GAN Object Contour Labeling Accuracy via Image Segmentation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing weakly supervised learning methods for labeling and segmenting objects in images suffer from low accuracy in determining the contour of target objects.

Innovation Solution

A method and apparatus that utilize a generative adversarial network (GAN) to improve the accuracy of object contour labeling. The method involves acquiring target image features, inputting them into a target generator within the GAN, and using the generated mask to label the object contour.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If weakly supervised learning methods with image-level classification tags are used, then the labeling process is simplified, but the accuracy of object contour determination deteriorates

Engineering Contradiction:

Improvelabeling process simplicityVSAvoidobject contour determination accuracy

Core Design Contradiction:

Ease of operationVSMeasurement precision

Solution Approach 1:

The patent segments the image into multiple local regions and performs classification independently for each region. This allows the model to capture local object characteristics more effectively while maintaining the simplicity of image-level tags, thereby improving contour determination accuracy without complicating the labeling process

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces a spatial dimension by dividing the image into multiple regions and processing them separately. This transforms the traditional single-image-level classification into a multi-region, multi-scale classification system, enabling more precise localization and contour determination while keeping the input labels simple

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

2Device complexity

If existing weakly supervised learning methods are used, then the training process is simpler, but the segmentation accuracy of target objects deteriorates

Engineering Contradiction:

Improvetraining process complexityVSAvoidsegmentation accuracy

Core Design Contradiction:

Device complexityVSManufacturing precision

Solution Approach 1:

The patent performs preliminary region division and feature extraction before final classification. By pre-processing the image into multiple regions and extracting relevant features in advance, the model can focus computational resources on accurate segmentation, improving results without significantly increasing overall training complexity

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent employs a dynamic multi-scale analysis approach where the model adapts to different regions and objects within the image. This allows the segmentation process to be flexible and responsive to varying object characteristics, improving accuracy while maintaining a manageable training framework

Inventive Principle:

Principle #15Dynamics

Data Source

PatentUS12340603B2Method and apparatus for marking object outline in target image, and storage medium and electronic apparatus

Publication Date: 2025.06.24 SANECHIPS TECH CO LTD

AI summary

Disclosed are a method and apparatus for labeling an object contour in a target image, a non-transitory computer-readable storage medium and an electronic device. The method for labeling an object contour in a target image includes acquiring a target image feature of a target image, inputting the target image feature into a target generator, and acquiring a target mask of the target image generated by the target generator, wherein the target mask is used for labeling a contour of the target object.