High-Resolution Event Meshflow (HREM)

• High-Resolution Event Meshflow (HREM) is a framework that reconstructs spatially coherent and high-resolution motion fields from sparse or coarse event-driven observations.
• It encompasses multiple formulations, including an event-camera dataset with the EEMFlow architecture, graph-based super-resolution, and reduced-order real-time velocimetry approaches.
• The approach delivers fast, accurate flow estimation (e.g., 142.9 FPS and reduced EPE) and supports practical applications like video stabilization, SLAM, and autonomous navigation.

Searching arXiv for the cited HREM-related papers to ground the article. High-Resolution Event Meshflow (HREM) denotes a recent family of formulations for recovering spatially coherent, higher-resolution motion fields from event-driven or otherwise coarse flow observations. In its most explicit usage, HREM is a high-resolution event-camera dataset and supervision pipeline for event-based meshflow and optical flow at $1280\times720$, accompanied by the Efficient Event-based MeshFlow (EEMFlow) network, the Confidence-induced Detail Completion (CDC) module, and the HREM+ multi-density extension (Luo et al., 5 Oct 2025). In the supplied literature, the same acronym is also used for a mesh-based super-resolution architecture derived from multiscale graph neural networks (Barwey et al., 2024) and for a reduced-order reconstruction framework for real-time Event-Based Imaging Velocimetry (rt-EBIV) (Franceschelli et al., 5 May 2026). The shared technical theme is super-resolution or refinement of flow information under sparsity, reduced sampling, or event-based sensing.

1. Terminology and scope

The designation HREM is not used in a single, fully standardized way across the supplied literature. Instead, it appears in three closely related senses: as a dataset for event-based meshflow learning, as a graph-based super-resolution architecture on mesh elements, and as a reduced-order real-time reconstruction framework for event-based velocimetry. This suggests that the acronym functions more as a thematic label for high-resolution flow recovery than as a unique canonical method.

Usage of HREM	Paper	Core object
High-Resolution Event Meshflow dataset and learning pipeline	(Luo et al., 5 Oct 2025)	Dataset, EEMFlow, CDC, HREM+
HREM super-resolution architecture	(Barwey et al., 2024)	Synchronized message passing, CSP, unpooling, FSP
HREM framework for rt-EBIV	(Franceschelli et al., 5 May 2026)	POD mapping, linear dynamics, KF/LSE/LSE+VR

In the event-camera setting, meshflow is defined as a sparse, spatially smooth motion field defined at the vertices of a regular grid. Compared with dense optical flow, it is presented as a representation that dramatically reduces redundancy and computational cost while still capturing non-rigid and multi-planar motion (Luo et al., 5 Oct 2025). In the fluid and velocimetry settings, HREM denotes super-resolution of flow fields from coarse or reduced observations rather than a particular event-camera dataset (Barwey et al., 2024).

2. Dataset construction and meshflow formulation

In (Luo et al., 5 Oct 2025), HREM was created to enable supervised learning of event-based meshflow and optical flow at high resolution, in the presence of dynamic objects, complex motions, and varying illumination. The dataset statistics reported are 100 virtual scenes rendered in Blender (40 indoor, 60 outdoor), 20,000 training pairs, 8,000 test pairs, and spatial resolution: $1280\times720$. Four test sub-sequences are defined by scene type and motion speed: Outdoor_slow (mean motion 0–30 px), Outdoor_fast (30–100 px), Indoor_slow (0–20 px), and Indoor_fast (20–100 px). Event data are captured at two temporal windows: $dt=1$ (60 Hz meshflow) and $dt=4$ (15 Hz) (Luo et al., 5 Oct 2025).

The synthesis pipeline begins with camera trajectories $\Gamma(t)$ with 6-DoF generated via PyBullet. Blender renders high-frame-rate videos $I(u,\tau_k)$ and ground-truth dense optical flow $F_{\tau_k\rightarrow\tau_{k+1}}$. Meshflow ground truth is then derived by placing a $16\times16$ regular mesh on each image, collecting a dense flow vector at each cell center, propagating that vector to the $3\times3$ block of nearby vertices, and applying two median filters: a first median filter $f_1$ on each vertex and a second $1280\times720$0 median filter $1280\times720$1 on the resulting sparse field to enforce spatial smoothness. The output is a sparse $1280\times720$2 motion field $1280\times720$3, bilinearly upsampled for display (Luo et al., 5 Oct 2025).

Event streams are synthesized by three event simulators (ESIM, V2E, DVS-Voltmeter). Each event $1280\times720$4 satisfies

$1280\times720$5

where $1280\times720$6, $1280\times720$7, and $1280\times720$8 is the contrast threshold. To select among simulators, the pipeline computes the Image of Warped Events (IWE) contrast,

$1280\times720$9

and chooses the simulator with maximum $dt=1$0 (Luo et al., 5 Oct 2025).

A further ingredient is event density. HREM+ extends HREM by synthetically varying the event-generation threshold $dt=1$1 to produce event data densities $dt=1$2. Density is defined from a voxel grid $dt=1$3 over $dt=1$4 temporal bins as

$dt=1$5

This explicit parameterization of density is central to the later ADM module and to the paper’s robustness study across varying event densities (Luo et al., 5 Oct 2025).

The meshflow representation itself is formulated on an $dt=1$6 mesh with motion vectors $dt=1$7 at each vertex. The stated goal is a sparse field that is both consistent with the underlying dense flow $dt=1$8 and spatially smooth. For clarity, the two median filters may be viewed as minimizing

$dt=1$9

subject to matching propagated candidate vectors, although the implementation uses the two cascaded $dt=4$0 median filters rather than explicit optimization (Luo et al., 5 Oct 2025).

3. EEMFlow architecture for event-based meshflow

The principal learning architecture associated with HREM is Efficient Event-based MeshFlow (EEMFlow). Its inputs are two voxelized event volumes $dt=4$1. The network has three major stages: a pyramid encoder, Dilated Feature Correlation (DFC), and feature fusion with a lightweight decoder (Luo et al., 5 Oct 2025).

The Pyramid Encoder has three levels $dt=4$2 operating at downsample factors $dt=4$3. Each level uses shared $dt=4$4 convolutions to extract features $dt=4$5. At each scale, Dilated Feature Correlation forms a cost volume

$dt=4$6

where $dt=4$7 is a sparse “dilated” search grid,

$dt=4$8

This masking preserves a large receptive field while lowering cost. The reported comparison is that standard correlation with radius $dt=4$9 has cost $\Gamma(t)$0, whereas DFC achieves the same range with approximately 50% fewer operations (Luo et al., 5 Oct 2025).

The decoder stacks the cost volumes with pooled features and uses grouped “Shuffle” convolutions to regress a full-resolution meshflow $\Gamma(t)$1. The paper’s ablations state that Shuffle conv vs. standard/depthwise yields the best trade-off in decoder, and that DFC vs. standard correlation provides large radius at low cost (Luo et al., 5 Oct 2025).

This architecture is explicitly lightweight. The reported meshflow model has 1.24 M parameters, 7 ms/inference (142.9 FPS), and Avg EPE = 5.50 ($\Gamma(t)$2), 24.45 ($\Gamma(t)$3) on HREM. Relative to ERAFT [Gehrig et al. 2021], the paper reports $\Gamma(t)$4 params, $\Gamma(t)$5 time, $\Gamma(t)$6 EPE (Luo et al., 5 Oct 2025).

4. Dense-flow extension, CDC, and density adaptation

When EEMFlow is upgraded to predict dense optical flow, the architecture uses a coarse-to-fine residual refinement and introduces the Confidence-induced Detail Completion (CDC) module at each upsampling stage (Luo et al., 5 Oct 2025). At decoder level $\Gamma(t)$7, after bilinear upsampling of the current flow estimate $\Gamma(t)$8, the network warps features, forms a correlation, and regresses a residual:

$\Gamma(t)$9

CDC takes the upsampled flow $I(u,\tau_k)$0 and processes it through two branches. The Self-Corrector branch is a 5-layer dense conv net on concatenated features $I(u,\tau_k)$1 that predicts a residual $I(u,\tau_k)$2 and a confidence map $I(u,\tau_k)$3. The Self-Correlation branch is a small self-attention module that computes weights $I(u,\tau_k)$4 to detect feature-space patterns matching blurring errors in $I(u,\tau_k)$5. The corrections are combined as

$I(u,\tau_k)$6

with $I(u,\tau_k)$7 empirically, and the final upsampled flow is

$I(u,\tau_k)$8

The ablations report improvements from self-corrector alone (0.81$I(u,\tau_k)$90.79 EPE), self-correlation alone (0.89$F_{\tau_k\rightarrow\tau_{k+1}}$00.79), and both (0.79$F_{\tau_k\rightarrow\tau_{k+1}}$10.75) at 39.2 FPS (Luo et al., 5 Oct 2025).

