Pith. sign in

REVIEW 3 major objections 8 minor 1 cited by

MEMFOF: High-Resolution Training for Memory-Efficient Multi-Frame Optical Flow Estimation

T0 review · 3 major / 8 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read MEMFOF claims a three-frame optical flow model that runs at native 1080p on 2.09 GB of GPU memory and reaches state-of-the-art accuracy on Spring, Sintel clean, and KITTI-2015.

desk verdict Solid memory-efficiency engineering with a real 2.09 GB result, but the SOTA claims need a fuller comparison table before they hold. read the letter →

arxiv 2506.23151 v1 pith:5257Z5QE submitted 2025-06-29 cs.CV cs.AIcs.MM

Vladislav Bargatin , Egor Chistov , Alexander Yakovenko , Dmitriy Vatolin This is my paper · ORCID
classification cs.CVcs.AIcs.MM
keywords opticalflowmulti-framememory-efficientinferencehigh-resolutiontraining1080pRAFT-stylearchitecturescorrelationvolumebenchmarkevaluation
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

Optical flow, the per-pixel motion between video frames, normally becomes memory-hungry at high resolution because RAFT-style correlation volumes grow quadratically with image size, so many methods downsample or crop to fit. MEMFOF claims a better trade-off: by computing correlation volumes at 1/16 resolution instead of 1/8 and processing three frames with bidirectional flows, it cuts inference memory to 2.09 GB at 1080p while improving accuracy. The authors also argue that training on 2x-bilinearly-upsampled versions of standard datasets, on full frames rather than crops, aligns training motion magnitudes with real FullHD video and prevents underfitting on large motions. If correct, this would be the first multi-frame flow method that combines native FullHD training and runtime with state-of-the-art results on Spring, Sintel clean, and KITTI-2015, reporting 1px 3.289 on Spring, EPE 0.963 on Sintel clean, and Fl-all 2.94% on KITTI-2015.

What carries the argument

The load-bearing mechanism is the reduced-resolution dual correlation volume: two all-pairs correlation volumes at 1/16 input resolution instead of the standard 1/8, which cuts the memory of the two volumes from about 10.4 GB to 0.65 GB at 1080p, with feature dimension raised from 256 to 1024 and update-block dimension raised from 128 to 512 to preserve information. Around this sit three-frame bidirectional flow estimation, in which flows from the central frame to both neighbors are refined jointly, a training protocol that 2x-upsamples datasets and trains on full 1080p frames, and inference-time reuse of feature maps and correlation volumes across consecutive video frames.

What would settle it

Evaluate the 2x-upsampled-trained MEMFOF on a ground-truth optical flow video set whose motion is shifted toward very large displacements, such as fast panning FullHD footage, and compare its 1px and EPE against a model trained on native-resolution data; if the gap over native training shrinks or reverses, the motion-distribution-transfer assumption underpinning the training strategy would be falsified.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a RAFT-style multi-frame optical flow architecture can be made memory-efficient enough to run and train at native 1080p without sacrificing accuracy. Reducing the correlation volume resolution from 1/8 to 1/16, widening feature channels to compensate, and extending SEA-RAFT to estimate bidirectional flows from a triplet of frames cuts runtime memory to 2.09 GB and training memory to 28.5 GB. Combined with a training protocol that bilinearly upsamples existing datasets by 2x and trains on full frames, this yields state-of-the-art results: first place on Spring with a 1px outlier rate of 3.289, EPE 0.963 on Sintel clean, and Fl-all 2.94% on KITTI-2015, outperforming more memory-hungry two-frame and multi-frame methods. The paper attributes the gains to aligning training motion statistics with the FullHD domain and to temporal coherence from three-frame bidirectional processing.

Load-bearing premise

The load-bearing premise is that bilinearly upsampling low-resolution training videos by 2x makes their motion statistics similar enough to real FullHD video that the model learns to handle FullHD motion; if real high-resolution motion lies outside that upsampled range, the accuracy gains may not carry over to other high-resolution domains.

