Pith. sign in

REVIEW 3 major objections 8 minor 52 references

EDCFlow: Exploring Temporally Dense Difference Maps for Event-based Optical Flow Estimation

T0 review · 3 major / 8 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read EDCFlow shows that temporally dense feature differences between adjacent event frames, fused with a low-resolution cost volume, give high-accuracy event-based optical flow at 1/4 resolution with lower compute than cost-volume-only methods.

desk verdict Solid event-flow efficiency win with clean ablations; the temporal-offset worry in Eq. (4) cancels under constant velocity, so the main caveats are reproducibility, not correctness. read the letter →

arxiv 2506.03512 v1 pith:2NRJY4DV submitted 2025-06-04 cs.CV

Daikun Liu , Lei Cheng , Teng Wang , changyin Sun This is my paper
classification cs.CV
keywords event-basedopticalflowfeaturedifferencemapscostvolumemulti-scaletemporalaggregationattention-basedfusionhigh-resolutionlightweightnetworkrefinementmodule
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that event-based optical flow can be estimated at 1/4 resolution, with better accuracy and lower compute than cost-volume-only models, by exploiting temporally dense feature differences between adjacent event frames. The authors argue that cost volumes capture global correspondences but are expensive and miss intermediate motion, while feature differences are cheap, detail-rich, and noise-prone, so the two should be fused. Their EDCFlow network computes multi-scale difference maps at high resolution, looks up a low-resolution cost volume, and adaptively combines both motion signals to iteratively refine flow. If the claim holds, it gives event-based flow a better accuracy/efficiency operating point and a plug-and-play refinement for existing recurrent flow networks.

What carries the argument

The central object is the multi-scale temporal feature difference layer. After dividing the event stream into g+1 short time windows and extracting shared-weight features, the layer warps each intermediate feature toward the reference with flow scaled by i/g, reduces channels, and computes dense differences D^s_j = F^l_{(j+1)s} - F^f_{js} at sampling strides s=1,2,5. Depthwise separable 3D convolutions aggregate these maps in space and time, an attention module weighs the scales, and a channel-attention fusion combines the resulting difference motion feature with the upsampled correlation motion feature. This machinery is what lets the network encode continuous intermediate motion at high resolution without building multi-frame high-resolution cost volumes.

What would settle it

Construct or record an event sequence with known strong rotation or acceleration within the 20 ms window and compare EDCFlow's end-point error against a variant whose intermediate warps use a nonlinear motion model; if the linear-warp version does not degrade on such sequences, the assumption survives, and if it degrades sharply, the central mechanism fails exactly where the assumption breaks.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that the complementarity between temporally dense feature differences and a cost volume is enough to make the cost volume need only run at 1/8 resolution while flow is estimated at 1/4 resolution. EDCFlow warps each of the g+1 intermediate event-frame features toward the reference using the linear scaling f_{0->i}=(i/g)f, computes multi-scale difference maps, encodes them with depthwise separable 3D convolutions, and fuses the resulting high-resolution motion features with upsampled correlation features through channel attention. On DSEC this yields an end-point error of 0.72 pixels, angular error 2.65 degrees, and 1-pixel error 10.0, with 2.5M parameters and 247 G MACs, improving on the temporally dense cost-volume method TMA while using less compute, and matching the high-resolution IDNet-4 while being far lighter. The same module appended to RAFT-like event networks at 1/4 resolution improves their motion-boundary detail at modest extra cost.

Load-bearing premise

The load-bearing premise is that within a short event window (20 ms on DSEC, 10/40 ms on MVSEC) motion is linear, so each intermediate feature map can be warped by a fraction i/g of the final flow; under rotation, acceleration, or occlusion inside that window the warped features misalign and the difference maps encode spurious motion.

