Attention-Based Learning for Fluid State Interpolation and Editing in a Time-Continuous Framework

TL;DR

This paper presents FluidsFormer, a transformer-based method that interpolates fluid states in a continuous-time framework, improving animation smoothness and sharpness by combining PITT and residual neural networks.

Contribution

Introducing FluidsFormer, a novel transformer-based approach that combines PITT and RNNs for fluid state interpolation in a continuous-time setting.

Findings

Effective interpolation of smoke and liquids.

Enhanced temporal smoothness and sharpness in animations.

Promising initial experimental results.

Abstract

In this work, we introduce FluidsFormer: a transformer-based approach for fluid interpolation within a continuous-time framework. By combining the capabilities of PITT and a residual neural network (RNN), we analytically predict the physical properties of the fluid state. This enables us to interpolate substep frames between simulated keyframes, enhancing the temporal smoothness and sharpness of animations. We demonstrate promising results for smoke interpolation and conduct initial experiments on liquids.