Unified Parameter Management for Multitask Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multitask recognition systems require multiple operation parameters for each combination of tasks, leading to an increase in total parameter quantity and inefficient execution when selectively executing specific recognition tasks.
Innovation Solution
A data processing apparatus with a storage unit for parameter groups, a selection unit to choose recognition tasks, a holding unit to store these parameters, and an execution unit to execute tasks sequentially using the stored parameters, allowing for efficient execution of multiple recognition tasks with a single set of operation parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple operation parameters are prepared for each combination of selected tasks, then task execution flexibility is improved, but total parameter quantity increases
Solution Approach 1:
The patent implements a unified parameter management system where a single set of operation parameters can serve multiple recognition tasks. The control unit selectively applies the same parameters across different tasks (object detection, classification, segmentation), eliminating the need for separate parameter sets for each task combination. This universal parameter approach maintains task execution flexibility while significantly reducing the total parameter quantity stored in memory.
2Adaptability or versatility
If software is replaced to execute a specific recognition task, then task specialization is improved, but execution efficiency deteriorates
Solution Approach 1:
The patent employs dynamic task selection within a fixed neural network architecture. Instead of replacing software for different tasks, the system dynamically configures which tasks are executed by controlling the inference process. The control unit enables or disables specific task outputs (detection, classification, segmentation) based on input images and requirements, allowing task specialization without software replacement. This dynamic approach maintains execution efficiency while achieving task specialization.
3Measurement precision
If multiple parameter groups are transferred to holding unit, then task execution accuracy is improved, but data transfer overhead increases
Solution Approach 1:
The patent extracts and eliminates redundant parameter transfers by implementing a single parameter loading mechanism. Instead of transferring multiple parameter groups for different tasks, the system loads one unified parameter set from memory to the holding unit (register/file). The control unit then routes this single parameter set to the appropriate tasks, completely removing the data transfer overhead associated with multiple parameter groups while preserving task execution accuracy.
Data Source
AI summary
A data processing apparatus includes a storage unit configured to store a plurality of types of parameter groups to be used in a plurality of types of recognition tasks, a selection unit configured to select two or more recognition tasks to be executed from among the plurality of types of recognition tasks, a holding unit configured to hold parameter groups, a transfer unit configured to transfer parameter groups to be used in the two or more recognition tasks in sequence from the storage unit to the holding unit, and an execution unit configured to execute the two or more recognition tasks in sequence using the parameter groups held in the holding unit.