Detection Score Map Quality Screening for Segmentation Output

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing deep learning-based detection and segmentation algorithms often fail to produce acceptable results when analyzing data far from their training dataset distribution, leading to reduced user confidence in the tools utilizing these algorithms.

Innovation Solution

A system and method that automatically detects the quality of detection and segmentation algorithm outputs by analyzing intermediate outputs at multiple operating points, using a detection score map to compute features, and applying a classifier to predict whether the final output meets a detection precision threshold, preventing poor-quality outputs from being shown to end users.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If deep learning-based detection and segmentation algorithms are applied to data far from training dataset distribution, then the algorithms can process diverse inputs, but the output quality deteriorates and becomes unreliable

Engineering Contradiction:
Improveability to process diverse inputsVSAvoidoutput quality
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The system performs preliminary quality assessment of algorithm outputs before presenting them to users. By computing features from detection score maps and using a trained classifier to predict quality metrics, the system proactively identifies and filters poor-quality outputs before they reach the end user, preventing unreliable results from being displayed.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system implements a feedback mechanism where the quality assessment results are fed back into the workflow to control output display. The classifier's predictions about detection precision and other quality metrics provide feedback that determines whether algorithm outputs should be shown to users, creating a closed-loop quality control system.

Inventive Principle:
Principle #23Feedback

2Reliability

If algorithm outputs are automatically quality-assessed before display, then user confidence and reliability improve, but processing time and system complexity increase

Engineering Contradiction:
Improveoutput quality assuranceVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The quality assessment system is segmented into distinct functional components: feature computation from detection score maps, classification model inference for quality prediction, and decision logic for output filtering. This modular segmentation allows each component to be optimized independently and integrated into existing workflows without requiring complete system redesign.

Inventive Principle:
Principle #1Segmentation

3Manufacturing precision

If quality assessment is performed on all algorithm outputs, then only high-quality results are displayed, but processing time increases

Engineering Contradiction:
Improvedetection precision threshold complianceVSAvoidprocessing throughput
Core Design Contradiction:
Manufacturing precisionVSProductivity

Solution Approach 1:

The system performs quality assessment at multiple operating points by computing features from detection score maps and evaluating multiple quality metrics (detection precision, false positive rate, recall) rather than a single metric. This partial assessment approach focuses computational resources on the most critical quality dimensions, achieving sufficient quality control without exhaustive analysis of all possible parameters.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP4434003B1Detecting unacceptable detection and segmentation algorithm output
Publication Date: 2026.03.18 INTERNATIONAL BUSINESS MACHINE CORPORATION
  • EP4434003B1 patent drawingFigure 1
  • EP4434003B1 patent drawingFigure 2
  • EP4434003B1 patent drawingFigure 3

AI summary

In an approach for automatically detecting whether an output of a detection and segmentation algorithm is of an acceptable quality, a processor receives an image. A processor applies a detection stage of a detection and segmentation algorithm to the image. A processor computes a set of features from a detection score map output by the detection stage of the detection and segmentation algorithm by analyzing the detection score map at more than one different operating points. A processor inputs the set of features into a classifier that predicts whether a final output of the detection and segmentation algorithm will be of an acceptable quality, wherein the acceptable quality is defined based on whether a detection precision threshold has been reached. A processor receives an output of the classifier.