MESSENGER: Memory-Enhanced Sequential Scene Flow Estimation via Autoregressive Next-Frame Forecasting

Jiuming Liu; Jianing Li; Mengmeng Liu; Hongyang He; Hesheng Wang; Per Ola Kristensson

cs.CVOct 7, 2026

MESSENGER: Memory-Enhanced Sequential Scene Flow Estimation via Autoregressive Next-Frame Forecasting

Authors: Jiuming Liu, Jianing Li, Mengmeng Liu, Hongyang He, Hesheng Wang, Per Ola Kristensson

Organizations: University of Cambridge · Shanghai Jiao Tong University · University of Twente · University of Warwick

Abstract

Scene flow can capture low-level 3D motion displacements in dynamic scenarios. Early pairwise estimators relying on instantaneous two-frame motion lack long-term temporal correlation and also struggle with poor extrapolation ability in future prediction. Although some recent methods attempt to explore multi-frame scene flow estimation in a sequence-to-sequence manner, they typically suffer from heavy computational overhead with increasing input frames and long-horizon prediction degradation due to ineffective motion propagation. To address these problems, we propose a novel memory-enhanced sequential scene flow pipeline, called MESSENGER. To sufficiently mine long-term temporal dependencies naturally within consecutive sequences, a memory buffer is designed by explicitly storing multiple history flow estimates and latent states. For each input frame, the temporally stored flows and states are correlated and retrieved to predict the current initialized flow in a next-frame forecasting manner. Furthermore, we develop an uncertainty-aware reweighting module to filter unreliable retrievals and mitigate accumulated errors. Extensive experiments on nuScenes and Argoverse 2 demonstrate state-of-the-art performance of our MESSENGER, reducing EPE3D by 71.6% on nuScenes and 67.7% on Argoverse 2 in long-horizon future extrapolation. This superiority can be attributed to our designed autoregressive forecasting paradigm, which naturally forces the network to progressively learn the next-frame distribution based on history observations. Code will be released at https://github.com/liujiuming123/Messenger.

Figures & tables

Figure 1 : Comparison with previous multi-frame scene flow methods. ( Left ) Compared to sequence-to-sequence methods with heavy computational costs, our method adopt an autoregressive sequential estimation paradigm, which retrieves multiple history flow priors from a memory buffer. ( Right ) Compared to recent SOTA flow prediction method GenFlow3D Li et al. (2025) suffering from degraded accuracy in longer-horizon predictions, our uncertainty-aware memory reweighting can mitigate long-term compounding errors by measuring history-present motion discrepancy.

Figure 2 : GPU memory comparison during training.

Figure 3 : Overall pipeline of our proposed MESSENGER.

Method	Reference	Dataset	Scene Flow Estimation
			$\mathcal{SF}_{1}$			$\mathcal{SF}_{2}$
			EPE3D $\downarrow$	Acc3DS $\uparrow$	Out3D $\downarrow$	EPE3D $\downarrow$	Acc3DS $\uparrow$	Out3D $\downarrow$
PointPWC-Net Wu et al. (2020)	ECCV’20	nuScenes	0.0386	0.7668	0.6255	0.0386	0.7695	0.6212
FLOT Puy et al. (2020)	ECCV’20		0.0297	0.8561	0.5071	0.0293	0.8604	0.5049
PV-RAFT Wei et al. (2021)	CVPR’21		0.0500	0.7570	0.6055	0.0492	0.7598	0.6024
Bi-PointFlowNet Cheng and Ko (2022)	ECCV’22		0.0291	0.8676	0.4924	0.0291	0.8696	0.4913

Table 1 : Comparison of scene flow estimation results on nuScenes Caesar et al. (2020) and Argoverse 2 Wilson et al. (2023) . We follow Li et al. (2025) , inputting

H=3

consecutive frames to estimate two corresponding flows in between.

Figure 4 : Qualitative comparison in scene flow estimation and prediction on nuScenes. From up to bottom, rows show ground truth, GenFlow3D Li et al. (2025) , and our predictions across time. Scene flow is visualized with hue indicating direction and saturation representing speed. Comparison is highlighted with orange circles. Please zoom in for details.

nuScenes	Methods	$\mathcal{SF}_{H+1}$		$\mathcal{SF}_{H+2}$		$\mathcal{SF}_{H+3}$		$\mathcal{SF}_{H+4}$
	Methods	EPE3D $\downarrow$	Acc3DS $\uparrow$	EPE3D $\downarrow$	Acc3DS $\uparrow$	EPE3D $\downarrow$	Acc3DS $\uparrow$	EPE3D $\downarrow$	Acc3DS $\uparrow$
	GenFlow3D Li et al. (2025)	0.029	0.948	0.046	0.889	0.065	0.830	0.088	0.754
	Ours	0.015	0.969	0.018	0.938	0.021	0.915	0.025	0.883
Argoverse 2	Methods	$\mathcal{SF}_{H+1}$		$\mathcal{SF}_{H+2}$		$\mathcal{SF}_{H+3}$		$\mathcal{SF}_{H+4}$
	Methods	EPE3D $\downarrow$	Acc3DS $\uparrow$	EPE3D $\downarrow$	Acc3DS $\uparrow$	EPE3D $\downarrow$	Acc3DS $\uparrow$	EPE3D $\downarrow$	Acc3DS $\uparrow$
	GenFlow3D Li et al. (2025)	0.049	0.651	0.105	0.279	0.140	0.172	0.192	0.130

Table 2 : Comparison of scene flow prediction results on nuScenes Caesar et al. (2020) and Argoverse 2 Wilson et al. (2023) . We set the observed frames as

H

= 6 and predict future frames with the intervals of

k=1,2,3,4

Figure 5 : Qualitative comparison in scene flow estimation and prediction on Argoverse 2. From up to bottom, rows show ground truth, GenFlow3D Li et al. (2025) , and our predictions across time. Scene flow is visualized with hue indicating direction and saturation representing speed. Comparison is highlighted with orange circles. Please zoom in for details.

