Object Detection Abnormality Checks Using Saliency Inconsistency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection models struggle to detect various types of abnormalities, including partial occlusion, out-of-distribution, and adversarial patches, due to specialized algorithms being designed for individual types, making it difficult to identify abnormalities other than the targeted ones.
Innovation Solution
An abnormality detection function that measures the inconsistency between intermediate and final stages of the object detection pipeline using saliency maps and detection results, calculating values to quantify the effect of any type of abnormality on the decision-making process, enabling detection of any type of abnormality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If specialized algorithms are designed for each individual type of abnormalities, then detection precision for targeted abnormality types is improved, but adaptability to detect other types of abnormalities deteriorates
Solution Approach 1:
The patent applies universality by designing a single abnormality detection model that can detect multiple types of abnormalities (adversarial patches, occlusions, out-of-distribution objects) without requiring separate specialized algorithms for each type. The model uses a unified approach based on saliency map inconsistency to achieve multi-functional abnormality detection capability.
Solution Approach 2:
The patent changes the detection parameter from abnormality-specific features to a general inconsistency metric between saliency maps and detection results. By measuring whether salient regions consistently correspond to detected objects across different abnormality types, the system achieves broad adaptability while maintaining detection precision through a standardized measurement approach.
2Adaptability or versatility
If a unified approach is used to detect any type of abnormality, then adaptability is improved, but measurement precision for specific abnormality types may deteriorate
Solution Approach 1:
The patent employs feedback by using saliency maps (generated by XAI techniques) to verify and refine the detection results. The inconsistency measurement between saliency maps and detection outcomes provides a feedback mechanism that maintains high precision across different abnormality types by continuously checking whether the detected objects align with the model's attention regions.
3Reliability
If specialized detection methods are used for specific abnormalities, then detection reliability for targeted types is improved, but device complexity increases due to multiple algorithms
Solution Approach 1:
The patent reduces device complexity by replacing multiple specialized detection algorithms with a single unified abnormality detection model. This universal model maintains reliability for various abnormality types while simplifying the system architecture, reducing computational overhead, and eliminating the need to manage multiple separate detection pipelines.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An abnormality detection program causes a computer to execute a process including calculating a first value and a second value, based on a detection result of an object detected from an input scene to an object detection model and a saliency map for the input scene obtained by an XAI technique, the first value indicating a value of the saliency map for the entire input scene, the second value indicating a value of the saliency map in a region other than a region of the detected object in the input scene, and detecting an abnormality in the input scene based on the calculated first and second values.