN121397272B_ABST
Patent Text Reader

Abstract

Embodiments of this disclosure relate to a method, apparatus, and storage medium for video processing. The method includes: reconstructing a first video frame in a video by decoding a bitstream, the first video frame corresponding to a first resolution; providing the first video frame to a super-resolution model to determine a first intermediate feature corresponding to a first processing stage of the super-resolution model; determining a feature reuse mode corresponding to the first video frame; in response to a feature reuse mode indication, performing super-resolution processing using cached feature information, acquiring a second intermediate feature corresponding to a second video frame without performing a second processing stage of the super-resolution model; determining a fusion feature based on the first and second intermediate features; and generating a third video frame corresponding to the first video frame based on the fusion feature, the second resolution of the third video frame being higher than the first resolution. In this manner, embodiments of this disclosure can improve video processing efficiency while maintaining image quality.
Need to check novelty before this filing date? Find Prior Art