Editorial extensions

If this is right

  • At 2.09 GB inference memory, MEMFOF can run at native 1080p on consumer GPUs, removing the need to downsample or tile high-resolution inputs.
  • Training at native 1080p without cropping lets the model see full-frame global motion context during training, not just local crops.
  • Videos can be processed more cheaply by caching and reusing feature maps and correlation volumes across overlapping frame pairs.
  • Memory efficiency in the RAFT family does not have to come at the cost of accuracy, since MEMFOF beats heavier multi-frame methods on the reported benchmarks.
  • The simple training-only change of upsampling low-resolution data by 2x is a recipe other optical flow methods could adopt to handle large motions at high resolution.

Reading between the lines

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

  • The memory-saving recipe of 1/16 correlation volumes with wider features is separable from the three-frame idea, so dropping it into any RAFT-style two-frame model would make the trade-off independently testable.
  • The authors observe that accuracy degrades when moving from 3 to 5 frames and attribute this to context and update network capacity, which suggests a testable extension: scale those modules with temporal width to see whether longer temporal context helps.
  • The 2x upsampling training trick addresses a distribution mismatch that also affects other high-resolution dense prediction tasks, so it may transfer to stereo matching or video frame interpolation, which face the same low-resolution-to-FullHD gap.
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 introduces MEMFOF, a multi-frame optical flow method that extends SEA-RAFT to three frames with bidirectional flow prediction, reduces the correlation volume to 1/16 resolution, increases feature dimensions, adds a GMA module, and trains on 2x-upsampled datasets with full-frame crops. The central claims are a low GPU memory footprint (2.09 GB at 1080p inference, 28.5 GB for training) and state-of-the-art accuracy on Spring, Sintel clean, and KITTI-2015 benchmarks. The paper includes systematic ablations of training resolution, frame count, correlation volume resolution, and inference-time optimizations, and releases code.

Significance. If the reported memory and accuracy numbers hold, MEMFOF addresses a practical bottleneck: multi-frame optical flow at native FullHD resolution with memory consumption within consumer GPU limits. The training strategy of upsampling standard datasets to better match high-resolution motion distributions is well motivated by the histogram analysis and supported by the ablations. The paper's strength is its systematic empirical evaluation: Tables 4, 5, 6, and 9 cover the key design choices, and the memory reduction is corroborated by the architecture change (1/16 correlation volume) plus the ablation comparison. The code release and official leaderboard use add to reproducibility. However, the state-of-the-art claims depend on the completeness of the benchmark comparisons, which is the main risk.

major comments (3)
  1. [4.2, Tables 2 and 3] The state-of-the-art claims are not fully supported by the tabulated comparisons. The abstract and Section 1 claim that MEMFOF 'achieves state-of-the-art performance' and 'leads' Sintel clean with EPE 0.963 and achieves the 'best Fl-all error' on KITTI-2015 at 2.94%, but Tables 2 and 3 list only a curated subset of methods. Methods cited elsewhere in the paper, such as FlowFormer++, DIP, and GMFlow, are absent from both tables. The official Spring, Sintel, and KITTI leaderboards contain far more submissions, so the assertions of 'first', 'leads', and 'best' are not verifiable from the manuscript alone. This is load-bearing because the 'without sacrificing accuracy' part of the central contribution rests directly on these benchmark numbers. Please either include all published results for the listed benchmarks (or explicitly state that the rankings refer to the leaderboard at the time of submission and cite the leaderboard entries), or qualify the claims to be relative to the methods actually compared.
  2. [4.1 and Tables 1, 2, 5, 6] The memory and runtime numbers are reported as single measurements without a description of the measurement protocol, variance, or number of runs. For example, the headline '2.09 GB' inference memory and '28.5 GB' training memory are cited in the abstract, but the paper does not state whether these are peak allocated memory, reserved memory, or measured with torch.cuda.max_memory_allocated, nor does it indicate how many repetitions were averaged. Since the memory-efficiency claim is a central contribution, the measurement methodology should be specified so the results are reproducible; reporting at least the protocol and ideally mean+-std across runs would strengthen the paper.
  3. [Abstract, 3.2, and Table 1] The claim that training is performed 'without the need for cropping or downsampling' is contradicted by Table 1. The table lists explicit crop sizes for every training stage (e.g., [480,960] for TartanAir, [864,1920] for Things and TSKH, [872,1920] for Sintel-ft, [750,1920] for KITTI-ft); only the Spring-ft stage uses [1080,1920], which matches a full 1080p frame. Thus the full-frame, no-crop training applies only to the final fine-tuning stage. Please reword the abstract and Section 3.2 to accurately describe that the method can fine-tune at native 1080p without cropping or downsampling, while the pre-training stages still use crops.
minor comments (8)
  1. [4.3, Figure 4 caption] Typo: 'Plese refer' should be 'Please refer'.
  2. [4.1, evaluation metrics] The metric is called 'Fl-score' in the text, while Table 3 uses 'Fl-all' and the KITTI benchmark uses 'Fl-all'; please use consistent terminology and define the relationship.
  3. [Table 8 caption] Typo: 'sumbissions' should be 'submissions'.
  4. [6. Acknowledgments] Typo: 'identifer' should be 'identifier'.
  5. [3.2] The strided convolution applied to the 1/8-resolution features is not fully specified; please provide the kernel size, stride, and output channel details, since the feature dimension increase (Df from 256 to 1024) is a key part of the design.
  6. [3.1, context network] The context network is described only by its inputs and outputs; a brief architectural description (or an explicit reference to the SEA-RAFT context network) would help readers understand the three-frame extension.
  7. [1 or 4.2] The statement that this is the 'first multi-frame optical flow method designed for memory efficiency at FullHD' is a strong novelty claim; given that StreamFlow also focuses on streamlined multi-frame processing, it would be more accurate to say 'to our knowledge, the first multi-frame method to operate at FullHD with a few GB of memory', with a discussion of how MEMFOF differs.
  8. [4.2, KITTI/Sintel protocol] For Sintel and KITTI, the paper states that inputs are bilinearly upscaled by 2x and outputs downscaled by 2x; please discuss whether this pre-processing could put MEMFOF at an advantage or disadvantage relative to methods operating at native resolution, and whether the reported memory numbers apply to the upscaled inference or native resolution.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's contributions are empirical and validated on external benchmarks and ablations, with no derivation that reduces to its inputs.