Editorial extensions

If this is right

  • High-resolution (1/4) event flow is achievable with a low-resolution (1/8) cost volume when dense feature differences supply the local motion detail.
  • Temporally dense cost volumes are not necessary: O(T N C) difference maps can replace O(T N^2 C) correlation searches without losing accuracy.
  • The learned fusion generalizes from simulation (Blinkflow) to real driving data with the smallest accuracy drop among compared methods.
  • Appending EDCFlow as a refinement stage after E-RAFT or TMA reduces EPE and sharpens motion boundaries at 1/4 resolution with only a modest increase in parameters and MACs.
  • At equal accuracy, EDCFlow uses about one fifth the computation of IDNet-4 and runs faster, improving deployability on resource-constrained platforms.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same difference-map-plus-cost-volume fusion could be transferred to frame-based optical flow, where high-resolution refinement currently relies on expensive upsampled cost volumes; the linear-warp difference layer would need no temporal event splitting.
  • The linear-motion assumption in Eq. (4) is the main fragility; a natural test is whether fitting a second-order (accelerated) motion model for the intermediate warps reduces error on fast rotational or oscillatory sequences.
  • The method suggests a cheaper alternative to multi-frame cost volumes in other event tasks such as depth or ego-motion estimation, where dense temporal differences can encode motion between reference states.
  • Because the difference layer's cost scales as O(T N C) and the cost volume is kept at 1/8, further resolution increases become feasible if the correlation upsampling noise noted in the ablation is addressed.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 8 minor

Summary. The paper proposes EDCFlow, an event-based optical flow network that fuses high-resolution (1/4) temporally dense feature difference maps with a low-resolution (1/8) cost volume. The target event stream is split into g time windows; each window's features are warped to the reference using a flow scaled by the temporal index, multi-scale difference maps are computed and adaptively fused with correlation features, and a GRU iteratively updates the flow. Experiments on DSEC and MVSEC report state-of-the-art or comparable accuracy together with lower model size and MACs than prior RAFT-like event-based methods, and the method is also proposed as a plug-and-play refinement module for existing RAFT-like networks.

Significance. If the reported results hold, EDCFlow provides a useful accuracy/efficiency operating point for event-based optical flow, in particular by estimating directly at 1/4 resolution without a high-resolution cost volume. The ablation study is a genuine strength: each component (difference branch, correlation branch, multi-scale attention, and channel attention) is shown to contribute, and the plug-and-play refinement experiments are practically valuable. The main caveats are that the core linear-motion warp underlying the difference layer is under-validated, and the reported accuracy differences are small and come from single-run evaluations. These issues are addressable with targeted experiments and uncertainty quantification, so the central claim is plausible but not yet fully established.

major comments (3)
  1. [3.3, Eq. (4)] The linear-motion scaling f^{k-1}_{0->i} = (i/g) f^{k-1} is the mechanism that warps every intermediate feature map F_i to the reference before computing difference maps, so it is load-bearing for the central claim. The temporal alignment of v_0 and v_1,...,v_g is never defined explicitly: the paper does not state how dt relates to g or where the temporal centroid of each window lies. On the natural reading (v_0 over [t-dt, t] and v_1,...,v_g partitioning [t, t+1] with g*dt = 1), the specific worry that v_0's centroid is offset from t does not produce a systematic error under constant-velocity motion, because f_{0->g} spans the same one-unit duration as the supervised flow f^{t->t+1}. However, the constant-velocity assumption itself is load-bearing and is only acknowledged as a future-work limitation. I ask for (a) an explicit statement of the temporal indexing and the exact relationship between f_{0->g} and the supervised flow, and (b) a quantitative validation on trajectories with acceleration or rotation, for example synthetic events with known nonlinear motion or a variant that estimates per-window flows instead of using Eq. (4), to show that the difference signal does not degrade sharply when the linearity assumption is violated.
  2. [Tables 1, 4-6] All quantitative claims rest on single-run evaluations. The DSEC improvement over TMA is 0.72 vs 0.74 EPE, and the ablation gains (W/o Diff 0.82, W/o Corr 0.83, W/o SE 0.74, W/o MSAttn 0.74, Ours 0.72) are of similar or smaller magnitude. Without error bars or multiple seeds, it is not possible to tell whether the reported ordering is significant. Since the public DSEC benchmark does not allow repeated submissions, please report mean and standard deviation over at least three seeds on a local held-out split (as in the supplementary error-map experiment) for the main DSEC numbers and for the core ablations, and state the evaluation protocol explicitly.
  3. [Table 1] The efficiency comparison is central to the claim of 'better performance with lower complexity,' but the paper does not state how MACs and runtime were obtained for the baselines. Differences in input resolution, implementation, or source (original paper vs re-measurement) can change MACs materially, and the table includes both measured and claimed numbers without distinction. Specify for each row whether the number is from the original paper or measured here, the exact input size used for MAC/runtime computation, and the runtime measurement protocol (GPU, batch size, number of runs).
minor comments (8)
  1. [3.1] The sentence 'We briefly review RFAT [35]' contains a typo; it should be 'RAFT'.
  2. [3.2] The sentence 'along with the reference event stream E_{t->t+1}' should read 'E_{t-dt->t}' to match the problem setup.
  3. [3.4, Eq. (7)] The loss exponent uses the index i, but the summation index is k; it should be 0.8^{K-k}.
  4. [Table 3] The row for 'IDNet-4' cites [21] but should cite [39], since IDNet is [39].
  5. [Table 3] The 'Reduc.' column shows values such as -0.61, which are absolute EPE gaps rather than percentage reductions; please define this column clearly.
  6. [Table 2] The rows 'Ours (+in1/2/3)' are not directly comparable with baselines trained only on outdoor day2; if kept, mark them with a separate symbol and state that the day2-only rows are the ones used for direct comparison.
  7. [Tables 1 and 3] Please state in the captions whether MACs and runtime are measured for all rows under identical conditions or taken from the original publications, and specify the input resolution used.
  8. [Table 6] The abbreviations 'Add.', 'Concat', 'GRU', and 'DWConv3d' in the first column are not defined in the table caption or the main text; please define them.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the flow predictions are supervised against held-out ground truth, and neither the difference maps nor the cost volume are defined in terms of the target flow.

