Deformable CNN Segmentation Training With Reduced Pixel Annotation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The high cost and low efficiency of manual pixel-level annotation in training semantic image segmentation networks, particularly for images with high resolutions, result in excessive training costs and inefficiencies.

Innovation Solution

A method for training a semantic image segmentation network using a deformable convolutional neural network and an offset network, allowing image-level annotation to reduce the need for manual pixel-level annotation, by alternately training an image classification network and an offset network to determine model parameters based on classification loss functions, thereby obtaining a semantic image segmentation model.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If manual pixel-level annotation is performed for training semantic image segmentation networks, then the model performance is improved, but the annotation cost and time consumption increase excessively

Engineering Contradiction:
Improvemodel performanceVSAvoidannotation time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The method performs preliminary image-level classification to generate candidate content regions before conducting pixel-level segmentation. This preliminary action provides a rough guide for subsequent detailed annotation, reducing the overall annotation time while maintaining segmentation accuracy.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The annotation process is segmented into two stages: first, image-level classification to identify content regions; second, pixel-level segmentation within those regions. This segmentation of the annotation process allows efficient use of manual effort by focusing detailed work only where needed.

Inventive Principle:
Principle #1Segmentation

2Reliability

If manual pixel-level annotation is performed for training semantic image segmentation networks, then the model performance is improved, but the annotation cost increases excessively

Engineering Contradiction:
Improvemodel performanceVSAvoidannotation cost
Core Design Contradiction:
ReliabilityVSEase of manufacture

Solution Approach 1:

The method performs preliminary image-level classification to generate candidate content regions before conducting pixel-level segmentation. This preliminary action provides a rough guide for subsequent detailed annotation, reducing the overall annotation time while maintaining segmentation accuracy.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The annotation process is segmented into two stages: first, image-level classification to identify content regions; second, pixel-level segmentation within those regions. This segmentation of the annotation process allows efficient use of manual effort by focusing detailed work only where needed.

Inventive Principle:
Principle #1Segmentation

3Ease of manufacture

If image-level annotation is used for training, then the annotation cost is reduced, but the model performance may be compromised

Engineering Contradiction:
Improveannotation costVSAvoidmodel performance
Core Design Contradiction:
Ease of manufactureVSReliability

Solution Approach 1:

The method performs preliminary image-level classification to generate candidate content regions before conducting pixel-level segmentation. This preliminary action provides a rough guide for subsequent detailed annotation, reducing the overall annotation time while maintaining segmentation accuracy.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The annotation process is segmented into two stages: first, image-level classification to identify content regions; second, pixel-level segmentation within those regions. This segmentation of the annotation process allows efficient use of manual effort by focusing detailed work only where needed.

Inventive Principle:
Principle #1Segmentation

4Manufacturing precision

If full-image classification is performed to obtain segmented supervised information, then the segmentation can be achieved, but the training efficiency remains low due to high annotation requirements

Engineering Contradiction:
Improvesegmentation accuracyVSAvoidtraining efficiency
Core Design Contradiction:
Manufacturing precisionVSProductivity

Solution Approach 1:

The method performs preliminary image-level classification to generate candidate content regions before conducting pixel-level segmentation. This preliminary action provides a rough guide for subsequent detailed annotation, reducing the overall annotation time while maintaining segmentation accuracy.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The annotation process is segmented into two stages: first, image-level classification to identify content regions; second, pixel-level segmentation within those regions. This segmentation of the annotation process allows efficient use of manual effort by focusing detailed work only where needed.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentUS12579443B2Training semantic image segmentation model comprising deformable convolutional neural network
Publication Date: 2026.03.17 TENCENT TECHNOLOGY (SHENZHEN) CO LTD
  • US12579443B2 patent drawing
  • US12579443B2 patent drawing
  • US12579443B2 patent drawing

AI summary

A method for training an image classification model includes obtaining first prediction class annotation information of a first image by using an image classification network based on a first model parameter of an offset network being fixed; determining a second model parameter corresponding to the image classification network by using a classification loss function based on the image content class information and the first prediction class annotation information; obtaining second prediction class annotation information of the first image by using the offset network based on the second model parameter of the image classification network being fixed; determining a third model parameter corresponding to the offset network by using the classification loss function based on the image content class information and the second prediction class annotation information; and training a semantic image segmentation network model based on the second model parameter and the third model parameter.