C3I-SynMicrosaccade: A pipeline and dataset for microsaccade recognition using neuromorphic event camera streams.

Area of Science:

• Computer Vision
• Neuroscience
• Biomedical Engineering

Background:

• Microsaccades are crucial for visual perception and are challenging to model.
• Existing datasets may lack the specific characteristics needed for event-based analysis.

Purpose of the Study:

• To introduce the C3I-SynMicrosaccade dataset for event-based microsaccade modeling and classification.
• To provide a large-scale, balanced dataset for training and evaluating algorithms.

Main Methods:

• Generated high-resolution RGB sequences of microsaccades using Blender with specific trajectory and angular classes.
• Converted RGB frames to event-based data streams using the v2e simulator, mimicking neuromorphic sensors.
• Controlled temporal durations and event counts for class balance and bias elimination.

Main Results:

• Created a synthetic dataset of 175,000 event sequences for microsaccades.
• Dataset features seven distinct angular classes (0.5° to 2.0°) with controlled motion dynamics.
• Ensured class balance, eliminating bias toward motion magnitude.

Conclusions:

• The C3I-SynMicrosaccade dataset offers a controlled, reproducible framework for studying fixational eye movements.
• Enables robust evaluation of event-based algorithms under fine motion dynamics.
• Facilitates research in microsaccade recognition, neuromorphic vision, and synthetic-to-real transfer learning.