full rationale

The paper's derivation chain is self-contained. The final flow f^k is produced by a GRU from two motion feature sources: correlation motion features F^C looked up from the standard RAFT-style 4D cost volume C = F_0 F_g^T / sqrt(d) (Eq. 3), which is built directly from input voxel features, and difference motion features F^D from maps D^s_j = F^l_{(j+1)s} - F^f_{js} (Eq. 5), computed from fixed input features warped by the current iterate f^{k-1} via the linear scaling of Eq. (4). Neither input is defined in terms of the target flow: the cost volume is a function of events only, and the difference maps are constructed from input features; the dependence of the warps and lookups on f^{k-1} is the standard RAFT-style iterative refinement, in which the network learns a residual, not a circular re-statement of the input. Training uses a supervised L1 loss against ground-truth flow (Eq. 7) with exponentially increasing weights, and evaluation is on held-out DSEC and MVSEC splits plus a Blinkflow-to-DSEC sim-to-real transfer (Table 3), so no fitted parameter is relabeled as a prediction. The paper contains no self-citations: no reference in [1]-[48] is authored by Liu, Cheng, Wang, or Sun, so no load-bearing premise rests on a self-citation chain, and no uniqueness theorem is imported. Ablations (Tables 4-6) removing the difference layer, the correlation encoder, or the attention modules degrade accuracy by 14-15%, which is the opposite of a trivial reduction. The only notable caveat is the linear-motion assumption in Eq. (4), which the authors explicitly flag in the supplementary Future Work section ("we assume linear motion within short time windows"); if the temporal centroids of the reference window v_0 and the g partitions of E_{t->t+1} are offset, that scaling may be systematically biased. That is a correctness/robustness risk to be checked, not a circularity, because prediction quality is still measured against external ground truth.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim rests on standard learning assumptions plus the explicit linear-motion warping assumption and several hand-selected architectural hyperparameters. No new physical entities are introduced.

free parameters (6)
  • g (number of temporal windows) = 5
    Only 5 was used in final model; ablations in Table 10 compared 1, 3, 5, 15 and selected 5 on the DSEC public benchmark.
  • B (temporal bins per voxel grid) = 3 for DSEC and MVSEC dt=4; 1 for MVSEC dt=1
    Selected via Table 10; balances temporal resolution and computational cost.
  • s (multi-scale sampling strides) = [1, 2, 5]
    Ablation Table 5 shows the combination 1,2,5 gives EPE 0.72 vs 0.77/0.74/0.79 for single scales.
  • r (channel reduction ratio) = 1
    Table 12 compares r=1,2,8 and 'w/o'; r=1 selected as best accuracy on benchmark.
  • K (GRU iterations) = 6
    Table 9 shows EPE stabilizes at 6 iterations (0.72) with no gain at 8.
  • GRU hidden dimension = 96
    Stated in implementation details; not ablated.
assumptions (4)
  • domain assumption Linear motion within each short time window
    Invoked in Sec 3.3, Eq. (4): f_{0->i} = (i/g) f. Used to warp intermediate feature maps before computing differences; fails under acceleration, rotation, or occlusion.
  • domain assumption Voxel grid with bilinear temporal interpolation represents event streams
    Equation (1)-(2), standard representation from Zhu et al. [47]; used to convert events into CNN input.
  • domain assumption Event camera measures brightness change with polarity and timestamp
    Underlies the event representation and the notion of difference maps; standard in event-based vision.
  • standard math Cost volume correlation computed as dot product normalized by sqrt(d_bar)
    Equation (3), standard RAFT construction; used for low-resolution long-range matching.

