Semantic Segmentation Validity Evaluation for Reliable Image Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The accuracy of semantic segmentation by trained models is not 100%, leading to potential misclassification, and users often focus on statistical information without verifying the validity of the segmentation, risking incorrect conclusions.
Innovation Solution
An information processing apparatus that includes an evaluation unit to assess the validity of semantic segmentation based on image feature values, displays warnings for unreliable results, and allows for correction and re-training of the model using corrected output images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multiple applications are operated simultaneously in a smartphone, then functionality and user productivity are improved, but memory resources are consumed and input methods may be mistakenly switched
Solution Approach 1:
The patent segments the input method control by creating application-specific input method settings. Each application can have its own designated input method, allowing the system to manage multiple applications simultaneously without global input method conflicts. This segmentation enables precise control over which input method is active in which application context.
Solution Approach 2:
The patent implements local quality by allowing different input methods to be assigned to different applications based on their specific needs. Instead of a uniform input method across all applications, the system provides localized input method configurations for each application, optimizing input experience while preventing unwanted input method switches in specific contexts.
2Adaptability or versatility
If input method switching is enabled for application transitions, then input flexibility is improved, but erroneous input method switching occurs due to ambiguous timing
Solution Approach 1:
The patent applies preliminary action by determining the input method for the target application before the actual application transition occurs. The system identifies which application will become active and pre-configures its designated input method, ensuring that the correct input method is ready and prevents erroneous switching during the transition process.
Solution Approach 2:
The patent implements feedback mechanisms to monitor and confirm application transition events. By detecting whether an application transition has actually occurred and verifying the target application's identity, the system can accurately determine whether input method switching is appropriate, providing feedback control to prevent erroneous switches.
3Adaptability or versatility
If input method is determined by application transition events, then input method adaptability is improved, but processing complexity increases due to event detection and timing judgment
Solution Approach 1:
The patent enables applications to self-identify and declare their own input method requirements. Each application can specify its designated input method in its configuration, eliminating the need for the system to complexly analyze transition events and determine appropriate input methods. The application itself provides the information needed for input method selection.
Solution Approach 2:
The patent introduces an intermediary configuration layer that stores application-specific input method assignments. This intermediary structure acts as a lookup table or registry that maps applications to their designated input methods, simplifying the decision-making process by providing direct access to pre-configured input method assignments without complex event analysis.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An information processing apparatus includes an acquisition unit that acquires an output image output from a trained model as a result of causing the trained model to perform semantic segmentation in which discrimination of a class which is a type of an object appearing in an input image is performed on a pixel-by-pixel basis, an evaluation unit that evaluates validity of the semantic segmentation based on the output image, and a display controller that performs control such that an evaluation result indicating that the semantic segmentation does not have validity is displayed in a case where the evaluation unit evaluates that the semantic segmentation does not have validity.