Neural Network Feature Map Quantization for Video Coding Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video signal coding technologies do not efficiently utilize neural network feature maps for improved coding efficiency.
Innovation Solution
A neural network-based method and device that generates and quantizes feature maps using multiple neural networks, considering the distribution type and structure of the networks, including uniform, Gaussian, and Laplace distributions, and classifies feature maps based on their attributes and spatial similarities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multiple neural networks are used for feature map generation, then coding efficiency is improved, but device complexity increases
Solution Approach 1:
The feature map quantization process is segmented into multiple independent neural networks, each handling specific quantization tasks. This allows parallel processing of different feature map components, improving overall coding efficiency while maintaining manageable complexity through modular architecture
Solution Approach 2:
The multiple neural networks are designed with universal quantization capabilities that can handle various distribution types (uniform, Gaussian, Laplace) and different feature map characteristics. This multi-functionality reduces the need for separate specialized networks, thereby improving efficiency without proportionally increasing device complexity
2Measurement precision
If feature map quantization is performed based on distribution type, then quantization precision is improved, but processing time increases
Solution Approach 1:
The distribution type of feature maps is identified and classified in advance before quantization processing. This preliminary classification allows the system to pre-select the most appropriate quantization neural network, avoiding runtime analysis and reducing processing time while maintaining high quantization precision
Solution Approach 2:
Different quantization parameters and methods are applied based on the identified distribution type (uniform, Gaussian, Laplace). By changing parameters according to the specific distribution characteristics, the system achieves optimal quantization precision for each case without using a single time-consuming universal method
3Productivity
If feature maps are classified into multiple classes, then coding efficiency is improved, but computational complexity increases
Solution Approach 1:
Feature maps are classified into different classes based on their local characteristics such as distribution type and spatial properties. Each class receives customized quantization treatment tailored to its specific characteristics, improving coding efficiency through localized optimization without requiring complex global processing
Data Source
AI summary
A neural network-based signal processing method and device according to the present invention generates a feature map by means of a multilayer neural network comprising a plurality of neural networks, and performs quantization for the feature map, the quantization performed on the basis of the structure of the multilayer neural network or the attribute of the feature map.