how reviews work

0 comments
Cite this review

Pith. "Pith review of EDCFlow: Exploring Temporally Dense Difference Maps for Event-based Optical Flow Estimation." pith.science (2026). https://pith.science/paper/2NRJY4DV

@misc{pith2026250603512,
  author       = {Pith},
  title        = {Pith review of: EDCFlow: Exploring Temporally Dense Difference Maps for Event-based Optical Flow Estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2NRJY4DV}},
  note         = {Machine review of arXiv:2506.03512}
}
read the original abstract

Recent learning-based methods for event-based optical flow estimation utilize cost volumes for pixel matching but suffer from redundant computations and limited scalability to higher resolutions for flow refinement. In this work, we take advantage of the complementarity between temporally dense feature differences of adjacent event frames and cost volume and present a lightweight event-based optical flow network (EDCFlow) to achieve high-quality flow estimation at a higher resolution. Specifically, an attention-based multi-scale temporal feature difference layer is developed to capture diverse motion patterns at high resolution in a computation-efficient manner. An adaptive fusion of high-resolution difference motion features and low-resolution correlation motion features is performed to enhance motion representation and model generalization. Notably, EDCFlow can serve as a plug-and-play refinement module for RAFT-like event-based methods to enhance flow details. Extensive experiments demonstrate that EDCFlow achieves better performance with lower complexity compared to existing methods, offering superior generalization.

Figures

Figures reproduced from arXiv: 2506.03512 by the authors.

Figure 1
Figure 1. Illustration of difference between correlation and differ [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. a. Overview of EDCFlow, including three main components: 1) Feature Extraction. Feature and context encoder extract features from input. 2) Motion Encoding. A correlation encoder and multi-scale temporal feature difference layer, along with an attention layer, are utilized to iteratively generate representative motion features. 3) Flow Updates. A GRU recurrently updates the flow by decoding the fused motion feature.… view at source ↗
Figure 3
Figure 3. Qualitative results of flow predictions on DSEC [11]. Notable areas are marked by bounding boxes. Please zoom in for details. frame apart) and dt = 4 (4 grayscale image frames apart), using the outdoor day2 sequence for training and testing on indoor flying sequences and 800 samples of the out￾door day1. We evaluate flow accuracy using end-point￾error (EPE) on both datasets. For MVSEC, we additionally measure the pe… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Refinement strategy in existing RAFT-like methods. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 6
Figure 6. Figure 6: Illustration of motion feature maps. (a) Event data. [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 5
Figure 5. Figure 5: End-point-error (px) on DSEC vs. computational com [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 7
Figure 7. Figure 7: Qualitative results on the outdoor day1 sequence on MVSEC [45]. Please zoom in for details. EPE=0.32 EPE=0.30 EPE=0.47 Reference image Events Ground truth Output flow Error distribution map EPE=3.32 [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]
Figure 8
Figure 8. Figure 8: Visualization of error distribution maps. We present three high-quality flow estimation results with smaller EPE (the first three [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

52 extracted references · 45 canonical work pages

  1. [1]

    Asynchronous frameless event-based optical flow.Neural Networks, 27:32–37, 2012

    Ryad Benosman, Sio-Hoi Ieng, Charles Clercq, Chiara Bar- tolozzi, and Mandyam Srinivasan. Asynchronous frameless event-based optical flow.Neural Networks, 27:32–37, 2012. 2

    work page 2012

  2. [2]

    Real- time optical flow for vehicular perception with low-and high- resolution event cameras.IEEE Transactions on Intelligent Transportation Systems, 23(9):15066–15078, 2021

    Vincent Brebion, Julien Moreau, and Franck Davoine. Real- time optical flow for vehicular perception with low-and high- resolution event cameras.IEEE Transactions on Intelligent Transportation Systems, 23(9):15066–15078, 2021. 6

    work page 2021

  3. [3]

    A differentiable recurrent surface for asynchronous event-based data

    Marco Cannici, Marco Ciccone, Andrea Romanoni, and Matteo Matteucci. A differentiable recurrent surface for asynchronous event-based data. InECCV, pages 136–152. Springer, 2020. 2

    work page 2020

  4. [4]

