Binary Attribute Pre-training for Face Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object attribute classification methods, such as those using ImageNet pre-trained models, are ineffective for face attribute classification due to their focus on global category classification rather than specific attributes, leading to poor performance in distinguishing face attributes.
Innovation Solution
The proposed method involves acquiring binary class attribute data indicating whether an attribute is 'Yes' or 'No' for specific class labels and using this data for pre-training a model, allowing for efficient and accurate object attribute classification by focusing on attribute-specific relationships.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If ImageNet pre-trained models are used for object attribute classification, then the model can be pre-trained with general image data, but the classification accuracy for specific face attributes deteriorates because the model focuses on global category classification rather than specific attributes
Solution Approach 1:
The patent changes the training data parameters from general ImageNet images to binary class attribute data with 'Yes' or 'No' labels for specific attributes. This parameter change in data representation and labeling scheme enables the model to focus on specific attribute classification rather than global category classification, resolving the contradiction between pre-training availability and attribute classification accuracy
Solution Approach 2:
The patent performs preliminary action by pre-training the model specifically on binary class attribute data before the actual attribute classification task. This preliminary training with attribute-specific data prepares the model to focus on attribute features rather than global categories, improving subsequent attribute classification performance while maintaining the benefit of pre-training
2Productivity
If traditional pre-training methods are used, then the training process can be completed with general data, but the model performance in distinguishing specific face attributes deteriorates
Solution Approach 1:
The patent changes the data parameters from general images to binary class attribute data with explicit 'Yes' or 'No' labels. This parameter change maintains training efficiency while significantly improving attribute distinction performance by providing clear binary supervision signals that guide the model to focus on specific attribute features
Solution Approach 2:
The patent performs preliminary training on binary class attribute data before the main attribute classification task. This preliminary action with attribute-specific binary data establishes strong attribute feature representations in the model, improving reliability in distinguishing specific face attributes while maintaining training productivity
Data Source
AI summary
The present disclosure relates to method, apparatus and storage medium for object attribute classification model training. There proposes a method of training a model for object attribute classification, comprising steps of: acquiring binary class attribute data related to a to-be-classified attribute on which an attribute classification task is to be performed, wherein the binary class attribute data includes data indicating whether the to-be-classified attribute is “Yes” or “No” for each of at least one class label; and pre-training the model for object attribute classification based on the binary class attribute data.