Related Experiment Video
Updated: Jul 29, 2026
12:39
A Methodology for Capturing Joint Visual Attention Using Mobile Eye-Trackers
Published on: January 18, 2020
8.1K
Optimizing MobileNetV3 for multimodal eye gaze and emotion recognition via advanced pruning and quantisation
Gousia Habib1,2, Ishfaq Ahmad Malik2, Surbhi Sharma3
1Indian Institute of Technology Delhi, Delhi, India.
Scientific Reports
|October 30, 2025
Summary
This study optimizes MobileNet V3 for recognizing eye gaze, blinks, and emotions. The enhanced model offers accurate visual understanding in resource-constrained settings.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Human-Computer Interaction
Background:
- Visual cues like gaze, blinks, and expressions are crucial for interaction and analysis.
- Resource-constrained environments require efficient and accurate visual understanding models.
- MobileNet V3 provides a lightweight architecture suitable for these tasks.
Purpose of the Study:
- To optimize MobileNet V3 for enhanced visual understanding.
- To improve recognition accuracy for eye-gaze, eye-blinks, and emotional expressions.
- To develop a robust algorithm for real-world applications.
Main Methods:
- Leveraged MobileNet V3 architecture.
- Applied advanced model optimization techniques: pruning and quantization.
- Validated the approach using EyeGaze, Emotions, and Closed Eye datasets.
Main Results:
- The optimized MobileNet V3 model demonstrated accurate detection and analysis of eye gaze, blinks, and emotional expressions.
- Model optimization reduced computational complexity without compromising accuracy.
- The approach proved robust across diverse visual inputs and scenarios.
Conclusions:
- Optimized MobileNet V3 is a powerful tool for visual understanding tasks.
- The model is suitable for real-world applications requiring efficient and accurate analysis of visual cues.
- Developed code and trained models are available for reproducibility.
