Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Joint Flow And Feature Refinement Using Attention For Video Restoration

T0 review · 3 major / 5 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 Jointly refining optical flow and feature maps in alternating steps lets a single video-restoration network set state-of-the-art results on denoising, deblurring, and super-resolution, with gains up to 1.62 dB PSNR.

desk verdict Solid engineering contribution with plausible gains, but the key ablation is too weak to pin the improvement on the joint refinement mechanism. read the letter →

arxiv 2505.16434 v1 pith:GLKWE3QU submitted 2025-05-22 cs.CV cs.MM

Ranjith Merugu , Mohammad Sameer Suhail , Akshay P Sarashetti , Venkata Bharath Reddy Reddem , Pankaj Kumar Bajpai , Amit Satish Unde This is my paper
classification cs.CVcs.MM
keywords videorestorationopticalflowrefinementattentionmechanismtemporalconsistencyocclusion-awarelossdenoisingdeblurringsuper-resolution
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

The paper proposes a video restoration architecture, JFFRA, built on the idea that optical flow and visual features should be refined alternately instead of using a fixed flow computed on degraded frames. It claims this joint refinement, repeated across scales, lets the network use temporal information more reliably and reduces sensitivity to initial flow errors. The network also uses an occlusion-aware temporal loss that penalizes misalignment mainly in non-occluded, nearly static regions, where flicker is most visible. The paper reports state-of-the-art results on video denoising, deblurring, and super-resolution benchmarks, with the largest gain being 1.62 dB PSNR over prior methods on the CRVD denoising dataset.

What carries the argument

The Joint Flow and Feature Refinement (JFFR) module is the paper's central mechanism: within each block, a cost volume over a 9x9 search window measures feature dissimilarity after warping, a flow head predicts only the offset from the current flow, and offset-flow warping is followed by query-key-value attention over the concatenation of reference and warped neighbor features. The same flow and features then pass to the next scale, creating a loop in which each refinement improves the input to the other. The supporting mechanism is the occlusion-aware temporal loss, which multiplies the L1 difference between the restored reference frame and the warped restored neighbor by an exponential mask built from ground-truth frame differences, keeping the loss focused on regions that should match.

What would settle it

Retrain the ablation variants from Table 5 with the same iteration count, batch size, and learning-rate schedule as JFFRA, and check validation loss convergence; if the PSNR gaps shrink to under roughly 0.3 dB once convergence is matched, the joint-refinement claim is not supported.

Watch

Extended reading notes

Core claim

JFFRA's central claim is that restoration quality improves when flow estimation and feature enhancement are coupled as an iterative, alternating process at multiple scales. Starting from a pre-trained optical flow model and 3D-convolution features of the low-quality input, each Joint Flow and Feature Refinement block first corrects the flow through a cost-volume comparison and offset learning, then uses the corrected flow to warp neighboring frames and refine all features through attention. The refined features feed the next refinement stage, so alignment and restoration improve together. An occlusion-aware temporal loss, computed from ground-truth optical flow and an exponential occlusion mask, suppresses flicker by enforcing consistency in visible, non-occluded regions. The paper argues that this mutual reinforcement makes the method outperform end-to-end attention models across denoising, deblurring, and super-resolution.

Load-bearing premise

The load-bearing premise is that the ablated variants in the comparison were trained under the same budget and to a similar final convergence as the full model; if those versions were undertrained, the large PSNR gaps do not isolate the contribution of joint flow and feature refinement.

Editorial extensions

If this is right

  • Video restoration systems do not need to wait for a perfect initial flow estimate; correcting the flow on the fly from progressively cleaned features is sufficient.
  • Using only three input frames, the method can match or beat much larger recurrent models on super-resolution benchmarks such as REDS and Vimeo-90K.
  • The occlusion-aware temporal loss should reduce visible flicker in near-static regions, supported by a lower optical-flow warping error compared to training without the loss.
  • The same architecture applies to denoising, deblurring, and super-resolution, implying the joint refinement mechanism is not tied to one degradation type.