full rationale

MEMFOF is an empirical systems paper: its central claims are (1) a memory-efficient multi-frame architecture derived from SEA-RAFT with reduced correlation volumes, (2) a high-resolution training protocol using 2x-upsampled datasets, and (3) benchmark results on Spring, Sintel, and KITTI-2015. None of these claims is established by defining a quantity in terms of the target result. The memory numbers (2.09 GB inference, 28.5 GB training) are direct measurements on the proposed architecture and code, not fitted parameters renamed as predictions. The high-resolution training strategy is justified by motion histograms (Figure 4) and by ablations on the Spring training split (Table 4), which is legitimate experimental validation rather than circular reasoning. The benchmark results are quoted from external leaderboards; whether the comparison set is complete is a correctness/verifiability concern, not a circularity concern. No load-bearing self-citations, imported uniqueness theorems, or ansatz-smuggling citations were found. The paper explicitly states that its main changes relative to SEA-RAFT are skipping FlyingChairs, 2x upsampling, and larger crops, and it ablat4es these choices against alternatives; the evaluation is therefore self-contained and externally grounded.

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

The method depends on several hand-chosen hyperparameters (correlation scale, feature dimensions, upsampling factor, number of iterations) and domain assumptions about the benefits of multi-frame processing, bidirectional flows, and upsampled training. These are all ablated on the Spring train split, providing empirical support, but they are assumptions rather than proven properties. No new entities are introduced.

