Machine Learning Framework for Incomplete Image Defect Labeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for ensuring image quality in electronic catalogs rely heavily on human intervention, which is inefficient and cannot scale to handle large volumes of user-uploaded images, especially in scenarios of rapid catalog expansion and new market entries, due to incomplete labeling and class imbalance issues.
Innovation Solution
A machine learning framework that uses partial information for training, employing a novel loss function and label values to account for unknown ground truth labels, and a deep neural network architecture that predicts multiple image defects with high accuracy, even with incomplete human annotations, thereby automating the image review process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human reviewers manually label images to ensure complete ground truth information, then measurement precision is improved, but productivity deteriorates due to inability to scale
Solution Approach 1:
The system uses automated machine learning models to perform image defect detection without requiring human reviewers to manually examine and label each image, enabling the system to serve itself at scale while maintaining consistent quality standards
Solution Approach 2:
The patent replaces the mechanical human review process with an automated machine learning system that processes images through trained models, substituting human manual labor with algorithmic automation to achieve both high throughput and consistent precision
2Productivity
If automated machine learning systems are deployed to handle large volumes of images, then productivity is improved, but measurement precision deteriorates due to incomplete labeling and class imbalance
Solution Approach 1:
The system performs preliminary actions by pre-processing images to identify and handle potential defect regions before main classification, and by implementing data augmentation and synthetic sample generation to prepare balanced training data in advance, addressing class imbalance before the main detection task
Solution Approach 2:
The patent changes key parameters including loss function design to handle incomplete labels, data distribution through augmentation techniques, and model architecture parameters to optimize for multi-defect detection, transforming the approach to overcome precision deterioration
3Adaptability or versatility
If multiple image defects are detected simultaneously, then adaptability is improved, but device complexity increases due to multi-label classification requirements
Solution Approach 1:
The system segments the complex multi-defect detection task into separate binary classification sub-tasks for each defect type, where each sub-model handles one specific defect classification independently, reducing overall system complexity while maintaining multi-defect detection capability
Solution Approach 2:
The patent implements a universal base model architecture that can be applied across multiple defect types through consistent preprocessing and standardized output handling, allowing the same core system to perform multiple defect detection functions without proportionally increasing complexity
Data Source
AI summary
Machine learning techniques are disclosed for training a model to identify each of multiple different classes in images, based on training data where each training image may not be labeled in a complete manner with respect to the classes. The disclosed training techniques use a new label value to indicate when a ground truth value is unknown for a particular class, and do not penalize the machine learning model for output predictions that do not match the label value representing unknown ground truth. The disclosed processes may, for example, be used to train a model to detect each multiple types of image defects based on incomplete information provided by human reviewers who accept and reject images based on whether any of the types of image defects are found.