    Explicit motion disen- tangling for efficient optical flow estimation

    Changxing Deng, Ao Luo, Haibin Huang, Shaodan Ma, Jiangyu Liu, and Shuaicheng Liu. Explicit motion disen- tangling for efficient optical flow estimation. InProceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 9521–9530, 2023. 1

    work page 2023

  5. [5]

    Spatio-temporal recurrent networks for event-based optical flow estimation

    Ziluo Ding, Rui Zhao, Jiyuan Zhang, Tianxiao Gao, Ruiqin Xiong, Zhaofei Yu, and Tiejun Huang. Spatio-temporal recurrent networks for event-based optical flow estimation. InProceedings of the AAAI conference on artificial intelli- gence, pages 525–533, 2022. 2, 6

    work page 2022

  6. [6]

    Flownet: Learning optical flow with convolutional networks

    Alexey Dosovitskiy, Philipp Fischer, Eddy Ilg, Philip Hausser, Caner Hazirbas, Vladimir Golkov, Patrick Van Der Smagt, Daniel Cremers, and Thomas Brox. Flownet: Learning optical flow with convolutional networks. InPro- ceedings of the IEEE international conference on computer vision, pages 2758–2766, 2015. 1

    work page 2015

  7. [7]

    A unifying contrast maximization framework for event cam- eras, with applications to motion, depth, and optical flow estimation

    Guillermo Gallego, Henri Rebecq, and Davide Scaramuzza. A unifying contrast maximization framework for event cam- eras, with applications to motion, depth, and optical flow estimation. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 3867–3876,

  8. [8]

    Davison, J ¨org Conradt, Kostas Daniilidis, and Davide Scaramuzza

    Guillermo Gallego, Tobi Delbr ¨uck, Garrick Orchard, Chiara Bartolozzi, Brian Taba, Andrea Censi, Stefan Leutenegger, Andrew J. Davison, J ¨org Conradt, Kostas Daniilidis, and Davide Scaramuzza. Event-based vision: A survey.IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(1):154–180, 2022. 1

    work page 2022

Show all 52 references
  1. [9]

    Im2flow: Motion hallucination from static images for action recogni- tion

    Ruohan Gao, Bo Xiong, and Kristen Grauman. Im2flow: Motion hallucination from static images for action recogni- tion. InCVPR, pages 5937–5947, 2018. 1

    2018

  2. [10]

    End-to-end learning of repre- sentations for asynchronous event-based data

    Daniel Gehrig, Antonio Loquercio, Konstantinos G Derpa- nis, and Davide Scaramuzza. End-to-end learning of repre- sentations for asynchronous event-based data. InProceed- ings of the IEEE/CVF International Conference on Com- puter Vision, pages 5633–5643, 2019. 2

    2019

  3. [11]

    Dsec: A stereo event camera dataset for driv- ing scenarios.IEEE Robotics and Automation Letters, 6(3): 4947–4954, 2021

    Mathias Gehrig, Willem Aarents, Daniel Gehrig, and Davide Scaramuzza. Dsec: A stereo event camera dataset for driv- ing scenarios.IEEE Robotics and Automation Letters, 6(3): 4947–4954, 2021. 1, 2, 4, 5, 7

    2021

  4. [12]

    E-raft: Dense optical flow from event cam- eras

    Mathias Gehrig, Mario Millh ¨ausler, Daniel Gehrig, and Da- vide Scaramuzza. E-raft: Dense optical flow from event cam- eras. InInternational Conference on 3D Vision (3DV), 2021. 1, 2, 3, 4, 5, 6, 7, 8

    2021

  5. [13]

    Dense continuous-time optical flow from event cameras

    Mathias Gehrig, Manasi Muglikar, and Davide Scaramuzza. Dense continuous-time optical flow from event cameras. IEEE Transactions on Pattern Analysis and Machine Intel- ligence, 2024. 1, 2, 5, 6

    2024

  6. [14]

    Squeeze-and-excitation net- works

    Jie Hu, Li Shen, and Gang Sun. Squeeze-and-excitation net- works. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 7132–7141, 2018. 4

    2018

  7. [15]

    Flowformer: A transformer architecture for optical flow

    Zhaoyang Huang, Xiaoyu Shi, Chao Zhang, Qiang Wang, Ka Chun Cheung, Hongwei Qin, Jifeng Dai, and Hongsheng Li. Flowformer: A transformer architecture for optical flow. InEuropean conference on computer vision, pages 668–685. Springer, 2022. 2