Reading between the lines

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

  • A test the paper does not run: train JFFRA with a less accurate off-the-shelf flow model; if the gains shrink noticeably, the claimed robustness to initial flow quality is weaker than stated.
  • The temporal loss could double as an inference-time flicker detector by measuring masked L1 between restored frames, giving restorers a cheap consistency metric.
  • If the joint refinement loop is the actual driver of the gains, similar alternating refinement could be transplanted to other alignment-dependent video tasks such as frame interpolation or video inpainting.
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 / 5 minor

Summary. The paper proposes JFFRA, a video restoration framework that jointly and iteratively refines optical flow and frame features using attention, together with an occlusion-aware temporal consistency loss. The method is evaluated on video denoising (CRVD, DAVIS, Set-8), video deblurring (DVD), and video super-resolution (REDS, Vimeo-90K, Vid4, UDM10), reporting state-of-the-art or competitive PSNR/SSIM numbers, with a headline gain of 1.62 dB over prior SOTA on CRVD.

Significance. If the reported gains hold under controlled comparison, the joint flow-feature refinement idea is a useful and well-motivated contribution to video restoration, and the occlusion-aware temporal loss is a sensible mechanism for reducing flicker. The paper also covers a wide range of tasks and benchmarks, which strengthens the claim of generality. At the same time, the central mechanistic claim that joint iterative refinement drives the improvement rests on an ablation study that is not shown to be training-matched, and the architecture is under-specified in several places. The paper's quantitative results are promising, but the evidence as presented is not yet sufficient to fully support the attribution claim.

major comments (3)
  1. [Section 5, Table 5] The central mechanistic claim that joint iterative flow and feature refinement drives the improvement rests on the ablation gap between the full JFFR model (35.15 dB on DVD) and its three ablated variants (28.5–29.45 dB). The paper does not report whether these variants were trained with the same 700k-iteration, batch-48 budget as the full model, nor does it provide convergence curves, FLOPs, or parameter counts. A 5.7–6.65 dB drop from removing a component is atypically large for a module ablation in video restoration and could indicate undertraining or a different training schedule rather than the module's true contribution. Please provide matched-budget training curves and computational cost for each ablated variant, or substantially temper the attribution of the improvement to the JFFR mechanism.
  2. [Section 3.3] The JFFR module is under-specified. Equation (5) invokes a 'Flow Head' but its architecture, number of layers, and input/output channels are not described; the number of JFFR refinement stages and their correspondence to the encoder-decoder levels are not stated; and the cost volume in Eq. (3) is not tied to the actual feature resolution used in the experiments. Without these details the method cannot be reproduced or compared fairly with existing approaches. Please add a complete architecture description, including a table that lists each component, its channels, and the number of stages.
  3. [Section 3.4, Section 4.1] The temporal loss in Eq. (13) introduces a hyperparameter α that is described in the text as needing to be 'sufficiently large,' but Section 4.1 sets α=0.2, which is a small value for an exponential decay mask. The paper does not report any sensitivity analysis for this parameter, even though the ablation in Table 6 attributes a measurable effect to the temporal loss. Please provide a sensitivity study or a clear justification for the chosen value, and clarify whether the same α is used across all tasks.
minor comments (5)
  1. [Section 3.3] The text says 'a sequence of 3 consecutive frames (i.e., N=3)', but the framework uses 2N+1 frames; for 3 frames, N should be 1. Please correct this notation.
  2. [Section 2.2] The phrase 'progressively improvised over the original optical flow and degraded frames' should be 'progressively improved' rather than 'improvised.'
  3. [Section 4.2, Table 2] On Set-8, JFFRA is slightly below VRT at σ=30 and below ShiftNet at σ=40 and σ=50. The text says the method 'gives competitive performance,' which is fair, but the selective emphasis on the best results in the abstract should be balanced by acknowledging these cases.
  4. [Section 4.5, Figure 6] The complexity analysis refers to 'supplementary' for detailed FLOPs and parameter numbers, but the arXiv version does not include a supplementary file. Please report the concrete complexity values in the main text.
  5. [Section 1, Contribution 3] The claim that this is 'the first to demonstrate' that iterative joint flow-feature refinement outperforms end-to-end learning is very strong and is not supported by a comparison to prior work that has used iterative refinement in related settings. Please soften the claim or provide a more thorough discussion of prior art.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: JFFRA is trained end-to-end on benchmark data and evaluated on held-out test sets; no fitted constants double as predictions.

