Generative Data Augmentation for Skeleton Action Recognition

2026-04-16 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

Researchers propose a conditional generative pipeline for data augmentation in skeleton action recognition, addressing the high cost and labor intensity of collecting large, diverse, and well-annotated 3D skeleton datasets. The method learns the distribution of real skeleton sequences constrained by action labels, synthesizing diverse and high-fidelity data. This approach effectively generates skeleton sequences even with limited training samples, achieving competitive recognition performance in low-data scenarios and demonstrating strong generalization. The pipeline utilizes a Transformer-based encoder-decoder architecture, a generative refinement module, and a dropout mechanism to balance fidelity and diversity during sampling. Experiments on HumanAct12 and NTU-VIBE datasets show consistent accuracy improvements for multiple skeleton-based action recognition models in both few-shot and full-data settings.

Key takeaway

For research scientists developing skeleton-based action recognition models, consider integrating conditional generative data augmentation to mitigate the challenges of limited or expensive 3D skeleton datasets. This approach can significantly improve model accuracy, particularly in few-shot learning scenarios, by synthesizing diverse and high-fidelity training data. You should explore Transformer-based encoder-decoder architectures with refinement modules to balance data fidelity and diversity effectively.

Key insights

A conditional generative pipeline synthesizes diverse skeleton data to augment action recognition, improving performance in low-data settings.

Principles

• Generative models can overcome data scarcity.
• Balancing fidelity and diversity is key for synthetic data.

Method

The method employs a Transformer-based encoder-decoder with a generative refinement module and dropout mechanism to learn and synthesize skeleton sequences conditioned on action labels, ensuring diversity and fidelity.

In practice

• Apply generative augmentation for scarce 3D skeleton data.
• Use Transformer-based architectures for sequence generation.

Topics

• Generative Data Augmentation
• Skeleton Action Recognition
• Transformer-based Encoder-Decoder
• Few-Shot Learning
• HumanAct12 Dataset

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• Generative Pose Data Augmentation with KopiKat.co - #OpenCV Ep. 109 — Generative data augmentation with SMPL models can overcome 3D human pose dataset limitations for specialized applications.
• DanceCrafter: Fine-Grained Text-Driven Controllable Dance Generation via Choreographic Syntax — Fine-grained text-driven dance generation requires specialized syntax, high-quality data, and tailored model architecture.
• Prompt-to-Gesture: Measuring the Capabilities of Image-to-Video Deictic Gesture Generation — Image-to-video models can generate realistic, variable deictic gestures, augmenting scarce human-recorded data for improved downstream task performance.
• The Sequence Knowledge # 780: Synthetic Data for Image Models — Synthetic data generation is crucial for robust vision systems, overcoming real-world data limitations through controlled pixel creation.
• Unsupervised Skeleton-Based Action Segmentation via Hierarchical Spatiotemporal Vector Quantization — A hierarchical spatiotemporal vector quantization method improves unsupervised skeleton-based action segmentation by integrating spatial and temporal cues.
• Think Twice, Act Once: Verifier-Guided Action Selection For Embodied Agents — Explicit test-time verification with a specialized, generatively trained verifier significantly boosts embodied agent robustness.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.