The second major extension is HREM+, which contains events at uniformly sampled densities $F_{\tau_k\rightarrow\tau_{k+1}}$2. The stated empirical observation is that models trained at density $F_{\tau_k\rightarrow\tau_{k+1}}$3 perform best across real datasets. To generalize across densities, the paper introduces the Adaptive Density Module (ADM). Its Multi-Density Changer (MDC) is a 3-level encoder–decoder that maps an input $F_{\tau_k\rightarrow\tau_{k+1}}$4 to $F_{\tau_k\rightarrow\tau_{k+1}}$5, while the Multi-Density Selector (MDS) concatenates the original $F_{\tau_k\rightarrow\tau_{k+1}}$6 and $F_{\tau_k\rightarrow\tau_{k+1}}$7, applies a two-layer conv, and produces pixel-wise softmax weights $F_{\tau_k\rightarrow\tau_{k+1}}$8. The adjusted volume is

$F_{\tau_k\rightarrow\tau_{k+1}}$9

which is then fed to EEMFlow or EEMFlow+ (Luo et al., 5 Oct 2025).

Training uses $16\times16$0 regression at full resolution for both meshflow and optical flow:

$16\times16$1

For ADM, two auxiliary losses on HREM+ are defined using moderate-density $16\times16$2:

$16\times16$3

and

$16\times16$4

The total loss is

$16\times16$5

with $16\times16$6 and $16\times16$7 (Luo et al., 5 Oct 2025).

5. Reported empirical performance and ablations

The principal reported results for HREM concern meshflow estimation on the HREM dataset and optical-flow estimation on DSEC. For meshflow on HREM, EEMFlow achieves Avg EPE = 5.50 ($16\times16$8) and 24.45 ($16\times16$9), while with ADM these improve to 5.05 ($3\times3$0, $3\times3$1) and 22.43 ($3\times3$2, $3\times3$3) (Luo et al., 5 Oct 2025).

Setting	Metric summary	Reported runtime
EEMFlow on HREM	Avg EPE $3\times3$4 ($3\times3$5), $3\times3$6 ($3\times3$7)	7 ms, 142.9 FPS
EEMFlow + ADM on HREM	Avg EPE $3\times3$8 ($3\times3$9), $f_1$0 ($f_1$1)	not separately stated
EEMFlow+ + both CDC branches on DSEC	1PE $f_1$2, EPE $f_1$3, AE $f_1$4	39.2 FPS
EEMFlow+ + ADM on DSEC	1PE $f_1$5, 2PE $f_1$6, 3PE $f_1$7, EPE $f_1$8, AE $f_1$9	22.9 FPS

For dense optical flow on DSEC, EEMFlow+ without CDC is reported at 60.4 FPS, 1PE = 15.5, EPE = 0.89, and AE = 3.11. Adding both CDC branches yields 39.2 FPS, 1PE = 11.4, EPE = 0.75, and AE = 2.67, described as a state-of-the-art speed–accuracy trade-off. With ADM, the reported metrics become 22.9 FPS, 1PE = 10.3 ($1280\times720$00), 2PE = 3.68 ($1280\times720$01), 3PE = 2.12 ($1280\times720$02), EPE = 0.73, and AE = 2.70 (Luo et al., 5 Oct 2025).

The density-generalization results are central to the HREM+ extension. The reported finding is that training on HREM+ densities $1280\times720$03 achieves the best cross-dataset performance on MVSEC and DSEC. In addition, spatial/temporal guided subsampling on DSEC confirms that lower $1280\times720$04 degrades performance linearly; ADM mitigates up to 60% density reduction. The paper further reports that ADM also yields +8–14% gains when plugged into TMA, BFlow, etc. (Luo et al., 5 Oct 2025).

Among ablations, the paper states that softmax fusion in ADM outperforms average/direct outputs, and that bilinear upsampling of meshflow yields better global alignment than mesh-homography or kriging. In the conclusions, HREM+ is described as the first high-resolution ($1280\times720$05), dynamic, multi-density event dataset with both meshflow and dense optical flow ground truth, while EEMFlow is reported as achieving SOTA accuracy (Avg EPE = 5.50) at 142.9 FPS. The abstract additionally characterizes EEMFlow as 30x faster than a recent state-of-the-art flow method and states that ADM improves the performance of EEMFlow and EEMFlow+ by 8% and 10%, respectively (Luo et al., 5 Oct 2025).

The applications named in the paper are real-time video stabilization, SLAM, HDR imaging in challenging lighting, and autonomous navigation. The listed future directions are end-to-end fusion of images and events, learning to synthesize optimal densities on the fly, and extending meshflow to 3D scene flow with event arrays (Luo et al., 5 Oct 2025).

