Machine Learning Recognition Candidate Selection for Detection Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In object detection processing, when an object is not detected, it is difficult to analyze the detection result and improve the accuracy of the recognition model, as there is no available contribution information for object detection.
Innovation Solution
An information processing method that obtains recognition result candidates and their likelihoods from a machine learning-trained model, selects a candidate based on its relationship and likelihood to a designated part in the sensing data, and outputs this selected candidate for analysis, even if an object is not detected.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If object detection processing is performed using a trained recognition model, then detection accuracy is improved, but analysis capability is lost when no object is detected
Solution Approach 1:
The system performs analysis processing in advance by generating recognition result candidates and their contribution information even when the recognition model does not detect an object. This preliminary action ensures that contribution information is available before final detection decisions are made, allowing subsequent analysis of why detection failed or succeeded.
Solution Approach 2:
The system introduces recognition result candidates as an intermediary between the sensing data and the final detection result. These candidates include contribution information that mediates between the raw data and the detection outcome, enabling analysis even when no object is ultimately detected by providing intermediate results with explanatory power.
2Reliability
If only detected objects are used for analysis, then analysis reliability is improved, but analysis coverage deteriorates
Solution Approach 1:
The system performs analysis processing on recognition result candidates that may not be selected as final detection results. This partial action extends analysis coverage beyond only confidently detected objects to include borderline cases and near-misses, providing excessive analysis coverage that can be filtered or weighted by reliability metrics.
Solution Approach 2:
The system changes the parameter threshold by generating recognition result candidates with varying degrees of confidence scores. This allows the system to adjust between reliability and coverage by selecting candidates above different threshold levels, enabling flexible adaptation to different analysis needs without sacrificing either reliability or coverage completely.
3Loss of information
If recognition result candidates are generated and analyzed, then analysis capability is improved, but processing complexity increases
Solution Approach 1:
The system extracts only the necessary contribution information from the recognition processing pipeline - specifically, the contribution of each data value to the detection likelihood. This extraction approach avoids the complexity of analyzing entire recognition pipelines while still providing actionable contribution information for analysis purposes.
Solution Approach 2:
The system creates simplified copies of recognition results in the form of recognition result candidates that contain essential contribution information. These copies are lighter-weight representations that preserve the analytical value of full recognition outputs without requiring the full computational complexity of the original recognition model to be re-executed for analysis.
Data Source
AI summary
An information processing method performed by a computer includes: obtaining a plurality of recognition result candidates in sensing data and a likelihood of each of the plurality of recognition result candidates, the plurality of recognition result candidates and the likelihood being obtained by inputting the sensing data to a model that is trained by machine learning and performs recognition processing; obtaining an indication designating a part to be analyzed in the sensing data; selecting at least one recognition result candidate from the plurality of recognition result candidates, based on (i) a relationship between each of the plurality of recognition result candidates and the part and (ii) the likelihood of each of the plurality of recognition result, candidates; and outputting the at least one recognition result candidate that is selected.