arxiv_cv 95% Match Research Paper Computer Vision Researchers,Sports Analysts,Robotics Engineers,Healthcare Professionals,Machine Learning Engineers 1 day ago

A Topology-Aware Graph Convolutional Network for Human Pose Similarity and Action Quality Assessment

graph-neural-networks › graph-learning

📄 Abstract

Abstract: Action Quality Assessment (AQA) requires fine-grained understanding of human motion and precise evaluation of pose similarity. This paper proposes a topology-aware Graph Convolutional Network (GCN) framework, termed GCN-PSN, which models the human skeleton as a graph to learn discriminative, topology-sensitive pose embeddings. Using a Siamese architecture trained with a contrastive regression objective, our method outperforms coordinate-based baselines and achieves competitive performance on AQA-7 and FineDiving benchmarks. Experimental results and ablation studies validate the effectiveness of leveraging skeletal topology for pose similarity and action quality assessment.

Authors (1)

Minmin Zeng

Submitted

November 3, 2025

arXiv Category

cs.CV

arXiv PDF

Key Contributions

Proposes a topology-aware Graph Convolutional Network (GCN) framework, GCN-PSN, for human pose similarity and Action Quality Assessment (AQA). By modeling the human skeleton as a graph, it learns topology-sensitive pose embeddings using a Siamese architecture and contrastive regression, outperforming coordinate-based baselines.

Business Value

Enables more objective and accurate analysis of human movement in sports, rehabilitation, and performance monitoring, leading to better training and injury prevention.

Paper Metadata

Innovation Type

Algorithmic Improvement

Deployment Feasibility

Moderate. Requires skeleton data extraction, but GCNs are becoming more common.

Limitations Addressed

Difficulty in capturing fine-grained motion details and pose similarity using coordinate-based methods, lack of explicit consideration for skeletal topology.

Performance Gains

Outperforms coordinate-based baselines on AQA-7 and FineDiving benchmarks.

Technical Tags

Graph Convolutional Network (GCN)human pose similarityAction Quality Assessment (AQA)skeleton graphtopology-awareSiamese architecturecontrastive regressionpose embeddings

Research Topics

Graph Neural NetworksHuman Pose EstimationAction RecognitionComputer VisionDeep Learning

Methods & Architectures

Graph Convolutional Networks (GCN)Siamese architectureContrastive regression objectiveSkeleton graph representation Graph Convolutional Network (GCN)Siamese Network

Applications & Tasks

Sports Analytics Human Motion Analysis Robotics Healthcare (rehabilitation) Surveillance Accurate assessment of human pose similarityObjective evaluation of action qualityLeveraging skeletal topology for motion analysis Learning discriminative pose embeddingsAssessing human pose similarityEvaluating action quality

Datasets & Benchmarks

Datasets

AQA-7, FineDiving

Benchmarks

AQA-7 • FineDiving

pose similarity scoreaction quality scoreaccuracy

Related Fields

Computer VisionMachine LearningGraph Neural NetworksSports ScienceRobotics

Keywords

GCNhuman poseaction quality assessmentpose similarityskeletongraph neural networktopologySiamese networkcontrastive learningcomputer visionmotion analysis

Academic Context

#Graph Neural Networks#Human Pose Estimation#Action Recognition#Computer Vision#Deep Learning

Technology Stack

Frameworks & Libraries

Graph Convolutional Network (GCN)

Commercial Potential

Potential Products

Automated sports performance analysis toolsMotion analysis software for physical therapyHuman motion understanding modules for robots

Target Industries

SportsHealthcareFitnessRoboticsEntertainment

Use Case Examples

Analyzing the technique of athletes to identify areas for improvementMonitoring patient recovery progress through motion analysisEnabling robots to better understand human actions

Competitive Edge

Leverages graph structures and topology awareness to achieve superior performance in pose similarity and action quality assessment compared to traditional methods.

Market Opportunity

Growing markets for sports analytics, digital health, and AI-powered motion analysis.

Revenue Models

Licensing of analysis softwaredevelopment of specialized AI modules.

Resource Requirements

Compute Needs

Moderate to High (for training GCNs)

Data Requirements

Datasets with human motion capture data and action quality labels (e.g., AQA-7, FineDiving).

Deployment Constraints

Requires accurate skeleton extraction from video.

Scalability

GCNs are generally scalable with graph size and complexity.

Regulatory Considerations

None explicitly mentioned.

Production Readiness

Maturity Level

Research

Time to Market

1-3 years for specialized applications

Patent Potential

Low

View Full Paper Back to Papers