free parameters (6)
  • Correlation volume scale = 1/16 of input resolution
    Selected over 1/8 and 1/24 based on ablation in Table 5; reduces memory from 10.4 GB to 0.65 GB for two correlation volumes.
  • Feature dimension D_f = 1024
    Increased from 256 in SEA-RAFT to compensate for reduced correlation volume resolution; chosen via ablations.
  • Update block dimension D_c = 512
    Increased from 128; ablation shows 1px improves from 3.56 to 3.06 when going from 128 to 512.
  • Training data upsampling factor = 2x
    Bilinear upsampling of training data and flows; ablation shows full-frame upsampled training gives EPE 0.341 vs 0.430 at native resolution.
  • Number of iterative refinements N = 8
    Chosen as balance between speed (472 ms) and accuracy (3.061 1px); Table 7 shows diminishing returns after 8.
  • Mixture-of-Laplace gamma = 0.85
    Follows RAFT/SEA-RAFT; weights later predictions higher.
assumptions (4)
  • domain assumption Temporal information from adjacent frames improves optical flow accuracy.
    The multi-frame design assumes that using three frames helps estimate bidirectional flows better than two frames; Table 5 shows 3-frame (3.061) beats 2-frame (3.375) at the same D_c=512, supporting the assumption.
  • domain assumption Bidirectional flows share consistent boundaries of the central frame, which makes initial flow prediction easier.
    Proposed as explanation for why bidirectional training improves EPE by 14.75% over unidirectional. This is a hypothesis, not independently proved.
  • domain assumption 2x upsampling of low-resolution training data aligns motion distributions with FullHD inference.
    Used to justify training on 2x upsampled TartanAir, Things, KITTI, HD1K, and Sintel. Histograms in Figure 4 show improved overlap with Spring motion ranges, but this is a modeling assumption about distribution transfer.
  • ad hoc to paper A correlation volume at 1/16 resolution with higher-dimensional features can retain as much correspondence information as the standard 1/8 volume.
    The memory reduction relies on this trade-off. The ablations support it for tested datasets, but it is a design choice specific to this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MEMFOF: High-Resolution Training for Memory-Efficient Multi-Frame Optical Flow Estimation." pith.science (2026). https://pith.science/paper/5257Z5QE

@misc{pith2026250623151,
  author       = {Pith},
  title        = {Pith review of: MEMFOF: High-Resolution Training for Memory-Efficient Multi-Frame Optical Flow Estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5257Z5QE}},
  note         = {Machine review of arXiv:2506.23151}
}
read the original abstract

Recent advances in optical flow estimation have prioritized accuracy at the cost of growing GPU memory consumption, particularly for high-resolution (FullHD) inputs. We introduce MEMFOF, a memory-efficient multi-frame optical flow method that identifies a favorable trade-off between multi-frame estimation and GPU memory usage. Notably, MEMFOF requires only 2.09 GB of GPU memory at runtime for 1080p inputs, and 28.5 GB during training, which uniquely positions our method to be trained at native 1080p without the need for cropping or downsampling. We systematically revisit design choices from RAFT-like architectures, integrating reduced correlation volumes and high-resolution training protocols alongside multi-frame estimation, to achieve state-of-the-art performance across multiple benchmarks while substantially reducing memory overhead. Our method outperforms more resource-intensive alternatives in both accuracy and runtime efficiency, validating its robustness for flow estimation at high resolutions. At the time of submission, our method ranks first on the Spring benchmark with a 1-pixel (1px) outlier rate of 3.289, leads Sintel (clean) with an endpoint error (EPE) of 0.963, and achieves the best Fl-all error on KITTI-2015 at 2.94%. The code is available at https://github.com/msu-video-group/memfof.

Figures

Figures reproduced from arXiv: 2506.23151 by the authors.

Figure 1
Figure 1. Comparison with state-of-the-art optical flow meth [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of our method and FullHD inference results. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Qualitative comparison of MemFlow [4], StreamFlow [30], SEA-RAFT [32], and our method on Spring benchmark [18] crops, colorbar represents endpoint error. Our approach surpasses prior methods and demonstrates that: (1) multi-frame processing enhances temporal coherence, and (2) native Full HD resolution preserves local and global motion details. Crops are sourced from official leaderboard submissions. f k t→t+1; H an… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: We analyze motion patterns in optical flow datasets us [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Qualitative comparison of MemFlow-T, SEA-RAFT (L), and our method on the Sintel benchmark. Sourced from official [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: Qualitative comparison of MemFlow-T, SEA-RAFT (L), and our method on the KITTI-2015 benchmark. Sourced from official [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. MegaFlow: Zero-Shot Large Displacement Optical Flow

    cs.CV 2026-03 accept novelty 6.0 of 10

    MegaFlow reaches SOTA zero-shot optical flow (especially large motions) and competitive point tracking by global matching of pre-trained ViT features followed by lightweight multi-frame refinement.