full rationale

I examined the derivation chain of the paper. The initial optical flow is computed with a pre-trained RAFT model on the low-quality input frames (Section 3.2), which is an external, independently trained component, not a parameter fitted by JFFRA. The JFFR module then alternates between flow correction via a cost volume (Equation 5) and feature refinement via attention (Equations 6-11); this is a functional iterative architecture, not a logical circle, because the refined flow and features are network outputs used in the forward pass rather than fitted constants that are later renamed as predictions. The total loss (Equation 15) combines an L1 reconstruction term with an occlusion-aware temporal consistency loss defined using ground-truth optical flow and an exponential mask computed from ground-truth frames (Equations 12-14). This is a training-time supervision signal; it does not inject the test benchmark values into the model, and the test-time restoration does not use ground-truth flow or masks. The reported PSNR/SSIM improvements are measured on held-out test sets (CRVD, DAVIS, Set-8, REDS, Vimeo-90K, Vid4, UDM10, DVD) against externally published state-of-the-art methods. No load-bearing step is justified by a self-citation, no uniqueness theorem is imported from the authors' prior work, and no ansatz is smuggled in via citation. The ablation study in Table 5 does not report per-variant training budgets or convergence criteria, which is an experimental-validity concern about whether the ablations are comparable, but it is not a circularity: the full model and its variants are all trained on the same data and evaluated on the same held-out benchmarks, and the central result does not reduce by construction to the ablation setup. Therefore, the paper is self-contained with respect to circularity, and no circular step can be exhibited from the text with the required specificity.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central claim is empirical and rests on standard deep-learning assumptions: pretrained flow on degraded frames is a usable initialization, ground-truth flow at training time transfers to test time, and benchmark protocols are followed. No new physical entities are introduced. The main hand-set constants are loss weights and the occlusion-mask steepness.

free parameters (4)
  • alpha = 0.2
    Controls the steepness of the occlusion mask in Eq (13); set by hand and not swept.
  • w1, w2 = 0.2 each
    Weights balancing L1 and temporal losses in Eq (15); chosen without reported sensitivity analysis.
  • attention window size = 8
    Window size for attention computation (Sec 4.1); no ablation justifies this choice.
  • number of JFFR refinement stages
    Multi-scale iterative architecture described but number of iterations per scale not specified in the paper.
assumptions (3)
  • domain assumption Pre-trained RAFT flow on degraded frames provides a useful initialization for refinement.
    Invoked in Sec 3.2 when computing initial flow on LQ frames; if this fails, the refinement loop starts from poor alignment.
  • domain assumption Temporal loss computed with ground-truth flow and occlusion masks during training improves test-time temporal consistency.
    Sec 3.4 uses GT flow for the mask and warping; assumes this supervision signal transfers to restored outputs at inference where GT is unavailable.
  • domain assumption Standard benchmark datasets, splits, and degradation settings are used without modification.
    Sec 4 follows the VRT training pipeline; results depend on those protocols.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Joint Flow And Feature Refinement Using Attention For Video Restoration." pith.science (2026). https://pith.science/paper/GLKWE3QU

@misc{pith2026250516434,
  author       = {Pith},
  title        = {Pith review of: Joint Flow And Feature Refinement Using Attention For Video Restoration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GLKWE3QU}},
  note         = {Machine review of arXiv:2505.16434}
}
read the original abstract

Recent advancements in video restoration have focused on recovering high-quality video frames from low-quality inputs. Compared with static images, the performance of video restoration significantly depends on efficient exploitation of temporal correlations among successive video frames. The numerous techniques make use of temporal information via flow-based strategies or recurrent architectures. However, these methods often encounter difficulties in preserving temporal consistency as they utilize degraded input video frames. To resolve this issue, we propose a novel video restoration framework named Joint Flow and Feature Refinement using Attention (JFFRA). The proposed JFFRA is based on key philosophy of iteratively enhancing data through the synergistic collaboration of flow (alignment) and restoration. By leveraging previously enhanced features to refine flow and vice versa, JFFRA enables efficient feature enhancement using temporal information. This interplay between flow and restoration is executed at multiple scales, reducing the dependence on precise flow estimation. Moreover, we incorporate an occlusion-aware temporal loss function to enhance the network's capability in eliminating flickering artifacts. Comprehensive experiments validate the versatility of JFFRA across various restoration tasks such as denoising, deblurring, and super-resolution. Our method demonstrates a remarkable performance improvement of up to 1.62 dB compared to state-of-the-art approaches.

