Hyfluid

HyFluid: A Neural Framework for Fluid Dynamics Inference from Sparse Multi-View Videos

HyFluid is an innovative neural-based approach designed to infer fluid density and velocity fields from sparse, multi-view video inputs. Unlike traditional neuro-fluid dynamics reconstruction methods, HyFluid demonstrates superior accuracy in estimating fluid density while simultaneously revealing underlying velocity information, effectively addressing the inherent visual challenges posed by fluid velocity estimation.

Key Features

The method introduces a comprehensive solution for physically plausible velocity field inference through the integration of physics-based loss functions. This approach not only handles the complex turbulent nature of fluid dynamics but also employs a hybrid neural velocity representation:

• A base neural velocity field that captures most of the irrotational energy components.
• Vortex particle velocities to simulate and model the remaining turbulent velocity characteristics.

This dual approach ensures accurate and physically consistent fluid dynamics reconstruction, making HyFluid highly suitable for a variety of applications involving 3D incompressible flows.

Applications

HyFluid’s versatility extends across multiple domains:

• Fluid Resimulation and Editing: Rebuilding fluid fields from sparse multi-view video inputs.
• Future Prediction: Predicting the future evolution of fluid dynamics.
• Neural Dynamic Scene Synthesis: Generating synthetic fluid scenes through neural networks.

Functional Capabilities

The system offers a range of powerful functionalities:

• Infer 3D fluid density and velocity fields from sparse, multi-view video inputs.
• Visualize the reconstructed 3D fluid fields for better understanding and analysis.
• Resimulate new viewpoints to explore different perspectives of the fluid dynamics.
• Predict future fluid behavior with high accuracy.
• Enable dynamic neural scene synthesis for creative applications.

By combining advanced neural networks with physics-based constraints, HyFluid provides a robust framework for tackling complex fluid dynamics problems in various real-world scenarios.