Object Detection False Positive Suppression via Feature Gallery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep neural networks used in object detection often produce false positive results, which can be critical in applications like medical diagnostics and autonomous vehicle operation, where accuracy is paramount.
Innovation Solution
A human-in-the-loop review process is implemented to identify and suppress false positives by creating a gallery of feature representations from incorrectly detected objects, allowing the system to match and exclude similar detections during inference without retraining the model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If a deep neural network object detection model is used to detect objects in images, then the model can automatically classify objects, but the model produces false positive results where it incorrectly classifies objects
Solution Approach 1:
The system implements a feedback mechanism where human reviewers verify detected objects and provide corrections. These corrections are fed back into the system to update the gallery of false positive feature representations, allowing the system to learn from human feedback and improve its detection accuracy over time without requiring complete model retraining
Solution Approach 2:
The patent introduces an intermediary gallery structure that stores feature representations of false positive objects. This gallery acts as a mediator between the detection model and the final output, allowing the system to compare detected objects against known false patterns and filter out false positives before presenting results to the user
2Productivity
If the object detection model returns all detected objects with bounding boxes and probability scores, then the system provides comprehensive detection results, but the system presents false positive results to the user
Solution Approach 1:
The system performs preliminary action by pre-computing and storing feature representations of false positive objects in a gallery before actual detection occurs. During detection, the system uses this pre-prepared gallery to quickly identify and filter false positives, maintaining comprehensive detection results while improving accuracy through pre-established false positive patterns
3Reliability
If the system retrain the model to reduce false positives, then the accuracy improves, but the process requires significant time and computational resources
Solution Approach 1:
The patent extracts only the relevant information needed to reduce false positives by isolating and storing only the feature representations of false positive objects in the gallery. This extraction approach allows the system to address the false positive problem without retraining the entire model, significantly reducing time and computational resources required while maintaining improved accuracy
Data Source
AI summary
Disclosed herein is a system and method for reducing false positives in object detection frameworks. A human form view of objects detected by the object detection framework and indicates the object is a false positive. When an indication of a false positive been received, a feature representation of displayed object stored in the gallery. During an inference or testing phase, the gallery is searched for a feature representation matching the feature representation of the detected objects, and, if a match is found, the detected object is deemed to be a false positive and is not displayed to the user.