Figures

Figures reproduced from arXiv: 2505.16434 by the authors.

Figure 1
Figure 1. The proposed JFFRA framework achieves the state [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Architecture of the proposed joint flow and feature refinement using attention framework(JFFRA). [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Visual comparison on video denoising task. The JFFRA effectively remove noise while preserving structure and sharpness. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Visual comparison on video super-resolution ( [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Visual comparison on video deblurring task. The restored image using JFFRA is more sharper and closer to ground truth. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Complexity Analysis of JFFRA with other state-of-the [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Impact of flow refinement in retrieving the information [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

Discussion (0). Sign in 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. StatsMerging: Statistics-Guided Model Merging via Task-Specific Teacher Distillation

    cs.LG 2025-06 conditional novelty 5.0 of 10

    StatsMerging predicts per-layer merging coefficients from weight statistics and teacher pseudo-labels, achieving 94.5% average accuracy across eight vision tasks, 5.1 points above WEMoE.

Reference graph

Works this paper leans on

62 extracted references · 53 canonical work pages · cited by 1 Pith paper

  1. [1]

    Restoration of video frames from a single blurred image with motion understanding

    Dawit Mureja Argaw, Junsik Kim, Francois Rameau, Chaon- ing Zhang, and In So Kweon. Restoration of video frames from a single blurred image with motion understanding. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 701–710, 2021. 1

    work page 2021

  2. [2]

    Video denoising via empirical bayesian estimation of space-time patches.Journal of Mathematical Imaging and Vision, 60:70–93, 2018

    Pablo Arias and Jean-Michel Morel. Video denoising via empirical bayesian estimation of space-time patches.Journal of Mathematical Imaging and Vision, 60:70–93, 2018. 6, 7

    work page 2018

  3. [3]

    Patch-based video denoising with optical flow estimation

    Antoni Buades, Jose-Luis Lisani, and Marko Miladinovi ´c. Patch-based video denoising with optical flow estimation. IEEE Transactions on Image Processing, 25(6):2573–2586,

    work page

  4. [4]

    Real- time video super-resolution with spatio-temporal networks and motion compensation

    Jose Caballero, Christian Ledig, Andrew Aitken, Alejandro Acosta, Johannes Totz, Zehan Wang, and Wenzhe Shi. Real- time video super-resolution with spatio-temporal networks and motion compensation. In Proceedings of the IEEE con- ference on computer vision and pattern recognition , pages 4778–4787, 2017. 2

    work page 2017

  5. [5]

    Video super-resolution transformer

    Jiezhang Cao, Yawei Li, Kai Zhang, and Luc Van Gool. Video super-resolution transformer. arXiv, 2021. 2

    work page 2021

  6. [6]

    Reference-based image super-resolution with deformable attention trans- former

    Jiezhang Cao, Jingyun Liang, Kai Zhang, Yawei Li, Yulun Zhang, Wenguan Wang, and Luc Van Gool. Reference-based image super-resolution with deformable attention trans- former. In European conference on computer vision , 2022. 2

    work page 2022

  7. [7]

    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 Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4947–4956, 2021. 6, 7

    work page 2021

  8. [8]

    Basicvsr++: Improving video super- resolution with enhanced propagation and alignment

    Kelvin CK Chan, Shangchen Zhou, Xiangyu Xu, and Chen Change Loy. Basicvsr++: Improving video super- resolution with enhanced propagation and alignment. In IEEE Conference on Computer Vision and Pattern Recog- nition, 2022. 3, 7

    work page 2022

Show all 62 references
  1. [9]

    Seeing motion in the dark

    Chen Chen, Qifeng Chen, Minh N Do, and Vladlen Koltun. Seeing motion in the dark. InIEEE International Conference on Computer Vision, pages 3185–3194, 2019. 6

