Multi-tier Visual Search System with Ranked Object Model Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing visual search systems face challenges in returning accurate and timely results due to the large size of image databases, as current techniques are inefficient in processing and retrieving data from these databases.
Innovation Solution
A multi-tier visual search system is implemented, where object models are divided into sets based on rank and distributed across client devices, servers, and human crowd resources, allowing for local, server-level, and crowd-sourced recognition, optimizing data access and processing latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If visual search is performed on large image databases using current techniques, then comprehensive search coverage is achieved, but search time becomes too long to return results accurately and timely
Solution Approach 1:
The patent segments the large image database into multiple tiers based on object model rank and access frequency. Tier 1 contains high-ranking, frequently accessed object models stored locally on client devices, while Tier 2 contains lower-ranking models stored on servers. This segmentation enables the system to quickly search local storage for common objects without querying the entire database, significantly reducing search time while maintaining comprehensive coverage.
Solution Approach 2:
The patent implements local quality by storing different portions of the object database at different locations with different access characteristics. High-priority object models are cached locally on client devices for immediate access, while lower-priority models remain on remote servers. This creates a quality gradient where frequently needed data has faster access locally, resolving the contradiction between database size and search speed.
2Speed
If object models are distributed across multiple tiers, then search speed improves, but system complexity increases
Solution Approach 1:
The patent employs dynamic object model ranking that adapts based on usage patterns and feedback. The system continuously updates the rank of object models based on recognition success rates and user interactions, dynamically adjusting which models are promoted to Tier 1 local storage and which remain in Tier 2 server storage. This dynamic approach optimizes recognition speed over time without requiring manual configuration, managing system complexity through automated adaptation.
Solution Approach 2:
The patent introduces a tiered architecture as an intermediary between the user and the complete object database. Rather than directly querying the entire database or relying on a single storage location, the system uses a multi-tier intermediary structure with clear protocols for querying Tier 1 locally and falling back to Tier 2 when needed. This intermediary layer simplifies the complexity by providing a structured, rules-based approach to distributed search.
Data Source
AI summary
A system for multi-tier visual recognition comprises a processor and a memory. The processor is configured to calculate a rank for each of a plurality of object models. The plurality of object models are used for visual recognition. Each of the plurality of object models has a size. The processor is configured to divide the plurality of object models into a plurality of sets based at least in part on the rank for each of the plurality of object models, and provide a first set of the plurality of sets to a first tier and a second set of the plurality of sets to a second tier. The memory coupled to the processor and configured to provide the processor with instructions.