Reference graph

Works this paper leans on

45 extracted references · 39 canonical work pages · cited by 1 Pith paper

  1. [1]

    Patchmatch: A randomized correspon- dence algorithm for structural image editing

    Connelly Barnes, Eli Shechtman, Adam Finkelstein, and Dan B Goldman. Patchmatch: A randomized correspon- dence algorithm for structural image editing. ACM Trans. Graph., 28(3):24, 2009. 3

    work page 2009

  2. [2]

    A naturalistic open source movie for op- tical flow evaluation

    Daniel J Butler, Jonas Wulff, Garrett B Stanley, and Michael J Black. A naturalistic open source movie for op- tical flow evaluation. In ECCV, pages 611–625. Springer,

    work page

  3. [3]

    Basicvsr: The search for essential compo- nents in video super-resolution and beyond

    Kelvin CK Chan, Xintao Wang, Ke Yu, Chao Dong, and Chen Change Loy. Basicvsr: The search for essential compo- nents in video super-resolution and beyond. In CVPR, pages 4947–4956, 2021. 1

    work page 2021

  4. [4]

    Memflow: Optical flow esti- mation and prediction with memory

    Qiaole Dong and Yanwei Fu. Memflow: Optical flow esti- mation and prediction with memory. InCVPR, pages 19068– 19078, 2024. 1, 2, 4, 5, 6, 7

    work page 2024

  5. [5]

    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. InICCV, pages 2758–2766, 2015. 1, 2, 5

    work page 2015

  6. [6]

    Two-frame motion estimation based on polynomial expansion

    Gunnar Farneb ¨ack. Two-frame motion estimation based on polynomial expansion. In Image Analysis: 13th Scandina- vian Conference, SCIA 2003 Halmstad, Sweden, June 29– July 2, 2003 Proceedings 13, pages 363–370. Springer, 2003. 2

    2003

  7. [7]

    Vision meets robotics: The kitti dataset

    Andreas Geiger, Philip Lenz, Christoph Stiller, and Raquel Urtasun. Vision meets robotics: The kitti dataset. The Inter- national Journal of Robotics Research , 32(11):1231–1237,

  8. [8]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In CVPR, pages 770–778, 2016. 5

    2016

Show all 45 references
  1. [9]

    Determining op- tical flow

    Berthold KP Horn and Brian G Schunck. Determining op- tical flow. Artificial intelligence, 17(1-3):185–203, 1981. 1, 2

    1981

  2. [10]

    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. In ECCV, pages 668–685. Springer, 2022. 2, 6, 7

    2022

  3. [11]

    Real-time intermediate flow estima- tion for video frame interpolation

    Zhewei Huang, Tianyuan Zhang, Wen Heng, Boxin Shi, and Shuchang Zhou. Real-time intermediate flow estima- tion for video frame interpolation. In ECCV, pages 624–642. Springer, 2022. 1

    2022

  4. [12]

    Flownet 2.0: Evolu- tion of optical flow estimation with deep networks

    Eddy Ilg, Nikolaus Mayer, Tonmoy Saikia, Margret Keuper, Alexey Dosovitskiy, and Thomas Brox. Flownet 2.0: Evolu- tion of optical flow estimation with deep networks. InCVPR, pages 2462–2470, 2017. 6, 7

    2017

  5. [13]

    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. In ICCV, pages 9772–9781,

  6. [14]

    Learning optical flow from a few matches

    Shihao Jiang, Yao Lu, Hongdong Li, and Richard Hartley. Learning optical flow from a few matches. In CVPR, pages 16592–16600, 2021. 1, 3

    2021

  7. [15]

    Face deblurring using dual camera fusion on mobile phones

    Wei-Sheng Lai, Yichang Shih, Lun-Cheng Chu, Xiaotong Wu, Sung-Fang Tsai, Michael Krainin, Deqing Sun, and Chia-Kai Liang. Face deblurring using dual camera fusion on mobile phones. ACM Transactions on Graphics (TOG), 41(4):1–16, 2022. 6

    2022

  8. [16]

    Video frame inter- polation via optical flow estimation with image inpainting

    Xiaozhang Liu, Hui Liu, and Yuxiu Lin. Video frame inter- polation via optical flow estimation with image inpainting. International Journal of Intelligent Systems , 35(12):2087– 2102, 2020. 1

    2020

  9. [17]

    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. 1, 2

    1981

  10. [18]

    Spring: A high-resolution high- detail dataset and benchmark for scene flow, optical flow and stereo