    2019

  2. [10]

    Videnn: Deep blind video denoising

    Michele Claus and Jan van Gemert. Videnn: Deep blind video denoising. In IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2019. 3, 6

    2019

  3. [11]

    Non-local video denoising by cnn

    Axel Davy, Thibaud Ehret, Jean-Michel Morel, Pablo Arias, and Gabriele Facciolo. Non-local video denoising by cnn. arXiv preprint arXiv:1811.12758, 2018. 6

    2018 arXiv

  4. [12]

    Flownet: Learn- ing optical flow with convolutional networks

    Philipp Fischer, Alexey Dosovitskiy, Eddy Ilg, Philip H¨ausser, Caner Hazırbas ¸, Vladimir Golkov, Patrick Van der Smagt, Daniel Cremers, and Thomas Brox. Flownet: Learn- ing optical flow with convolutional networks. arXiv preprint arXiv:1504.06852, 2015. 2, 3

    2015 arXiv

  5. [13]

    Recurrent back-projection network for video super- resolution

    Muhammad Haris, Gregory Shakhnarovich, and Norimichi Ukita. Recurrent back-projection network for video super- resolution. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3897–3906,

  6. [14]

    Bidirectional recurrent convolutional networks for multi-frame super- resolution

    Yan Huang, Wei Wang, and Liang Wang. Bidirectional recurrent convolutional networks for multi-frame super- resolution. Advances in neural information processing sys- tems, 28, 2015. 2

    2015

  7. [15]

    Deep video super-resolution network using dynamic upsampling filters without explicit motion compen- sation

    Younghyun Jo, Seoung Wug Oh, Jaeyeon Kang, and Seon Joo Kim. Deep video super-resolution network using dynamic upsampling filters without explicit motion compen- sation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 3224–3232, 2018. 7

    2018

  8. [16]

    Video object segmentation with language referring expressions

    Anna Khoreva, Anna Rohrbach, and Bernt Schiele. Video object segmentation with language referring expressions. In Computer Vision–ACCV 2018: 14th Asian Conference on Computer Vision, Perth, Australia, December 2–6, 2018, Re- vised Selected Papers, Part IV 14, pages 123–141. Springer,

    2018

  9. [17]

    Towards real-world event-guided low- light video enhancement and deblurring

    Taewoo Kim, Jaeseok Jeong, Hoonhee Cho, Yuhwan Jeong, and Kuk-Jin Yoon. Towards real-world event-guided low- light video enhancement and deblurring. arXiv preprint arXiv:2408.14916, 2024. 1, 2

    2024 arXiv

  10. [18]

    Spatio-temporal transformer network for video restoration

    Tae Hyun Kim, Mehdi SM Sajjadi, Michael Hirsch, and Bernhard Scholkopf. Spatio-temporal transformer network for video restoration. In Proceedings of the European con- ference on computer vision (ECCV) , pages 106–122, 2018. 7

    2018

  11. [19]

    Arvo: Learning all-range volumetric correspondence for video deblurring

    Dongxu Li, Chenchen Xu, Kaihao Zhang, Xin Yu, Yiran Zhong, Wenqi Ren, Hanna Suominen, and Hongdong Li. Arvo: Learning all-range volumetric correspondence for video deblurring. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 7721–77...

    2021

  12. [20]

    A simple baseline for video restoration with grouped spatial- temporal shift

    Dasong Li, Xiaoyu Shi, Yi Zhang, Ka Chun Cheung, Simon See, Xiaogang Wang, Hongwei Qin, and Hongsheng Li. A simple baseline for video restoration with grouped spatial- temporal shift. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages ...

    2023

  13. [21]

    Swinir: Image restoration us- ing swin transformer

    Jingyun Liang, Jiezhang Cao, Guolei Sun, Kai Zhang, Luc Van Gool, and Radu Timofte. Swinir: Image restoration us- ing swin transformer. InProceedings of the IEEE/CVF inter- national conference on computer vision , pages 1833–1844,

  14. [22]

    Recurrent video restoration trans- former with guided deformable attention

    Jingyun Liang, Yuchen Fan, Xiaoyu Xiang, Rakesh Ranjan, Eddy Ilg, Simon Green, Jiezhang Cao, Kai Zhang, Radu Timofte, and Luc V Gool. Recurrent video restoration trans- former with guided deformable attention. Advances in Neu- ral Information Processing Systems, 35:378–393, 20...

