Dynamic Margin Neural Network for Face Attribute Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In face attribute recognition, the use of a fixed margin value to expand feature angles between categories is ineffective due to varying feature distances, leading to high errors and reduced recognition accuracy, especially in systems like crowd analysis and identity verification.
Innovation Solution
A method for generating a neural network that determines a dynamically changing margin value based on semantic relationships between attributes, using a predefined table to reflect different semantic distances between categories, thereby adjusting feature distances and improving recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a fixed margin value is used to expand feature angles between categories, then the feature distance between different face categories is expanded, but the recognition accuracy deteriorates due to varying feature distances between different face attribute categories
Solution Approach 1:
The patent applies dynamics by transforming the fixed margin value into a dynamic margin value that changes based on the semantic relationship between attribute categories. The margin value is no longer static but adapts according to the semantic distance, allowing the system to handle varying feature distances between different face attribute categories effectively.
Solution Approach 2:
The patent changes the parameter of margin value from a fixed constant to a variable parameter determined by semantic relationships. By introducing semantic distance as a determining factor, the margin value parameter is adjusted dynamically to match the varying feature distances between different attribute categories, thereby improving recognition accuracy.
2Ease of manufacture
If the same fixed margin value is applied to all face attribute categories, then the implementation is simple, but the recognition accuracy deteriorates due to inability to measure different feature distances between categories
Solution Approach 1:
The patent changes the margin value parameter from a uniform fixed value to category-specific dynamic values based on semantic relationships. This parameter transformation allows the system to account for different feature distances between attribute categories while maintaining a systematic approach through predefined semantic relationships.
Solution Approach 2:
The patent applies local quality by assigning different margin values to different attribute category pairs based on their specific semantic relationships. Instead of a uniform approach, each category pair receives a tailored margin value that reflects its local semantic distance, improving recognition accuracy for each specific case.
3Device complexity
If a fixed margin value is used, then the computational process is simple, but the number of errors between labeled and predicted attributes increases due to large feature distance variations
Solution Approach 1:
The patent transforms the fixed margin parameter into a dynamic parameter that varies with semantic relationships. This change introduces a more sophisticated computational approach where the margin value is determined by semantic distance calculations, thereby reducing errors despite increased computational complexity.
Solution Approach 2:
The patent incorporates feedback mechanisms by using semantic relationship information to adjust margin values. The system continuously adapts the margin parameter based on the semantic distance between attribute categories, creating a feedback loop that improves recognition accuracy by accounting for feature distance variations.
Data Source
AI summary
The present disclosure includes a method, apparatus and system for generating a neural network and a non-transitory computer readable storage medium storing instructions. The method comprises: recognizing at least an attribute of an object in a sample image according to a feature extracted from the sample image, using the neural network; determining a loss function value at least according to a margin value determined based on a semantic relationship between attributes, wherein the semantic relationship is obtained from a predefined table at least according to a real attribute and the recognized attribute of the object, wherein the predefined table is composed of the attributes and the semantic relationship between the attributes; updating a parameter in the neural network according to the determined loss function value. When using the neural network generated according the present disclosure, the accuracy of object attribute recognition can be improved.