6. Related HREM formulations in fluid super-resolution and rt-EBIV

In the supplied literature, HREM also labels a four-stage super-resolution architecture built from synchronized message passing, a coarse-scale processor (CSP), graph-unpooling, and a fine-scale processor (FSP), summarized from Barwey et al.’s work on mesh-based fluid-flow super-resolution (Barwey et al., 2024). In that formulation, synchronized message passing modifies a baseline graph-convolution layer by averaging aggregated features across coincident nodes:

$1280\times720$06

with node and edge updates performed by $1280\times720$07 and $1280\times720$08. The two MLPs are reported to employ ELU activations and layer-norm, and residual connections inside the MLP itself. The CSP embeds a query coarse element graph and up to 26 neighbors in a structured hexahedral stencil into a latent representation using $1280\times720$09 synchronized message-passing layers. A KNN-based interpolation with $1280\times720$10 performs graph-unpooling from coarse to fine latent graphs, and the FSP applies $1280\times720$11 standard (unsynchronized) message-passing layers as a fine-scale corrector before decoding a residual velocity field in $1280\times720$12 (Barwey et al., 2024).

The numerical experiments reported in that summary are restricted to Taylor–Green Vortex in a periodic cube, solved by NekRS with spectral-element hexahedral mesh, at Re = 1600 and 3200. Training uses three time-levels near peak dissipation ($1280\times720$13), coarsened from $1280\times720$14 DNS to $1280\times720$15. Two model variants are given: Model 1 (pure coarse-scale): $1280\times720$16, $1280\times720$17 and Model 2 (multiscale): $1280\times720$18 coarse layers + $1280\times720$19 fine layers, with neighborhood sizes 0, 6 or 26 coarse-element neighbors. The reported findings are that both models dramatically outperform classical spectral-element interpolation, that at Re = 1600 the multiscale variant with neighbors reduces element-MSE by approximately 20–30%, that at Re = 3200 the purely coarse-scale model is nearly as accurate, and that errors scale roughly with Reynolds number and concentrate where coarse field has high local variance. The same summary states that time-extrapolation and Reynolds-extrapolation remain reasonable, though upward Re-extrapolation is more challenging (Barwey et al., 2024).

A distinct HREM formulation appears in “Real-Time Estimation of High-Resolution Flow Fields and Reduced-Order Coordinates from Event-Based Imaging Velocimetry” (Franceschelli et al., 5 May 2026). There, HREM couples a very fast, low-resolution event-based velocimetry pipeline with an offline-trained mapping and dynamical model that reconstructs high-resolution velocity fields in real time. Offline, paired LR and HR fields are used for Proper Orthogonal Decomposition (POD) on both resolutions, yielding $1280\times720$20 and $1280\times720$21, and for identification of an LR$1280\times720$22HR linear mapping

$1280\times720$23

together with a one-step linear dynamical operator

$1280\times720$24

Online, each LR snapshot is projected onto the LR basis, an estimated HR coefficient vector is produced, and the HR field is reconstructed by

$1280\times720$25

That rt-EBIV HREM compares three estimators: a direct Kalman filter (KF), Linear Stochastic Estimation + KF (LSE), and Variance-Rescaled LSE + KF (LSE+VR). The computational costs per snapshot are stated as $1280\times720$26 for LR projection, $1280\times720$27 for reduced-state estimation, and $1280\times720$28 for HR field reconstruction, with the reconstruction step dominating. On the jet dataset at $1280\times720$29, a C++ microbenchmark reports that the added time per frame remains below 0.8 ms in the worst case, and that if only reduced coordinates are needed, the latent pipeline runs in approximately $1280\times720$30, described as negligible compared with raw cross-correlation latency. Quantitatively, on the jet case, normalized RMSE is reported as cubic interpolation $1280\times720$31, KF $1280\times720$32, LSE $1280\times720$33, and LSE+VR $1280\times720$34; on the channel flow, cubic $1280\times720$35 versus LSE $1280\times720$36. The paper also reports that HREM better preserves turbulent kinetic energy, temporal spectra, spatial spectra, and temporal autocorrelation than cubic interpolation (Franceschelli et al., 5 May 2026).

A common misconception would be to treat these three HREM instances as interchangeable. The supplied literature does not support that interpretation. Rather, HREM refers to a set of related high-resolution flow-recovery strategies: one centered on event-camera meshflow datasets and lightweight neural estimation (Luo et al., 5 Oct 2025), one on graph-based mesh super-resolution in computational fluid dynamics (Barwey et al., 2024), and one on reduced-order real-time reconstruction for event-based velocimetry (Franceschelli et al., 5 May 2026). What unifies them is the objective of reconstructing dynamically meaningful, higher-resolution flow structure from sparse, localized, or low-resolution inputs.