    2022

  15. [23]

    Vrt: A video restoration transformer

    Jingyun Liang, Jiezhang Cao, Yuchen Fan, Kai Zhang, Rakesh Ranjan, Yawei Li, Radu Timofte, and Luc Van Gool. Vrt: A video restoration transformer. IEEE Transactions on Image Processing, 2024. 1, 2, 3, 6, 7, 8

    2024

  16. [24]

    Video denoising, deblocking, and en- hancement through separable 4-d nonlocal spatiotemporal transforms

    Matteo Maggioni, Giacomo Boracchi, Alessandro Foi, and Karen Egiazarian. Video denoising, deblocking, and en- hancement through separable 4-d nonlocal spatiotemporal transforms. IEEE Transactions on Image Processing, 2012. 6

    2012

  17. [25]

    Efficient multi-stage video denoising with recurrent spatio-temporal fusion

    Matteo Maggioni, Yibin Huang, Cheng Li, Shuai Xiao, Zhongqian Fu, and Fenglong Song. Efficient multi-stage video denoising with recurrent spatio-temporal fusion. In IEEE Conference on Computer Vision and Pattern Recogni- tion, 2021. 2, 3

    2021

  18. [26]

    Deep multi-scale convolutional neural network for dynamic scene deblurring, 2018

    Seungjun Nah, Tae Hyun Kim, and Kyoung Mu Lee. Deep multi-scale convolutional neural network for dynamic scene deblurring, 2018. 6, 7

    2018

  19. [27]

    Ntire 2019 challenge on video deblurring and super- resolution: Dataset and study

    Seungjun Nah, Sungyong Baik, Seokil Hong, Gyeongsik Moon, Sanghyun Son, Radu Timofte, and Kyoung Mu Lee. Ntire 2019 challenge on video deblurring and super- resolution: Dataset and study. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition work...

    2019

  20. [28]

    Cascaded deep video deblurring using temporal sharpness prior

    Jinshan Pan, Haoran Bai, and Jinhui Tang. Cascaded deep video deblurring using temporal sharpness prior. InProceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3043–3051, 2020. 7

    2020

  21. [29]

    Meshflow video denoising

    Zhihang Ren, Jiajia Li, Shuaicheng Liu, and Bing Zeng. Meshflow video denoising. In 2017 IEEE International Conference on Image Processing (ICIP), pages 2966–2970,

    2017

  22. [30]

    Unsupervised deep video denoising

    Dev Yashpal Sheth, Sreyas Mohan, Joshua L Vincent, Ra- mon Manzorro, Peter A Crozier, Mitesh M Khapra, Eero P Simoncelli, and Carlos Fernandez-Granda. Unsupervised deep video denoising. In Proceedings of the IEEE/CVF international conference on computer vision , pages 1759– 17...

    2021

  23. [31]

    Recurrent video deblurring with blur-invariant motion estimation and pixel volumes

    Hyeongseok Son, Junyong Lee, Jonghyeop Lee, Sunghyun Cho, and Seungyong Lee. Recurrent video deblurring with blur-invariant motion estimation and pixel volumes. ACM Transactions on Graphics (TOG), 40(5):1–18, 2021. 7

    2021

  24. [32]

    Temp- former: Temporally consistent transformer for video denois- ing

    Mingyang Song, Yang Zhang, and Tunc ¸ O Aydın. Temp- former: Temporally consistent transformer for video denois- ing. In European conference on computer vision, pages 481–

  25. [33]

    Understand- ing lstm–a tutorial into long short-term memory recurrent neural networks

    Ralf C Staudemeyer and Eric Rothstein Morris. Understand- ing lstm–a tutorial into long short-term memory recurrent neural networks. arXiv preprint arXiv:1909.09586, 2019. 2

    1909 arXiv

  26. [34]

