Hierarchical Clustering Tree for Feature Extraction Model Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing self-supervised learning methods based on contrastive learning fail to model hierarchical semantic structures in natural image datasets, resulting in features of different images being far apart in the feature space, which limits their applicability to higher-level classification scenarios.
Innovation Solution
A method for generating a feature extraction model that involves obtaining candidate images, determining a clustering tree based on the feature extraction model and the candidate images, generating target sample pairs comprising image and feature pairs, and training the feature extraction model using these pairs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If self-supervised learning based on contrastive learning is used, then local semantics can be learned from different sample images of the same granularity, but the features of different images become far apart in the feature space and hierarchical semantic structure cannot be modeled
Solution Approach 1:
The patent segments the feature learning process into multiple hierarchical levels by constructing a clustering tree with different granularity layers. Each layer learns semantics at a specific level, from fine-grained local features to coarse-grained hierarchical structures, resolving the contradiction between local semantics accuracy and hierarchical modeling capability
Solution Approach 2:
The patent adds a hierarchical dimension to the feature space by organizing features into a multi-level clustering tree structure. This transforms the flat feature representation into a hierarchical one, enabling the model to capture both local semantics and global semantic relationships simultaneously
2Ease of operation
If contrastive learning processes each sample image with various differences uniformly, then the learning process is simple, but it is difficult to apply to higher-level scenarios requiring hierarchical semantics
Solution Approach 1:
The patent makes the learning process dynamic by adapting the granularity level according to the hierarchical position in the clustering tree. Different layers process images at different granularities, allowing the system to maintain simplicity at each level while achieving versatility across hierarchical levels through the dynamic adjustment of processing granularity
Data Source
AI summary
The disclosure relates to a method for generating a feature extraction model, and a method and apparatus for image feature extraction, the method comprising: obtaining a plurality of candidate images; determining a clustering tree corresponding to the plurality of candidate images based on a feature extraction model and the plurality of candidate images, wherein the clustering tree comprises clusters at a plurality of layers; generating a target sample pair based on the plurality of candidate images and the clustering tree, wherein the target sample pair comprises an image sample pair and a feature sample pair, the image sample pair is formed based on two different candidate images, and the feature sample pair is formed based on features of the candidate image and features of a cluster center of the clustering; and training the feature extraction model based on the target sample pair.