    2022

  8. [16]

    Ccmr: high resolution optical flow estimation via coarse-to-fine context-guided motion reasoning

    Azin Jahedi, Maximilian Luz, Marc Rivinius, and Andr ´es Bruhn. Ccmr: high resolution optical flow estimation via coarse-to-fine context-guided motion reasoning. InProceed- ings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 6899–6908, 2024. 1, 2

    2024

  9. [17]

    Learning to estimate hidden motions with global motion aggregation

    Shihao Jiang, Dylan Campbell, Yao Lu, Hongdong Li, and Richard Hartley. Learning to estimate hidden motions with global motion aggregation. InProceedings of the IEEE/CVF international conference on computer vision, pages 9772– 9781, 2021. 2

    2021

  10. [18]

    Tea: Temporal excitation and aggregation for action recognition

    Yan Li, Bin Ji, Xintian Shi, Jianguo Zhang, Bin Kang, and Limin Wang. Tea: Temporal excitation and aggregation for action recognition. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 909–918, 2020. 2

    2020

  11. [19]

    Blinkflow: A dataset to push the limits of event-based optical flow estimation

    Yijin Li, Zhaoyang Huang, Shuo Chen, Xiaoyu Shi, Hong- sheng Li, Hujun Bao, Zhaopeng Cui, and Guofeng Zhang. Blinkflow: A dataset to push the limits of event-based optical flow estimation. In2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages...

    2023

  12. [20]

    Flow-guided sparse transformer for video deblurring.arXiv preprint arXiv:2201.01893,

    Jing Lin, Yuanhao Cai, Xiaowan Hu, Haoqian Wang, You- liang Yan, Xueyi Zou, Henghui Ding, Yulun Zhang, Radu Timofte, and Luc Van Gool. Flow-guided sparse transformer for video deblurring.arXiv preprint arXiv:2201.01893,

    arXiv

  13. [21]

    Tma: Temporal motion aggregation for event-based optical flow

    Haotian Liu, Guang Chen, Sanqing Qu, Yanping Zhang, Zhi- jun Li, Alois Knoll, and Changjun Jiang. Tma: Temporal motion aggregation for event-based optical flow. InProceed- ings of the IEEE/CVF International Conference on Com- puter Vision, pages 9685–9694, 2023. 1, 2, 4, 5, 6, 7, 8

    2023

  14. [22]

    Block-matching optical flow for dynamic vision sensors: Algorithm and fpga implemen- tation

    Min Liu and Tobi Delbruck. Block-matching optical flow for dynamic vision sensors: Algorithm and fpga implemen- tation. In2017 IEEE International Symposium on Circuits and Systems (ISCAS), pages 1–4. IEEE, 2017. 2

    2017

  15. [23]

    Abmof: A novel optical flow algorithm for dynamic vision sensors.arXiv preprint arXiv:1805.03988, 2018

    Min Liu and Tobi Delbruck. Abmof: A novel optical flow algorithm for dynamic vision sensors.arXiv preprint arXiv:1805.03988, 2018. 2

    2018 arXiv

  16. [24]

    Learning by distillation: a self-supervised learning frame- work for optical flow estimation.IEEE transactions on pat- tern analysis and machine intelligence, 44(9):5026–5041,

    Pengpeng Liu, Michael R Lyu, Irwin King, and Jia Xu. Learning by distillation: a self-supervised learning frame- work for optical flow estimation.IEEE transactions on pat- tern analysis and machine intelligence, 44(9):5026–5041,

  17. [25]

    Oiflow: Occlusion-inpainting op- 9 tical flow estimation by unsupervised learning.IEEE Trans- actions on Image Processing, 30:6420–6433, 2021

    Shuaicheng Liu, Kunming Luo, Nianjin Ye, Chuan Wang, Jue Wang, and Bing Zeng. Oiflow: Occlusion-inpainting op- 9 tical flow estimation by unsupervised learning.IEEE Trans- actions on Image Processing, 30:6420–6433, 2021. 2

    2021

  18. [26]

    Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017

    I Loshchilov. Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017. 5

    2017 arXiv

  19. [27]

    An iterative image reg- istration technique with an application to stereo vision

    Bruce D Lucas and Takeo Kanade. An iterative image reg- istration technique with an application to stereo vision. In IJCAI’81: 7th international joint conference on Artificial in- telligence, pages 674–679, 1981. 2

    1981

  20. [28]

    Single image optical flow estimation with an event camera

    Liyuan Pan, Miaomiao Liu, and Richard Hartley. Single image optical flow estimation with an event camera. In 2020 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), pages 1669–1678. IEEE, 2020. 2

    2020

  21. [29]

    Taming contrast max- imization for learning sequential, low-latency, event-based optical flow

    Federico Paredes-Vall ´es, Kirk YW Scheper, Christophe De Wagter, and Guido CHE De Croon. Taming contrast max- imization for learning sequential, low-latency, event-based optical flow. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 9695–9705, ...