    Deep video deblurring for hand-held cameras

    Shuochen Su, Mauricio Delbracio, Jue Wang, Guillermo Sapiro, Wolfgang Heidrich, and Oliver Wang. Deep video deblurring for hand-held cameras. In Proceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 1279–1288, 2017. 2, 7, 8

    2017

  27. [35]

    Gated spatio-temporal attention-guided video deblurring

    Maitreya Suin and AN Rajagopalan. Gated spatio-temporal attention-guided video deblurring. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7802–7811, 2021. 7

    2021

  28. [36]

    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 Proceedings of the IEEE conference on computer vision and pattern recognition, pages 8934–8943,

  29. [37]

    Dvdnet: A fast network for deep video denoising

    Matias Tassano, Julie Delon, and Thomas Veit. Dvdnet: A fast network for deep video denoising. InIEEE International Conference on Image Processing, 2019. 2, 3, 6

    2019

  30. [38]

    Fastdvdnet: Towards real-time deep video denoising without flow estima- tion

    Matias Tassano, Julie Delon, and Thomas Veit. Fastdvdnet: Towards real-time deep video denoising without flow estima- tion. In IEEE Conference on Computer Vision and Pattern Recognition, 2020. 3, 6

    2020

  31. [39]

    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

  32. [40]

    Tdan: Temporally-deformable alignment network for video super-resolution

    Yapeng Tian, Yulun Zhang, Yun Fu, and Chenliang Xu. Tdan: Temporally-deformable alignment network for video super-resolution. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 3360–3369, 2020. 2

    2020

  33. [41]

    Self-supervised video object segmenta- tion with distillation learning of deformable attention

    Quang-Trung Truong, Duc Thanh Nguyen, Binh-Son Hua, and Sai-Kit Yeung. Self-supervised video object segmenta- tion with distillation learning of deformable attention. arXiv preprint arXiv:2401.13937, 2024. 2

    2024 arXiv

  34. [42]

    Patch craft: Video denoising by deep modeling and patch matching

    Gregory Vaksman, Michael Elad, and Peyman Milanfar. Patch craft: Video denoising by deep modeling and patch matching. In IEEE International Conference on Computer Vision, 2021. 2, 6

    2021

  35. [43]

    Patch craft: Video denoising by deep modeling and patch matching

    Gregory Vaksman, Michael Elad, and Peyman Milanfar. Patch craft: Video denoising by deep modeling and patch matching. In IEEE International Conference on Computer Vision, 2021. 2

    2021

  36. [44]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. Advances in neural information processing systems, 30, 2017. 2

    2017

  37. [45]

    Edvr: Video restoration with enhanced deformable convolutional networks

    Xintao Wang, Kelvin CK Chan, Ke Yu, Chao Dong, and Chen Change Loy. Edvr: Video restoration with enhanced deformable convolutional networks. In IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2019. 2, 6, 7

    2019

  38. [46]

    Occlusion aware unsupervised learning of optical flow

    Yang Wang, Yi Yang, Zhenheng Yang, Liang Zhao, Peng Wang, and Wei Xu. Occlusion aware unsupervised learning of optical flow. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4884–4893,

  39. [47]

    Neural video depth stabilizer

    Yiran Wang, Min Shi, Jiaqi Li, Zihao Huang, Zhiguo Cao, Jianming Zhang, Ke Xian, and Guosheng Lin. Neural video depth stabilizer. In Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision , pages 9466–9476,

  40. [48]

    Vision transformer with deformable attention

    Zhuofan Xia, Xuran Pan, Shiji Song, Li Erran Li, and Gao Huang. Vision transformer with deformable attention. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 4794–4803, 2022. 2

    2022

  41. [49]

    Monocular relative depth percep- tion with web stereo data supervision

    Ke Xian, Chunhua Shen, Zhiguo Cao, Hao Lu, Yang Xiao, Ruibo Li, and Zhenbo Luo. Monocular relative depth percep- tion with web stereo data supervision. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion, pages 311–320, 2018. 8

    2018

  42. [50]

    Deep video deblurring using sharpness features from exemplars

    Xinguang Xiang, Hao Wei, and Jinshan Pan. Deep video deblurring using sharpness features from exemplars. IEEE Transactions on Image Processing, 29:8976–8987, 2020. 7

