Contour Gap Closure for Accurate Instance Delineation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing semantic segmentation models struggle to accurately delineate individual objects when they overlap or are adjacent, often requiring domain-specific enhancements and postprocessing, which are not generic and can be imperfect.
Innovation Solution
A Closed Contour Generic Model (CCGM) and an Object Delineation System (ODS) that includes preprocessing and postprocessing modules to close gaps in contours, using a semantic segmentation model trained on diverse images with incomplete contours, to create closed contours and distinct object segments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If standard semantic segmentation models are used to classify pixels by category, then pixel classification accuracy is improved, but the ability to separate overlapping or touching objects deteriorates
Solution Approach 1:
The patent applies segmentation by dividing the pixel classification task into two distinct stages: first, semantic segmentation to classify pixels by category; second, contour-based instance segmentation to separate individual objects. This two-stage segmentation approach resolves the contradiction by handling category classification and instance delineation as separate processing steps, allowing each to optimize for its specific goal.
Solution Approach 2:
The patent introduces an intermediary element - the contour map - that bridges semantic segmentation and instance segmentation. The contour map serves as a mediator that captures object boundaries and guides the separation of overlapping instances, enabling accurate instance delineation while preserving the category classification accuracy achieved by semantic segmentation models.
2Manufacturing precision
If postprocessing procedures are applied to separate overlapping objects, then instance separation capability is improved, but system complexity increases
Solution Approach 1:
The patent applies preliminary action by generating the contour map during the semantic segmentation training phase, before instance separation is needed. The contour map is pre-computed and stored, allowing rapid instance segmentation without requiring complex postprocessing procedures. This approach reduces system complexity by preparing separation guidance in advance rather than adding complex real-time processing.
Solution Approach 2:
The patent uses copying by creating a contour map that replicates object boundary information from the semantic segmentation output. This contour map serves as a simplified copy or representation of the scene's geometric structure, enabling instance separation through straightforward contour-based operations rather than complex iterative postprocessing.
3Manufacturing precision
If domain-specific enhancements are made to semantic segmentation models, then performance on specific tasks is improved, but adaptability to different domains deteriorates
Solution Approach 1:
The patent applies universality by designing a domain-agnostic instance segmentation approach that works across different object types and datasets. The contour map generation and contour-based instance segmentation methodology is universally applicable regardless of the specific domain, allowing the same system to handle diverse tasks without domain-specific calibration while maintaining high performance through the generalizable contour representation.
Data Source
AI summary
A computer-implemented method includes receiving an initial contour image comprising contour information of a plurality of objects, the contour information contains one or more gaps thereby constituting a first number of gaps, utilizing a machine learning model trained to close gaps in contours to close at least one gap in the initial contour image and creating a closed contour image comprising contour information of the plurality of objects where a quantity of gaps in the closed contour image is smaller than the first number. A system includes an object delineation system (ODS) with a Closed Contour Generic Model (CCGM) machine learning model trained to close gaps in contours configured to receive an initial contour image that includes contour information with one or more gaps and utilize the CCGM to close at least one gap and create a closed contour image including contour information with less gaps.