Image Instance Clustering for Similar-Object Training Datasets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems for generating training image datasets suffer from inaccuracies, inefficiencies, and limited functionality, particularly in accurately and efficiently creating datasets with visually and semantically similar objects for machine learning models.
Innovation Solution
An instance extraction system that utilizes stratified sampling, metadata analysis, and content and color embeddings to intelligently group similar objects into clusters, deduplicate them, and build machine learning models using these clusters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional crowd-sourcing systems are used to classify and label digital images, then human feedback can be obtained for model training, but the process suffers from inaccuracies and inefficiencies in generating comprehensive datasets with visually and semantically similar objects
Solution Approach 1:
The system performs self-service by automatically generating training datasets through unsupervised clustering of image embeddings, eliminating the need for human crowd-sourcing while maintaining high accuracy in grouping visually and semantically similar objects
Solution Approach 2:
The patent replaces the mechanical human labeling process with an automated computational system that uses neural network embeddings and clustering algorithms to classify and group images, significantly improving both accuracy and efficiency
2Adaptability or versatility
If traditional image dataset generation methods are used, then some object classes can be captured, but the system lacks functionality to efficiently create comprehensive datasets with multiple unique classes of visually and semantically similar objects
Solution Approach 1:
The system achieves universality by creating a multi-functional pipeline that simultaneously handles image embedding generation, clustering, deduplication, and dataset creation, enabling comprehensive processing of diverse object classes with a single integrated system
Solution Approach 2:
The patent applies segmentation by dividing the complex task of dataset generation into distinct modular components: image embedding generation, clustering algorithm execution, duplicate detection, and final dataset assembly, making the system more manageable and adaptable
3Reliability
If comprehensive image datasets with multiple object classes are generated, then machine learning model training can be improved, but the computing resources and processing time required increase significantly
Solution Approach 1:
The system uses partial action by generating embeddings only for salient objects detected in images rather than processing entire images, and by using unsupervised clustering on sampled data, reducing computing resources while maintaining model training quality
Solution Approach 2:
The patent applies preliminary action by pre-computing and storing image embeddings in a database before clustering, allowing the clustering process to work with pre-processed features rather than raw images, significantly reducing the computational burden during dataset generation
Data Source
AI summary
Methods, systems, and non-transitory computer readable media are disclosed for accurately and efficiently generating groups of images portraying semantically similar objects for utilization in building machine learning models. In particular, the disclosed system utilizes metadata and spatial statistics to extract semantically similar objects from a repository of digital images. In some embodiments, the disclosed system generates color embeddings and content embeddings for the identified objects. The disclosed system can further group similar objects together within a query space by utilizing a clustering algorithm to create object clusters and then refining and combining the object clusters within the query space. In some embodiments, the disclosed system utilizes one or more of the object clusters to build a machine learning model.