    2020

  43. [51]

    Video enhancement with task-oriented flow

    Tianfan Xue, Baian Chen, Jiajun Wu, Donglai Wei, and William T Freeman. Video enhancement with task-oriented flow. International Journal of Computer Vision, 127:1106– 1125, 2019. 2, 3, 6, 7

    2019

  44. [52]

    Video enhancement with task-oriented flow

    Tianfan Xue, Baian Chen, Jiajun Wu, Donglai Wei, and William T Freeman. Video enhancement with task-oriented flow. International Journal of Computer Vision, 127:1106– 1125, 2019. 6, 7

    2019

  45. [53]

    Progressive fusion video super-resolution network via exploiting non-local spatio-temporal correlations

    Peng Yi, Zhongyuan Wang, Kui Jiang, Junjun Jiang, and Ji- ayi Ma. Progressive fusion video super-resolution network via exploiting non-local spatio-temporal correlations. InPro- ceedings of the IEEE/CVF international conference on com- puter vision, pages 3106–3115, 2019. 6, 7

    2019

  46. [54]

    Deep it- erative down-up cnn for image denoising

    Songhyun Yu, Bumjun Park, and Jechang Jeong. Deep it- erative down-up cnn for image denoising. In IEEE Con- ference on Computer Vision and Pattern Recognition work- shops, pages 0–0, 2019. 6

    2019

  47. [55]

    Joint learning of blind video denoising and optical flow estimation

    Songhyun Yu, Bumjun Park, Junwoo Park, and Jechang Jeong. Joint learning of blind video denoising and optical flow estimation. In 2020 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition Workshops (CVPRW), pages 2099–2108, 2020. 2

    2020

  48. [56]

    A review of recurrent neural networks: Lstm cells and net- work architectures

    Yong Yu, Xiaosheng Si, Changhua Hu, and Jianxun Zhang. A review of recurrent neural networks: Lstm cells and net- work architectures. Neural computation, 31(7):1235–1270,

  49. [57]

    Supervised raw video denoising with a benchmark dataset on dynamic scenes

    Huanjing Yue, Cong Cao, Lei Liao, Ronghe Chu, and Jingyu Yang. Supervised raw video denoising with a benchmark dataset on dynamic scenes. In IEEE Conference on Com- puter Vision and Pattern Recognition, 2020. 5, 6

    2020

  50. [58]

    Blur-aware spatio-temporal sparse transformer for video deblurring

    Huicong Zhang, Haozhe Xie, and Hongxun Yao. Blur-aware spatio-temporal sparse transformer for video deblurring. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 2673–2681, 2024. 7

    2024

  51. [59]

    Adversarial spatio-temporal learning for video deblurring

    Kaihao Zhang, Wenhan Luo, Yiran Zhong, Lin Ma, Wei Liu, and Hongdong Li. Adversarial spatio-temporal learning for video deblurring. IEEE Transactions on Image Processing, 28(1):291–301, 2018. 7

    2018

  52. [60]

    Prac- tical blind denoising via swin-conv-unet and data synthesis

    Kai Zhang, Yawei Li, Jingyun Liang, Jiezhang Cao, Yulun Zhang, Hao Tang, Radu Timofte, and Luc Van Gool. Prac- tical blind denoising via swin-conv-unet and data synthesis. arXiv preprint arXiv:2203.13278, 2022. 3

    2022 arXiv

  53. [61]

    Spatio-temporal filter adaptive network for video deblurring

    Shangchen Zhou, Jiawei Zhang, Jinshan Pan, Haozhe Xie, Wangmeng Zuo, and Jimmy Ren. Spatio-temporal filter adaptive network for video deblurring. InIEEE International Conference on Computer Vision, pages 2482–2491, 2019. 7

    2019

  54. [62]

    De- formable convnets v2: More deformable, better results

    Xizhou Zhu, Han Hu, Stephen Lin, and Jifeng Dai. De- formable convnets v2: More deformable, better results. In IEEE Conference on Computer Vision and Pattern Recogni- tion, 2019. 2

    2019

Pith tools

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