Face Recognition Training With Quality-Adaptive Margins
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing facial recognition systems struggle with low-quality images, as they often rely on uniform training that emphasizes all data types equally, leading to poor performance on low-quality datasets due to unidentifiable images that focus on irrelevant visual characteristics.
Innovation Solution
A facial recognition system that adaptively assigns different margin functions to samples based on their image quality and recognition difficulty, using feature norm as a proxy to emphasize hard samples in high-quality images and de-emphasize unidentifiable samples in low-quality images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If uniform training is applied to all face images regardless of quality, then the training process is simple and consistent, but recognition performance on low-quality datasets deteriorates due to unidentifiable images dominating the training loss
Solution Approach 1:
The patent applies different margin functions to different training samples based on their image quality. High-quality images use one margin function while low-quality images use another, allowing the system to adapt to local characteristics of each sample rather than applying a uniform training approach throughout the entire dataset
Solution Approach 2:
The margin function is dynamically selected based on the quality assessment of each image sample. The system transitions from a static, uniform margin function to a dynamic selection process where the appropriate margin function changes depending on the input image quality, enabling adaptive training behavior
2Measurement precision
If the model emphasizes all training samples equally, then the training procedure is straightforward, but the model learns irrelevant visual characteristics from unidentifiable low-quality images instead of meaningful identity features
Solution Approach 1:
The patent changes the margin parameter based on image quality assessment. By adjusting the margin function parameter according to whether an image is high or low quality, the system modifies the training objective to either emphasize or de-emphasize samples based on their ability to provide meaningful identity information
Solution Approach 2:
Different margin functions are applied to different regions of the training data space based on image quality. This local differentiation ensures that only samples with sufficient quality contribute meaningfully to identity feature learning, while low-quality samples that would introduce noise are handled differently
Data Source
AI summary
A method of training a facial recognition system with a plurality of image samples of a training set includes determining an image quality of each of the image samples in the training set, assigning a margin function to each of the image samples based on the image quality of each image sample, classifying the image samples, determining a prediction, determining a loss based on the prediction and the margin function, generating gradients based on the loss and changing weights in the classifier based on the gradients.