    Lukas Mehl, Jenny Schmalfuss, Azin Jahedi, Yaroslava Nali- vayko, and Andr ´es Bruhn. Spring: A high-resolution high- detail dataset and benchmark for scene flow, optical flow and stereo. In CVPR, pages 4981–4991, 2023. 1, 2, 4, 5

    2023

  11. [19]

    Object scene flow for autonomous vehicles

    Moritz Menze and Andreas Geiger. Object scene flow for autonomous vehicles. In CVPR, pages 3061–3070, 2015. 2, 5

    2015

  12. [20]

    Recurrent partial kernel network for efficient optical flow estimation

    Henrique Morimitsu, Xiaobin Zhu, Xiangyang Ji, and Xu- Cheng Yin. Recurrent partial kernel network for efficient optical flow estimation. In AAAI, pages 4278–4286, 2024. 1, 6, 7

    2024

  13. [21]

    A benchmark dataset and evaluation methodology for video object segmentation

    Federico Perazzi, Jordi Pont-Tuset, Brian McWilliams, Luc Van Gool, Markus Gross, and Alexander Sorkine-Hornung. A benchmark dataset and evaluation methodology for video object segmentation. In CVPR, pages 724–732, 2016. 3

    2016

  14. [22]

    Representation flow for action recognition

    AJ Piergiovanni and Michael S Ryoo. Representation flow for action recognition. In CVPR, pages 9945–9953, 2019. 1

    2019

  15. [23]

    A fusion approach for multi- frame optical flow estimation

    Zhile Ren, Orazio Gallo, Deqing Sun, Ming-Hsuan Yang, Erik B Sudderth, and Jan Kautz. A fusion approach for multi- frame optical flow estimation. InIEEE Winter Conference on Applications of Computer Vision , pages 2077–2086. IEEE,

  16. [24]

    Playing for benchmarks

    Stephan R Richter, Zeeshan Hayder, and Vladlen Koltun. Playing for benchmarks. In ICCV, pages 2213–2222, 2017. 5, 1

    2017

  17. [25]

    The surprising effectiveness of diffusion models for optical flow and monocular depth estimation

    Saurabh Saxena, Charles Herrmann, Junhwa Hur, Abhishek Kar, Mohammad Norouzi, Deqing Sun, and David J Fleet. The surprising effectiveness of diffusion models for optical flow and monocular depth estimation. NeurIPS, 36:39443– 39469, 2023. 7

    2023

  18. [26]

    Videoflow: Exploiting temporal cues for multi-frame optical flow estimation

    Xiaoyu Shi, Zhaoyang Huang, Weikang Bian, Dasong Li, Manyuan Zhang, Ka Chun Cheung, Simon See, Hongwei Qin, Jifeng Dai, and Hongsheng Li. Videoflow: Exploiting temporal cues for multi-frame optical flow estimation. In ICCV, pages 12469–12480, 2023. 1, 2, 3, 6, 7

    2023

  19. [27]

    Flowformer++: Masked cost volume autoen- coding for pretraining optical flow estimation

    Xiaoyu Shi, Zhaoyang Huang, Dasong Li, Manyuan Zhang, Ka Chun Cheung, Simon See, Hongwei Qin, Jifeng Dai, and Hongsheng Li. Flowformer++: Masked cost volume autoen- coding for pretraining optical flow estimation. In CVPR, pages 1599–1610, 2023. 1

    2023

  20. [28]

    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. In CVPR, pages 8934–8943, 2018. 1, 2, 6, 7

    2018

  21. [29]

    Optical flow guided feature: A fast and robust motion representation for video action recognition

    Shuyang Sun, Zhanghui Kuang, Lu Sheng, Wanli Ouyang, and Wei Zhang. Optical flow guided feature: A fast and robust motion representation for video action recognition. In CVPR, pages 1390–1399, 2018. 1 9

    2018

  22. [30]

    Streamflow: streamlined multi- frame optical flow estimation for video sequences

    Shangkun Sun, Jiaming Liu, Huaxia Li, Guoqing Liu, Thomas Li, and Wei Gao. Streamflow: streamlined multi- frame optical flow estimation for video sequences. NeurIPS, 37:9205–9228, 2025. 1, 2, 3, 4, 5, 6, 7

