Interactive Scene Graph Search With GNN Node Alignment

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing graph matching algorithms are inefficient for interactive queries from humans and struggle with accurate one-to-one node correspondence, especially in large databases, making it difficult to visualize and interpret subgraph patterns effectively.

Innovation Solution

A system utilizing Graph Neural Networks (GNNs) for encoding topological and node attributes in graphs as fixed-length vectors, employing NeuroMatch for subgraph matching and NeuroAlign for node alignment, enabling efficient and accurate one-to-one node correspondence visualization.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional graph matching algorithms are used for subgraph pattern search, then completeness of matching can be achieved, but computational efficiency deteriorates significantly in large databases

Engineering Contradiction:
Improvematching accuracyVSAvoidsearch speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent replaces traditional mechanical graph matching algorithms with a neural network-based system. The GNN encoder learns graph representations and the similarity calculator computes matching scores through neural network operations, substituting the step-by-step mechanical matching process with parallel neural computation that achieves both accuracy and efficiency.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent transforms the graph matching problem from discrete structural comparison to continuous vector space comparison. By encoding graphs as continuous vectors through GNN and computing similarity in this transformed parameter space, the system achieves efficient computation while maintaining matching accuracy.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If detailed node-level analysis is performed for subgraph matching, then node correspondence accuracy improves, but computational complexity increases

Engineering Contradiction:
Improvenode correspondence accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent performs preliminary graph encoding into fixed-length vectors using GNN before the actual matching process. This pre-computation of graph representations allows the system to work with compact vectors rather than full graph structures during matching, reducing computational complexity while preserving node-level correspondence information.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces graph embedding vectors as an intermediary representation between the original graph structure and the matching process. These vectors serve as a compressed intermediary that captures essential structural and attribute information, enabling efficient comparison without directly processing the full graph complexity.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Ease of operation

If interactive query interface with visual feedback is provided, then user ease of operation improves, but system complexity increases

Engineering Contradiction:
Improveuser interaction easeVSAvoidsystem complexity
Core Design Contradiction:
Ease of operationVSDevice complexity

Solution Approach 1:

The patent creates visual copies and representations of the query graph and matching results through graphical user interface elements. The system generates visual feedback that replicates the graph structure and highlighting information, allowing users to interact with intuitive visual representations rather than complex data structures directly.

Inventive Principle:
Principle #26Copying

Data Source

PatentEP4152177B1Systems and methods for interactive image scene graph pattern search and analysis
Publication Date: 2026.01.21 ROBERT BOSCH GMBH
  • EP4152177B1 patent drawingFigure 1
  • EP4152177B1 patent drawingFigure 2~3
  • EP4152177B1 patent drawingFigure 4

AI summary

Methods and systems for providing an interactive image scene graph pattern search are provided. A user is provide with an image having a plurality of selectable segmented regions therein. The user selects one or more of the segmented regions to build a query graph. Via a graph neural network, matching target graphs are retrieved that contain the query graph from a target graph database. Each matching target graph has matching target nodes that match with the query nodes of the query graph. Matching target images from an image database are associated with the matching target graphs. Embeddings of each of the query nodes and the matching target nodes are extracted. A comparison of the embeddings of each query node with the embeddings of each matching target node is performed. The user interface displays the matching target images that are associated with the matching target graphs.