Hierarchical Domain-Based Object Recognition Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning systems trained with generic annotations face challenges in providing accurate and efficient object recognition when tasked with finer sub-classifications, as they require extensive manual annotation and can become resource-intensive and performance-degraded due to increased dataset size and complexity.
Innovation Solution
The implementation of a hierarchical domain-based object recognition service that uses sequential or parallel processing with domain-based inference models to identify objects at both general and sub-classification levels, reducing the need for human resources through automated annotation and normalization of results using a sliding window analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If machine learning systems use generic annotations for training, then training efficiency is improved, but classification precision deteriorates when finer sub-classifications are required
Solution Approach 1:
The patent segments the classification task into multiple hierarchical levels (e.g., general category, sub-category, specific type). Each level is handled by a separate machine learning model trained on annotations of appropriate granularity. This allows the system to maintain high training efficiency at coarser levels while achieving high classification precision at finer levels, resolving the contradiction between training efficiency and classification precision.
2Measurement precision
If manual annotation is performed to achieve finer sub-classifications, then classification precision is improved, but time consumption and resource usage increase
Solution Approach 1:
The patent applies preliminary action by first training models on coarser-grained annotations to establish general classifications. These pre-trained models then guide the annotation process for finer sub-classifications, reducing the need for extensive manual annotation at each level. The hierarchical structure allows the system to leverage previously performed annotation work, significantly reducing time consumption while maintaining high classification precision.
3Adaptability or versatility
If the dataset size is increased to cover more sub-classifications, then classification coverage is improved, but system complexity and performance degradation occur
Solution Approach 1:
The patent segments the large-scale classification problem into multiple manageable hierarchical levels, with each level handled by a dedicated model. This segmentation allows the system to achieve broad classification coverage across numerous sub-classes without requiring a single monolithic model, thereby controlling system complexity while improving adaptability.
Solution Approach 2:
The patent introduces a hierarchical dimension to the classification structure, organizing classes into multiple levels (e.g., parent categories, child categories, specific types). This dimensional organization allows the system to scale to cover more sub-classifications by adding hierarchical depth rather than expanding a flat classification structure, which would otherwise increase system complexity linearly.
Data Source
AI summary
A method, a device, and a non-transitory storage medium each provide an object recognition service that identifies an object within an image or a video. An object is identified according to a general classification based on a domain-based inference model, and is subsequently further identified according to a sub-classification of the general classification based on another domain-based inference model. The domain-based inference models are hierarchical. The object recognition of the object may be used in support of end user services.