    2023

  22. [30]

    Secrets of event-based optical flow

    Shintaro Shiba, Yoshimitsu Aoki, and Guillermo Gallego. Secrets of event-based optical flow. InEuropean Conference on Computer Vision, pages 628–645. Springer, 2022. 2, 5, 6

    2022

  23. [31]

    Super-convergence: Very fast training of neural networks using large learn- ing rates

    Leslie N Smith and Nicholay Topin. Super-convergence: Very fast training of neural networks using large learn- ing rates. InArtificial intelligence and machine learning for multi-domain operations applications, pages 369–386. SPIE, 2019. 5

    2019

  24. [32]

    Simultaneous opti- cal flow and segmentation (sofas) using dynamic vision sen- sor.arXiv preprint arXiv:1805.12326, 2018

    Timo Stoffregen and Lindsay Kleeman. Simultaneous opti- cal flow and segmentation (sofas) using dynamic vision sen- sor.arXiv preprint arXiv:1805.12326, 2018. 2

    2018 arXiv

  25. [33]

    Craft: Cross- attentional flow transformer for robust optical flow

    Xiuchao Sui, Shaohua Li, Xue Geng, Yan Wu, Xinxing Xu, Yong Liu, Rick Goh, and Hongyuan Zhu. Craft: Cross- attentional flow transformer for robust optical flow. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17602–17611, 2022. 2

    2022

  26. [34]

    Pwc-net: Cnns for optical flow using pyramid, warping, and cost volume

    Deqing Sun, Xiaodong Yang, Ming-Yu Liu, and Jan Kautz. Pwc-net: Cnns for optical flow using pyramid, warping, and cost volume. InCVPR, pages 8934–8943, 2018. 1, 2

    2018

  27. [35]

    Raft: Recurrent all-pairs field transforms for optical flow

    Zachary Teed and Jia Deng. Raft: Recurrent all-pairs field transforms for optical flow. InECCV, pages 402–419. Springer, 2020. 1, 2, 3, 4

    2020

  28. [36]

    Unsupervised learning of optical flow with cnn-based non-local filtering.IEEE Transactions on Image Processing, 29:8429–8442, 2020

    Long Tian, Zhigang Tu, Dejun Zhang, Jun Liu, Baoxin Li, and Junsong Yuan. Unsupervised learning of optical flow with cnn-based non-local filtering.IEEE Transactions on Image Processing, 29:8429–8442, 2020. 2

    2020

  29. [37]

    Learning dense and continuous optical flow from an event camera.IEEE Transactions on Image Processing, 31:7237–7251, 2022

    Zhexiong Wan, Yuchao Dai, and Yuxin Mao. Learning dense and continuous optical flow from an event camera.IEEE Transactions on Image Processing, 31:7237–7251, 2022. 1, 2, 6

    2022

  30. [38]

    Flow dynamics correction for action recognition

    Lei Wang and Piotr Koniusz. Flow dynamics correction for action recognition. InICASSP 2024-2024 IEEE Interna- tional Conference on Acoustics, Speech and Signal Process- ing (ICASSP), pages 3795–3799. IEEE, 2024. 1

    2024

  31. [39]

    Lightweight event-based optical flow estimation via iterative deblurring

    Yilun Wu, Federico Paredes-Vall ´es, and Guido CHE de Croon. Lightweight event-based optical flow estimation via iterative deblurring. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 14708–14715. IEEE, 2024. 1, 2, 4, 5, 6

    2024

  32. [40]

    Unifying flow, stereo and depth estimation.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(11):13941– 13958, 2023

    Haofei Xu, Jing Zhang, Jianfei Cai, Hamid Rezatofighi, Fisher Yu, Dacheng Tao, and Andreas Geiger. Unifying flow, stereo and depth estimation.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(11):13941– 13958, 2023. 1, 2