Methods	DifFlow3D Liu et al. (2024a)	GenFlow3D Li et al. (2025)	Ours
Latency	216.7ms	88.7ms	55.9ms

Table 3 : Latency comparison.

Table 4 : Ablation studies on the nuScenes dataset. We evaluate the impact of core architecture designs, memory lengths, and memory stored contents in the 4-th future prediction setting.

Figure 6 : Comparison of different input frames ( $H$ ) and extrapolation frames ( $K$ ) for future scene flow prediction on nuScenes.

Figure 7 : Visualization of uncertainty. We map the learned uncertainty values to a corresponding color strip.

Table 5 : Per-class analysis of estimation errors on nuScenes.

Explore similar work

Apr 21, 2026cs.CV

RAFT-MSF++: Temporal Geometry-Motion Feature Fusion for Self-Supervised Monocular Scene Flow

Monocular scene flow estimation aims to recover dense 3D motion from image sequences, yet most existing methods are limited to two-frame inputs, restricting temporal modeling and robustness to occlusions. We propose RAFT-MSF++, a self-supervised multi-frame framework that recurrently fuses temporal features to jointly estimate depth and scene flow. Central to our approach is the Geometry-Motion Feature (GMF), which compactly encodes coupled motion and geometry cues and is iteratively updated for effective temporal reasoning. To ensure the robustness of this temporal fusion against occlusions, we incorporate relative positional attention to inject spatial priors and an occlusion regularization module to propagate reliable motion from visible regions. These components enable the GMF to effectively propagate information even in ambiguous areas. Extensive experiments show that RAFT-MSF++ achieves 24.14% SF-all on the KITTI Scene Flow benchmark, with a 30.99% improvement over the baseline and better robustness in occluded regions. The code is available at https://github.com/sunzunyi/RAFT-MSF-PlusPlus.

Xunpei Sun, Zuoxun Hou, Yi Chang +2

School of Computer Science and Engineering, Sun Yat-sen University, Guangzhou, China 510006. · Beijing Institute of Space Mechanics and Electricity, Beijing, China 100094. · School of Artificial Intelligence and Automation, Huazhong University of Science and Technology, Wuhan, China 430074.

Optical Flow

Ask PDF

Jun 1, 2026cs.RO

DisFlow: Scene Flow from Distance Field for Object Pose, Velocity Tracking, and Dynamic Object Reconstruction

We present \emph{DisFlow}, a novel framework for online scene flow estimation from distance field that enables \emph{6DoF dynamic object pose estimation}, \emph{motion tracking}, and \emph{surface reconstruction}. The scene is represented by Gaussian Process Implicit Surfaces (GPIS), with surface normals serving as derivative constraints, enabling accurate signed distance computations near the surface and gradient queries with uncertainty. With this representation as a foundation, we compute a scene flow from the distance field that describes how surface points are transported over time in consecutive frames. Through our flow, we can estimate an object's pose and motion by incrementally registering a new observed point cloud via an elegant closed-form optimisation. Unlike prior methods that operate in the camera or world frame, our approach performs probabilistic fusion directly in the \emph{object frame}, where the object remains geometrically consistent over time. The tight coupling of the DisFlow method in space and time yields dense geometry, surface normals, object pose trajectories, velocities, and uncertainty, all at real-time rates. We evaluate DisFlow on dynamic object sequences and demonstrate that it achieves accurate pose and motion tracking while simultaneously reconstructing high-quality object surfaces. Code publicly available at https://github.com/LanWu076/disflow_ros2

Lan Wu, Sheila Sutjipto, Jennifer Wakulicz +1

Robotics Institute, University of Technology Sydney, Ultimo, NSW 2007, Australia

Visual Object Tracking Object Pose Estimation Dynamic Scene Reconstruction Point Cloud Surface Reconstruction

Ask PDF

Sep 11, 2026cs.CV

FreeFlow: A Bias-free Hierarchical Transformer for Optical Flow Estimation

Optical flow methods typically rely on task-specific inductive biases, such as correlation volumes, feature warping, and iterative refinement, among others, to reach high accuracy. While effective, such biases constrain the model to predefined heuristics, which can limit its expressivity and lead to more complex pipelines and additional computational cost. We present FreeFlow, a hierarchical transformer built without any flow-specific components, using instead a single feed-forward encoder--decoder. FreeFlow combines three attention variants: window attention for local processing, shifted-window attention for cross-window information exchange, and a global attention operating at a reduced resolution. The resulting architecture scales naturally with model capacity, enabling a consistent accuracy gain from small to large variants. Despite the absence of standard inductive biases, FreeFlow achieves state-of-the-art results on major benchmarks, including Sintel (0.68/1.48 EPE on Clean/Final), KITTI-2015 (3.23 Fl-all), and Spring (3.192 1px), while remaining memory efficient at 1080p inference.

Vladislav Bargatin, Alexander Yakovenko, Khaled Abud +1

AI Center, Lomonosov MSU, Moscow, Russia · Lomonosov Moscow State University, Moscow, Russia · MSU Institute for Artificial Intelligence, Moscow, Russia

Optical Flow Vision Transformer Efficient ViTs

Ask PDF