Object Recognition System with Connection Switch for Labeling Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The high human cost and data loss associated with manually labeling datasets for neural network-based video analysis systems, particularly in multi-task scenarios, where the need for diverse and large datasets increases the workload and risk of errors.
Innovation Solution
An object recognition system that includes a storage unit for images and labels, an estimator with an object feature extraction unit and task-specific identification units, and a learning control unit that adjusts connections between these units based on task types, reducing the need for extensive labeling by utilizing a connection switch and update degree storage to manage parameter updates and task losses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling is performed for multi-task neural network datasets, then recognition accuracy is improved, but human cost and time consumption increase significantly
Solution Approach 1:
The system performs self-labeling by automatically generating labels through the neural network's own prediction capabilities. The network processes images and generates labels without human intervention, allowing the system to serve itself in the labeling process, thus eliminating the need for manual labeling while maintaining accuracy improvement through multi-task learning
Solution Approach 2:
The system performs preliminary labeling actions by pre-processing images through the neural network to generate labels before actual recognition tasks. This preliminary action creates a labeled dataset that can be used for training, eliminating the need for subsequent manual labeling efforts while establishing the foundation for accurate recognition
2Measurement precision
If manual labeling is performed for multi-task neural network datasets, then recognition accuracy is improved, but human cost increases
Solution Approach 1:
The system performs self-labeling by automatically generating labels through the neural network's own prediction capabilities. The network processes images and generates labels without human intervention, allowing the system to serve itself in the labeling process, thus eliminating the need for manual labeling while maintaining accuracy improvement through multi-task learning
Solution Approach 2:
The system replaces the mechanical human labeling process with an automated computational system. The neural network automatically generates labels by processing images through its multi-task architecture, substituting human manual work with machine-based automation, thereby eliminating human costs while maintaining the ability to improve recognition accuracy
3Quantity of substance
If extensive manual labeling is performed, then data completeness is improved, but data loss due to human error increases
Solution Approach 1:
The system performs self-labeling by automatically generating labels through the neural network's own prediction capabilities. The network processes images and generates labels without human intervention, allowing the system to serve itself in the labeling process, thus eliminating the need for manual labeling while maintaining accuracy improvement through multi-task learning
Solution Approach 2:
The system replaces the mechanical human labeling process with an automated computational system. The neural network automatically generates labels by processing images through its multi-task architecture, substituting human manual work with machine-based automation, thereby eliminating human costs while maintaining the ability to improve recognition accuracy
Data Source
AI summary
In a data driven object recognition system and object recognition method, a connection relationship between an object feature extraction unit and a plurality of task-specific identification units is stored in a connection switch according to a type of task. The connection relationship is changed based on the connection information to suppress the amount of labeling in constructing a learning data set.