Synthetic Data in Human Analysis: A Survey.

Area of Science:

• Computer Vision
• Machine Learning
• Human-Computer Interaction

Background:

• Deep neural networks (DNNs) excel in human analysis tasks like recognition and re-identification.
• DNN performance relies heavily on large-scale training datasets.
• Acquiring real-world data for human analysis is challenging due to cost, time, and privacy concerns.

Purpose of the Study:

• To survey methodologies for generating and utilizing synthetic data in human analysis.
• To highlight the benefits of synthetic data as an alternative to real-world data collection.
• To provide an overview of current synthetic data generation models and datasets.

Main Methods:

• Literature review of synthetic data generation techniques for human analysis.
• Analysis of state-of-the-art methods and their applications.
• Compilation of publicly available synthetic datasets and generation models.

Main Results:

• Synthetic data generation is a viable and privacy-preserving alternative for training DNNs.
• Significant advancements in synthetic data generation methods have been observed.
• Numerous synthetic datasets and models are now accessible to researchers.

Conclusions:

• Synthetic data effectively addresses the data scarcity and privacy issues in human analysis.
• Further research is needed to overcome existing limitations and explore open problems.
• The survey provides a comprehensive resource for researchers and practitioners in the field.