tempoGAN

Abstract

Fig. 1. Our convolutional neural network learns to generate highly detailed, and temporally coherent features based on a low-resolution field containing a single time-step of density and velocity data. We introduce a novel discriminator that ensures the synthesized details change smoothly over time.We propose a temporally coherent generative model addressing the superresolution problem for fluid flows. Our work represents a first approach to synthesize four-dimensional physics fields with neural networks. Based on a conditional generative adversarial network that is designed for the inference of three-dimensional volumetric data, our model generates consistent and detailed results by using a novel temporal discriminator, in addition to the commonly used spatial one. Our experiments show that the generator is able to infer more realistic high-resolution details by using additional physical quantities, such as low-resolution velocities or vorticities. Besides improvements in the training process and in the generated outputs, these inputs offer means for artistic control as well. We additionally employ a physics-aware data augmentation step, which is crucial to avoid overfitting and to reduce memory requirements. In this way, our network learns to generate advected quantities with highly detailed, realistic, and temporally coherent features. Our method works instantaneously, using only a single time-step of low-resolution fluid data. We demonstrate the abilities of our (*) Similar amount of contributions.

Toward Stable, General Machine‐Learned Models of the Atmospheric Chemical System

“…Raissi et al (2017) found that learning Navier-Stokes integration closure parameters using a neural network outperformed standard approaches. Additional advances have been made by enforcing or encouraging physical constraints and symmetries, such as the conservation of mass and energy, temporal coherence, Lyapunov stability, and Newton's laws (Erichson et al, 2019;Karpatne et al, 2017;Kim et al, 2018;Lusch et al, 2018;Reichstein et al, 2019;Stewart & Ermon, 2016;Wang et al, 2017;Xie et al, 2018).…”

ScalarFlow

“…Looking ahead, we believe that there is a wide range of exciting future applications for our data. Beyond avenues for benchmarking, accuracy measurements, and novel reconstruction methods, we are looking forward to developments in the area of machine learning that this data will enable [Kim et al 2019;Sato et al 2018;Xie et al 2018].…”

“…To leverage machine learning in the context of simulations, researchers have, e.g., used machine learning to drive particle-based simulations [Ladický et al 2015], replaced the traditional pressure solve with pre-trained models [Tompson et al 2017], and augmented simulated data with learned descriptors [Chu and Thuerey 2017]. Others have focused on learning controllers for rigid body interactions [Ma et al 2018] or aimed for temporal coherence with adversarial training [Xie et al 2018]. This is a nascent field with growing interest, where our data set can provide a connection of simulations with the real world.…”

Efficient preservation and breakup of liquid sheets using screen-projected particles

“…In addition, techniques have been proposed that synthesize fluids using machine learning techniques to reduce computation time. However, since most of these techniques focus on improving efficiency, it is difficult to express high-quality liquid sheets [25][26][27][28][29].…”