    2023

  33. [41]

    V olumetric correspon- dence networks for optical flow.Advances in neural infor- mation processing systems, 32, 2019

    Gengshan Yang and Deva Ramanan. V olumetric correspon- dence networks for optical flow.Advances in neural infor- mation processing systems, 32, 2019. 2

    2019

  34. [42]

    Folt: Fast multiple object tracking from uav- captured videos based on optical flow

    Mufeng Yao, Jiaqi Wang, Jinlong Peng, Mingmin Chi, and Chao Liu. Folt: Fast multiple object tracking from uav- captured videos based on optical flow. InProceedings of the 31st ACM International Conference on Multimedia, pages 3375–3383, 2023. 1

    2023

  35. [43]

    Spatio- temporal deformable attention network for video deblurring

    Huicong Zhang, Haozhe Xie, and Hongxun Yao. Spatio- temporal deformable attention network for video deblurring. InECCV, pages 581–596. Springer, 2022. 1

    2022

  36. [44]

    Dip: Deep inverse patch- match for high-resolution optical flow

    Zihua Zheng, Ni Nie, Zhi Ling, Pengfei Xiong, Jiangyu Liu, Hao Wang, and Jiankun Li. Dip: Deep inverse patch- match for high-resolution optical flow. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8925–8934, 2022. 1, 2

    2022

  37. [45]

    The multi- vehicle stereo event camera dataset: An event camera dataset for 3d perception.IEEE Robotics and Automation Letters, 3 (3):2032–2039, 2018

    Alex Zihao Zhu, Dinesh Thakur, Tolga ¨Ozaslan, Bernd Pfrommer, Vijay Kumar, and Kostas Daniilidis. The multi- vehicle stereo event camera dataset: An event camera dataset for 3d perception.IEEE Robotics and Automation Letters, 3 (3):2032–2039, 2018. 2, 4, 6, 7, 1, 3

    2018

  38. [46]

    Ev-flownet: Self-supervised optical flow estimation for event-based cameras.arXiv preprint arXiv:1802.06898, 2018

    Alex Zihao Zhu, Liangzhe Yuan, Kenneth Chaney, and Kostas Daniilidis. Ev-flownet: Self-supervised optical flow estimation for event-based cameras.arXiv preprint arXiv:1802.06898, 2018. 2, 5, 6

    2018 arXiv

  39. [47]

    Unsupervised event-based learning of optical flow, depth, and egomotion

    Alex Zihao Zhu, Liangzhe Yuan, Kenneth Chaney, and Kostas Daniilidis. Unsupervised event-based learning of optical flow, depth, and egomotion. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 989–997, 2019. 2, 3, 6

    2019

  40. [48]

    Hao Zhuang, Zheng Fang, Xinjie Huang, Kuanxu Hou, Delei Kong, and Chenming Hu. Ev-mgrflownet: Motion-guided recurrent network for unsupervised event-based optical flow with hybrid motion-compensation loss.IEEE Transactions on Instrumentation and Measurement, 2024. 2, 6 10 EDCF...

    2024

  41. [49]

    They primarily perform pixel matching or refinement across multiple spatial reso- lutions in frames

    More comparison with SOTA methods Compared with frame-based methods.Several frame- based SOTA methods [16, 40, 44] focus on achieving high- resolution optical flow estimation. They primarily perform pixel matching or refinement across multiple spatial reso- lutions in frames. ...

  42. [50]

    7 presents a qual- itative comparison of our method with other methods on outdoor day1 sequence of the MVSEC [45]

    More Visualizations Qualitative Results on MVSEC.Fig. 7 presents a qual- itative comparison of our method with other methods on outdoor day1 sequence of the MVSEC [45]. Compared to DSEC dataset [11], MVSEC has lower resolution and sparser events (especially atdt= 1), and it la...

  43. [51]

    w/o” indicating conv1 removed. We setr= 1to bal- ance accuracy and computations. When compared to “w/o

    More Ablation Studies More ablation studies are also conducted on the DSEC dataset [11]. Iterations.Some existing methods [5, 35, 39] achieve bet- ter optical flow results through iterative refinement strate- gies, particularly for small objects with large displacements. The r...

  44. [52]

    Future Work In our work, we assume linear motion within short time windows (20 ms for DSEC and 10/40 ms for MVSEC), which shows good empirical performance and lower com- putational complexity. However, investigating alternative motion models, such as estimating higher temporal...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.