Object Detection Weight Redistribution for Multi-Unit Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection systems face accuracy issues when using multiple recognition devices with different performance levels, leading to potential lowering of final result accuracy due to varying recognition device accuracies.
Innovation Solution
An object detection device is developed, comprising multiple object detection units that output scores for partial image regions, a weight computation unit to merge scores using weight parameters, a weight redistribution unit to adjust weights, and a parameter correction unit to optimize weights and reduce loss, ensuring accurate object detection across different recognition devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple object detection units with different accuracies are used, then the coverage and versatility of object detection is improved, but the final detection accuracy may be lowered due to varying performance levels
Solution Approach 1:
The patent dynamically adjusts the weight parameters of each object detection unit based on their individual accuracies and the characteristics of the input image. By changing the weight parameters adaptively, the system can emphasize the outputs of more accurate detectors while reducing the influence of less accurate ones, thus resolving the contradiction between using diverse detectors and maintaining high accuracy
Solution Approach 2:
The weight computation unit dynamically computes optimal weights for each object detection unit based on the input image characteristics and detector performance levels. This dynamic adjustment allows the system to adaptively optimize the contribution of each detector, ensuring high accuracy while maintaining versatility
2Measurement precision
If weights are computed and redistributed for multiple object detection units, then the accuracy is improved, but the computational complexity increases
Solution Approach 1:
The system pre-computes accuracy metrics for each object detection unit and stores them for quick reference during weight computation. This preliminary preparation reduces the computational burden during actual detection, as the weight computation can leverage pre-analyzed detector performance rather than re-evaluating from scratch
Solution Approach 2:
The weight computation unit acts as an intermediary that simplifies the complex task of integrating multiple detector outputs. By introducing this intermediate layer that computes optimal weights based on detector accuracies and image characteristics, the system manages the complexity of combining multiple detectors while achieving improved accuracy
Data Source
AI summary
In the object detection device, the plurality of object detection units output a score indicating a probability that a predetermined object exists, for each partial region set with respect to image data inputted. The weight computation unit computes a weight for each of the plurality of object detection units by using weight computation parameters and based on the image data. The weights are used when the scores outputted by the plurality of object detection units are merged. The weight redistribution unit changes the weight for a predetermined object detection unit, among the weights computed by the weight computation unit, to 0 and output the weights. The merging unit merges the scores outputted by the plurality of object detection units for each of the partial regions, by using the weights computed by the weight computation unit and including the weight changed by the weight redistribution unit. The loss computation unit computes a difference between a ground truth label of the image data and the merged score merged by the merging unit as a loss. Then, the parameter correction unit corrects the weight computation parameters so as to reduce the loss.