Object Recognition Model Updating via Category Center Parameters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object recognition models face inefficiencies when adding new tasks, as re-training is required, which affects the recognition performance of original tasks and is time-consuming.
Innovation Solution
An object recognition model updating method that determines and updates category center parameters based on sample features from both original and new tasks, allowing the model to recognize new tasks without re-training, thereby maintaining original task performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the object recognition model is re-trained to add a new object recognition task, then the model can recognize the new task, but the recognition effect of the original task is affected and the training time is increased
Solution Approach 1:
The patent segments the model update process into two independent parts: (1) extracting category center parameters from new task samples without re-training the entire model, and (2) updating only the category center parameters while keeping the original model weights intact. This segmentation allows the model to acquire new task capabilities without compromising original task performance.
Solution Approach 2:
The patent changes the approach from re-training the entire model to updating only specific parameters (category center parameters). By extracting new category center parameters from the new task samples and merging them with the existing model's category center parameters, the model adapts to new tasks without altering the weights that govern original task recognition, thus maintaining reliability.
2Adaptability or versatility
If the object recognition model is re-trained to add a new object recognition task, then the model can recognize the new task, but the training time is increased
Solution Approach 1:
The patent extracts only the necessary information (category center parameters) from the new task samples without processing the entire training dataset through the full model. By extracting features and computing category center parameters directly from samples, the system avoids the time-consuming process of re-training the entire model while still achieving the goal of adding new task recognition capability.
Solution Approach 2:
The patent performs preliminary extraction of category center parameters from new task samples before integrating them into the model. This preliminary action prepares the necessary parameters in advance, allowing for rapid model update without the need for time-consuming re-training procedures.
3Reliability
If the original object recognition model is used without updating, then the recognition effect of original tasks is maintained, but the model cannot recognize new tasks
Solution Approach 1:
The patent makes the model multi-functional by enabling it to perform both original tasks and new tasks simultaneously. The updated model structure allows a single model to handle multiple object recognition tasks by incorporating category center parameters from different tasks, eliminating the need to choose between maintaining original performance and acquiring new capabilities.
Data Source
AI summary
An object recognition model updating method includes: obtaining a first object recognition model through training based on first training sample sets of first object recognition tasks, wherein the first training sample sets include first image samples of first objects of first categories in the first object recognition tasks; determining a first category center parameter of the first object recognition model based on first sample features of the first image samples; obtaining a second training sample set of a second object recognition task including second image samples of second objects of second categories in the second object recognition task; determining a second category center parameter based on the first sample features and second sample features of the second image samples; and updating the first category center parameter to the second category center parameter, to obtain a second object recognition model.