    2025

  23. [31]

    Raft: Recurrent all-pairs field transforms for optical flow

    Zachary Teed and Jia Deng. Raft: Recurrent all-pairs field transforms for optical flow. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23– 28, 2020, Proceedings, Part II 16, pages 402–419. Springer,

    2020

  24. [32]

    Sea-raft: Simple, efficient, accurate raft for optical flow

    Yihan Wang, Lahav Lipson, and Jia Deng. Sea-raft: Simple, efficient, accurate raft for optical flow. In ECCV, pages 36–

  25. [33]

    Croco v2: Improved cross-view completion pre- training for stereo matching and optical flow

    Philippe Weinzaepfel, Thomas Lucas, Vincent Leroy, Yohann Cabon, Vaibhav Arora, Romain Br ´egier, Gabriela Csurka, Leonid Antsfeld, Boris Chidlovskii, and J ´erˆome Revaud. Croco v2: Improved cross-view completion pre- training for stereo matching and optical flow. InICCV, pag...

    2023

  26. [34]

    Memory-efficient optical flow via radius-distribution orthogonal cost volume

    Gangwei Xu, Shujun Chen, Hao Jia, Miaojie Feng, and Xin Yang. Memory-efficient optical flow via radius-distribution orthogonal cost volume. arXiv preprint arXiv:2312.03790,

    arXiv

  27. [35]

    High-resolution optical flow from 1d attention and correlation

    Haofei Xu, Jiaolong Yang, Jianfei Cai, Juyong Zhang, and Xin Tong. High-resolution optical flow from 1d attention and correlation. In ICCV, pages 10498–10507, 2021. 1, 3, 5, 6, 7

    2021

  28. [36]

    Gmflow: Learning optical flow via global matching

    Haofei Xu, Jing Zhang, Jianfei Cai, Hamid Rezatofighi, and Dacheng Tao. Gmflow: Learning optical flow via global matching. In CVPR, pages 8121–8130, 2022. 1, 2

    2022

  29. [37]

    Quadratic video interpolation

    Xiangyu Xu, Li Siyao, Wenxiu Sun, Qian Yin, and Ming- Hsuan Yang. Quadratic video interpolation. NeurIPS, 32,

  30. [38]

    Improved two-stream model for human action recognition

    Yuxuan Zhao, Ka Lok Man, Jeremy Smith, Kamran Sid- dique, and Sheng-Uei Guan. Improved two-stream model for human action recognition. EURASIP Journal on Image and Video Processing, 2020:1–9, 2020. 1

    2020

  31. [39]

    Hybrid cost volume for memory-efficient optical flow

    Yang Zhao, Gangwei Xu, and Gang Wu. Hybrid cost volume for memory-efficient optical flow. InProceedings of the 32nd ACM International Conference on Multimedia, pages 8740– 8749, 2024. 1

    2024

  32. [40]

    Dip: Deep inverse patchmatch for high-resolution optical flow

    Zihua Zheng, Ni Nie, Zhi Ling, Pengfei Xiong, Jiangyu Liu, Hao Wang, and Jiankun Li. Dip: Deep inverse patchmatch for high-resolution optical flow. InCVPR, pages 8925–8934,

  33. [43]

    Definitions Here we will provide more detailed definitions used in the main text. 7.1. W AUC In optical flow, weighted area under curve (W AUC), origi- nally from VIPER [24], is formally defined as the integral 2 5 Z 5 0 f (x) · 5 − x 5 dx, (15) where f (x) is equal to the per...

    1920

  34. [44]

    Additional ablations In this section, we provide ablations or ablation data not included in the main text. 8.1. Number of iterative refinements We study our method’s behavior depending on the number of iterative refinements. The results are provided in Table 7. For a balance b...

  35. [45]

    Additional results In this section, we provide some other results that are not included in the main text. 9.1. Additional zero-shot evaluation Following previous works, we evaluate the zero-shot per- formance of our method after the ”Things” training stage on Sintel (train) an...

  36. [54]

    2, 4, 5, 6, 7

    Springer, 2024. 2, 4, 5, 6, 7

    2024

  37. [2022]

    3 10 MEMFOF: High-Resolution Training for Memory-Efficient Multi-Frame Optical Flow Estimation Supplementary Material

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.