Automatic Schelling Point Detection From Meshes.

Area of Science:

• Computer Graphics
• Perceptual Psychology
• Machine Learning

Background:

• Mesh Schelling points indicate human focus on 3D object regions, crucial for computer graphics and perceptual psychology.
• Current methods for detecting mesh Schelling points are time-consuming and costly, relying on participant observation studies.

Purpose of the Study:

• To develop an automatic and efficient method for detecting mesh Schelling points using deep learning.
• To introduce the Deep Schelling Network (DS-Net) as the first deep neural network for this task.

Main Methods:

• Utilized mesh convolution and pooling operations for feature extraction from 3D mesh objects.
• Developed DS-Net with a multi-scale fusion component and a region-specific loss function for end-to-end heatmap prediction.
• Trained and evaluated DS-Net on a dataset derived from participant observation studies.

Main Results:

• DS-Net effectively detects mesh Schelling points, demonstrating superior performance over existing state-of-the-art methods.
• Qualitative and quantitative evaluations confirmed the network's capability in identifying salient regions on 3D meshes.
• The proposed network achieved better regression of heatmaps compared to other deep learning models.

Conclusions:

• DS-Net offers an automated, efficient, and effective solution for detecting mesh Schelling points, overcoming limitations of traditional methods.
• This deep learning approach advances the field by providing a scalable tool for analyzing visual attention on 3D models.
• The findings have significant implications for both computer graphics applications and understanding human visual perception.