Weighted Labeling for Image Recognition Model Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The accuracy of image recognition models is affected by subjective labeling of training data, leading to inconsistent and inaccurate recognition results due to the reliance on a single labeling result with the highest occurrence times.
Innovation Solution
A method that determines a weight coefficient for each labeling result based on its occurrence times, allowing for a more accurate label determination and improved recognition model training by incorporating multiple labeling results and their corresponding weights.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the labeling result with the maximum occurrence times is used as the label of training data, then the labeling process is simple and efficient, but the accuracy and credibility of the recognition model is reduced due to subjective thinking influences
Solution Approach 1:
The patent changes the parameter of label determination from a discrete selection (single most frequent label) to a continuous weighted combination of multiple labels. By introducing weight coefficients that reflect the frequency and reliability of different labeling results, the system transforms the labeling approach to preserve more information while maintaining efficiency.
Solution Approach 2:
The patent creates a composite label structure by combining multiple labeling results with different weight coefficients. Instead of using a single homogeneous label, the system constructs a composite representation that incorporates the strengths of multiple labeling perspectives, thereby improving recognition accuracy without sacrificing labeling efficiency.
2Measurement precision
If multiple labeling results are incorporated with weight coefficients, then the recognition accuracy is improved, but the complexity of the labeling process increases
Solution Approach 1:
The system performs self-service by automatically calculating weight coefficients based on the occurrence frequencies of different labeling results. Rather than requiring manual intervention to determine weights or select labels, the system autonomously processes multiple labeling results and generates the weighted combination, thereby reducing the perceived complexity for users.
Solution Approach 2:
The patent implements a feedback mechanism where the weight coefficients are determined based on the statistical feedback from multiple labeling results. The system uses the occurrence times of each label as feedback to assign appropriate weights, creating a self-adjusting process that improves accuracy while keeping the methodology systematic and manageable.
3Quantity of substance
If artificial labeling is used for training data, then the training data can be obtained, but the subjective thinking of persons causes inconsistent labeling results
Solution Approach 1:
The patent merges multiple inconsistent labeling results into a unified weighted label representation. By combining several subjective labeling outcomes rather than selecting just one, the system preserves the diversity of human judgment while creating a more stable and comprehensive training label that reflects the consensus and variations in human perception.
Solution Approach 2:
The system cushions against the inconsistency of human labeling by pre-incorporating multiple labeling perspectives into the training data. Instead of relying on a single potentially biased label, the weighted combination approach prepares the training data to be more robust against subjective variations, thereby cushioning the recognition model from inconsistent labeling influences.
Data Source
AI summary
A method and apparatus for recognizing an image, and a storage medium are provided. The method includes: obtaining labeling results of training data to be labeled in a target training data set; determining a weight coefficient of each labeling result in response to that the training data includes a plurality of labeling results; determining a label of the training data based on labeling result and the weight coefficient corresponding to the labeling result, wherein the training data with the label is the target training data for training a recognition model; and recognizing the image based on the recognition model trained based on the target training data.