1: \begin{abstract}

2:   Deep Learning (DL) algorithms are becoming increasingly popular for the

3:   reconstruction of high-resolution turbulent flows (\emph{aka}

4:   super-resolution).  However, current DL approaches perform spatially

5:   \emph{uniform} super-resolution -- a key performance limiter for

6:   scalability of DL-based surrogates for Computational Fluid Dynamics (CFD).

7: 

8: To address the above challenge, we introduce \name, a deep

9: 	learning-based adaptive mesh refinement (AMR) framework for non-uniform

10:   super-resolution of turbulent flows.  \name\ divides the input low-resolution

11:   flow field into \emph{patches}, scores each patch, and predicts

12:   their target resolution.  As a result, it outputs a spatially non-uniform flow

13:   field, adaptively refining regions of the fluid domain to achieve the target

14:   accuracy. We train \name\ with Reynolds-Averaged Navier-Stokes (RANS)

15:   solutions from three different canonical flows, namely turbulent channel

16:   flow, flat plate, and flow around ellipses. \name\ shows remarkable

17:   discerning properties, refining areas with complex flow features, such as

18:   near-wall domains and the wake region in flow around solid bodies, while

19:   leaving areas with smooth variations (such as the freestream) in the

20:   low-precision range. Hence, \name\ demonstrates an excellent qualitative and quantitative

21:   alignment with the traditional OpenFOAM AMR solver. Moreover, it reaches the

22:   same convergence guarantees as the AMR solver while accelerating it by

23:   $3.2-5.5\times$, including unseen-during-training geometries and boundary

24:   conditions, demonstrating its generalization capacities.  Due to \name's

25:   ability to super-resolve only regions of interest, it predicts the same

26:   target $1024\times1024$ spatial resolution $7-28.5\times$ faster than

27:   state-of-the-art DL methods and reduces the memory usage by $4.4-7.65\times$,

28:   showcasing improved scalability. 

29: \end{abstract}

30: