ConMLP: MLP-Based Self-Supervised Contrastive Learning for Skeleton Data Analysis and Action Recognition.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Human action recognition is crucial for computer vision applications.
• Skeleton-based action recognition has advanced using deep learning, often with convolutional neural networks (CNNs).
• Existing methods face challenges with model complexity, computational cost, reliance on labeled data, and real-time application suitability.

Purpose of the Study:

• To address the limitations of conventional action recognition models.
• To propose an efficient and effective self-supervised learning framework for human action recognition.
• To develop a model suitable for real-world applications with limited computational resources and unlabeled data.

Main Methods:

• A multi-layer perceptron (MLP)-based self-supervised learning framework named ConMLP.
• Utilizing a contrastive learning loss function within the MLP architecture.
• Processing skeleton sequences for spatial and temporal feature extraction.

Main Results:

• ConMLP achieves a top-1 inference accuracy of 96.9% on the NTU RGB+D dataset, outperforming state-of-the-art self-supervised methods.
• The framework demonstrates reduced computational complexity and resource consumption compared to conventional deep learning models.
• ConMLP shows comparable performance to supervised methods when evaluated in a supervised manner.

Conclusions:

• ConMLP offers an efficient and effective solution for human action recognition using self-supervised learning.
• The framework's low computational requirements and ability to leverage unlabeled data make it suitable for real-time and embedded applications.
• ConMLP represents a significant advancement in self-supervised learning for skeleton-based action recognition.