Macro- and Micro-Expressions Facial Datasets: A Survey.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Automatic facial expression recognition (FER) is crucial for numerous applications.
• Effective FER solutions, especially for neural network training, require a comprehensive understanding of available datasets.
• Existing research increasingly focuses on spontaneous, real-world expressions ('in-the-wild' datasets).

Purpose of the Study:

• To provide a comprehensive review of over eighty facial expression datasets.
• To analyze both macro- and micro-expressions, with a focus on spontaneous and in-the-wild data.
• To assist researchers in selecting appropriate datasets by highlighting their pros and cons.

Main Methods:

• Systematic review of existing facial expression datasets.
• Categorization of datasets based on expression type (macro/micro) and context (spontaneous/in-the-wild).
• Analysis of dataset characteristics, including strengths and weaknesses.

Main Results:

• Identification and review of more than eighty facial expression datasets.
• Emphasis on spontaneous and in-the-wild datasets, reflecting current research trends.
• Discussion of potential applications and limitations of each dataset.

Conclusions:

• A clear overview of facial expression datasets is vital for developing and evaluating FER systems.
• The survey facilitates informed dataset selection for researchers in the field.
• Understanding dataset characteristics is key to advancing FER technology.