3D Model Projection for Multi-View Image Segmentation Annotation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image segmentation annotation methods face challenges in improving annotation quality and efficiency, particularly for irregularly shaped objects, due to human subjectivity and high workload in manual annotation, and the limitations of man-machine collaborative methods.

Innovation Solution

Construct a three-dimensional model of an object based on a multi-view picture set, adjust its position or orientation to align the projection with the image, and annotate the aligned projection as a segmentation result.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If full-manual annotation is used to annotate irregularly shaped objects, then annotation quality can be improved, but annotation time and workload increase significantly

Engineering Contradiction:
Improveannotation qualityVSAvoidannotation time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent creates a three-dimensional model as a digital copy of the object based on multi-view images, then projects this model onto the target image to generate annotation boundaries. This copying approach replaces time-consuming manual tracing while maintaining consistent and accurate boundaries across different views.

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent transitions from two-dimensional manual annotation to three-dimensional modeling. By constructing a 3D model from multi-view images and projecting it back to 2D, the system achieves automated boundary generation that is both time-efficient and consistent across different object orientations and views.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Stability of the object's composition

If manual annotation is used to ensure boundary uniformity across different views, then annotation quality is improved, but the workload and complexity increase

Engineering Contradiction:
Improveboundary uniformityVSAvoidannotation complexity
Core Design Contradiction:
Stability of the object's compositionVSDevice complexity

Solution Approach 1:

The three-dimensional model serves as a universal representation that can be projected onto any view of the object. This single 3D model handles multiple views and orientations, ensuring boundary uniformity across all perspectives without requiring separate manual annotation for each view.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent performs preliminary three-dimensional reconstruction from multi-view images before annotation is needed. This pre-established 3D model can then be quickly projected onto any target image, eliminating the need for repeated manual boundary drawing and ensuring consistency across different views.

Inventive Principle:
Principle #10Preliminary action

3Ease of operation

If man-machine collaborative annotation is used to reduce manual burden, then some automation is achieved, but the annotation efficiency remains limited due to required manual refining

Engineering Contradiction:
Improvemanual burdenVSAvoidannotation efficiency
Core Design Contradiction:
Ease of operationVSProductivity

Solution Approach 1:

The system performs automated three-dimensional reconstruction and projection without requiring manual intervention for boundary tracing. The algorithm automatically processes multi-view images, constructs the 3D model, and generates annotation boundaries, achieving full automation that significantly improves annotation efficiency.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP4700702A1Image annotation method and apparatus, and storage medium
Publication Date: 2026.02.25 BEIJING WODONG TIANJUN INFORMATION TECH CO LTD
  • EP4700702A1 patent drawingFigure 1~2
  • EP4700702A1 patent drawingFigure 3
  • EP4700702A1 patent drawingFigure 4

AI summary

The present disclosure relates to the field of image processing, and provides an image annotation method and apparatus, and a storage medium. The method comprises: on the basis of a multi-view picture set of an object, constructing a three-dimensional model of the object; adjusting the pose of the three-dimensional model of the object, so that the projection of the three-dimensional model of the object on a picture to be annotated completely overlaps an image of the object in the picture to be annotated; and annotating the projection when said projection completely overlaps said image as a segmentation result of the image of the object in the picture to be annotated. By adjusting the pose of the object in the picture to replace the annotation of dense point positions (hundreds and thousands of point positions) of the boundary of an object in one picture, annotation efficiency is improved. By constructing the three-dimensional model of the object, and using the projection of the unified three-dimensional model in different poses to annotate the object in the picture, the uniformity of the boundaries of the object at different viewing angles can be ensured, thereby improving annotation quality. Especially in a multi-view image segmentation scenario, higher annotation efficiency and annotation quality are achieved.