🤖 AI Summary
This work addresses the challenge of high-fidelity flow field estimation and scalar field temporal interpolation for scientific ensemble data (2D+time/3D+time) under conditions of missing or only sparse ground-truth flow fields. We propose the first end-to-end deep learning framework specifically designed for such data. Our method employs a modular architecture integrating multi-scale convolutional–deconvolutional networks with an adaptive, configurable loss function, enabling unified modeling across supervised, weakly supervised, and fully unsupervised flow estimation scenarios. Crucially, it generates dense, time-resolved flow fields even for datasets lacking any initial flow information. Extensive evaluation on both synthetic and real experimental datasets demonstrates significant improvements in flow estimation accuracy and temporal interpolation quality over conventional approaches. The framework establishes a new paradigm for scientific visualization in settings where flow priors are unavailable.
📝 Abstract
We present FLINT (learning-based FLow estimation and temporal INTerpolation), a novel deep learning-based approach to estimate flow fields for 2D+time and 3D+time scientific ensemble data. FLINT can flexibly handle different types of scenarios with (1) a flow field being partially available for some members (e.g., omitted due to space constraints) or (2) no flow field being available at all (e.g., because it could not be acquired during an experiment). The design of our architecture allows to flexibly cater to both cases simply by adapting our modular loss functions, effectively treating the different scenarios as flow-supervised and flow-unsupervised problems, respectively (with respect to the presence or absence of ground-truth flow). To the best of our knowledge, FLINT is the first approach to perform flow estimation from scientific ensembles, generating a corresponding flow field for each discrete timestep, even in the absence of original flow information. Additionally, FLINT produces high-quality temporal interpolants between scalar fields. FLINT employs several neural blocks, each featuring several convolutional and deconvolutional layers. We demonstrate performance and accuracy for different usage scenarios with scientific ensembles from both simulations and experiments.