Machine Vision Compression with Semantic Post-Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies prioritize human visual system characteristics for compression, which may not be optimal for machine vision tasks, leading to inefficiencies in machine task performance and computational challenges.
Innovation Solution
A post-processing network is applied to enhance semantic-related information in reconstructed visual signals without altering existing codecs, focusing on improving machine vision performance through methods like QP adaptive visual signal enhancement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If existing video coding technologies are used for compression, then compression efficiency is improved, but machine vision task performance deteriorates
Solution Approach 1:
The processing pipeline is segmented into three independent stages: (1) standard video coding for compression, (2) post-processing network for semantic enhancement, and (3) machine vision tasks. This allows each stage to optimize for its specific function without compromising the others.
Solution Approach 2:
A post-processing network is introduced as an intermediary component between the compressed video and machine vision tasks. This intermediary enhances semantic information in the reconstructed visual signal, bridging the gap between compression efficiency and machine vision performance requirements.
2Reliability
If post-processing network is added to enhance semantic information, then machine vision performance is improved, but device complexity increases
Solution Approach 1:
The post-processing network performs preliminary enhancement of semantic information in the reconstructed visual signal before it is fed to machine vision tasks. This preliminary action prepares the data in advance, reducing the computational burden on subsequent processing stages.
Solution Approach 2:
The system dynamically adjusts processing parameters based on the compression ratio and visual signal characteristics. The post-processing network adapts its operation based on the input signal quality, optimizing performance while managing computational resources efficiently.
Data Source
AI summary
A video processing method includes compressing and reconstructing an original visual signal to obtain a reconstructed visual signal; processing the reconstructed visual signal to obtain a post-processed visual signal; and feeding the post-processed visual signal to a machine task network