FlowNet: A Deep Learning Framework for Clustering and Selection of Streamlines and Stream Surfaces

Area of Science:

• Computer Vision
• Scientific Visualization
• Data Analysis

Background:

• Effective flow visualization requires identifying representative flow lines or surfaces.
• Existing methods struggle to address both flow lines and surfaces simultaneously.
• Automated selection and clustering of flow features are crucial for complex datasets.

Purpose of the Study:

• To introduce FlowNet, a unified deep learning framework for clustering and selecting both streamlines and stream surfaces.
• To enable efficient exploration and visual reasoning of flow field data.
• To provide a robust solution for representative feature identification in flow visualization.

Main Methods:

• Flow data converted into binary volumes.
• Autoencoder employed to learn latent feature descriptors.
• Dimensionality reduction and clustering of feature descriptors for analysis.
• Development of an interactive visual interface for exploration.

Main Results:

• Learned feature descriptors effectively represent flow lines and surfaces in latent space.
• Dimensionality reduction and clustering enable intuitive exploration.
• Validated effectiveness across diverse flow field datasets.
• Demonstrated superiority over state-of-the-art selection algorithms.

Conclusions:

• FlowNet offers a powerful, unified deep learning approach for flow line and surface selection.
• The framework enhances visual reasoning and exploration of flow data.
• FlowNet represents a significant advancement in flow visualization techniques.