Fluid Inverse Volumetric Modeling and Applications From Surface Motion.

Area of Science:

• Computer Graphics
• Fluid Dynamics
• Machine Learning

Background:

• Reconstructing fluid motion from observable data is challenging.
• Existing methods often struggle with temporal coherence and physical accuracy.

Purpose of the Study:

• To develop a framework for volumetric fluid reconstruction from free surface motion.
• To combine deep learning with physical simulation for accurate fluid reproduction.
• To enable physics-guided re-editing of fluid scenes.

Main Methods:

• Inferring surface velocities using deep learning (encoding/decoding spatiotemporal features).
• Generating volumetric velocity fields with a 3D Convolutional Neural Network (CNN).
• Estimating fluid physical properties with a dedicated neural network.
• Integrating reconstructed fields and properties into a physical simulator.

Main Results:

• Successful volumetric fluid reconstruction from both synthetic and real fluid data.
• Preservation of guiding motion and temporal coherence in reproduced fluid.
• Demonstrated effectiveness in graphics applications like fluid behavior reproduction and re-editing.
• Outperformed state-of-the-art methods in 3D fluid inverse modeling and animation.

Conclusions:

• The proposed framework offers a novel and effective approach to 3D fluid reconstruction.
• The amalgamation of deep learning and simulation enhances accuracy and applicability in computer graphics.
• This method advances fluid animation and inverse modeling capabilities.