Pith. sign in

REVIEW 3 major objections 3 minor 57 references

PriOr-Flow: Enhancing Primitive Panoramic Optical Flow with Orthogonal View

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

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

Pith's one-line read PriOr-Flow claims optical flow in equirectangular panoramas improves by processing an orthogonally rotated view of the same sphere, cutting reported endpoint error by about 30 percent on two benchmarks.

desk verdict A genuinely new orthogonal-view fusion idea for panoramic flow, with big reported gains, but the flow-field transformation is under-specified and must be clarified before the numbers can be trusted. read the letter →

arxiv 2506.23897 v3 pith:6D5Q7FCJ submitted 2025-06-30 cs.CV

Longliang Liu , Miaojie Feng , Junda Cheng , Jijun Xiang , Xuan Zhu , Xin Yang This is my paper · ORCID
classification cs.CV
keywords panoramicopticalflowequirectangularprojectionorthogonalviewdual-branchnetworkcostvolumelookuppolarregiondistortioniterativerefinement
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 sets out to show that the severe polar-region errors of equirectangular (ERP) panoramic optical flow can be removed by processing a second, orthogonally rotated copy of the same sphere alongside the original. Because a 90-degree rotation puts the second view's minimum distortion exactly where the original has its maximum, the network can harvest low-distortion motion evidence for the poles instead of guessing through the distortion. This matters for any application built on wide-field motion — virtual reality, autonomous driving, video interpolation and inpainting — because ERP remains the standard panoramic format. The best variant, PriOr-RAFT, is reported to reduce endpoint error by 30.0% on MPFDataset and 29.6% on FlowScape relative to the previous state of the art, with the polar regions contributing most of the gain.

What carries the argument

The load-bearing object is the orthogonal view: the same spherical scene re-projected into ERP after a 90-degree rotation about the x-axis, whose distortion pattern is complementary to the primitive view's, with minimal distortion exactly at the poles. Two mechanisms built on it carry the argument. DCCL is a lookup operator that indexes both cost volumes at spherical locations derived from a single flow estimate, so correlation cues from the low-distortion orthogonal volume suppress the distortion noise in the primitive volume's polar regions. ODDC is a fusion module that estimates both branches' confidence as the group-wise correlation between frame-1 features and warped frame-2 features, re-projects the orthogonal flow into primitive coordinates, and uses the confidences to gate how the orthogonal motion features compensate the primitive branch inside a ConvGRU. A spherical-area-weighted L1 loss supervises both branches so the non-uniform ERP sampling does not bias training toward the equator.

What would settle it

Take the best model and replace ODDC's learned confidence with either (a) the fixed distortion map of the primitive ERP view or (b) confidence computed in each branch's own coordinates; if polar-region EPE on FlowScape does not get worse under (a) or does not improve under (b), the mechanism attributed to ODDC is not carrying the gain. A second decisive experiment: train the full model with the orthogonal branch's loss term removed. If the 29.6% EPE reduction on FlowScape persists without supervision on the orthogonal branch, the low-distortion prior itself is not the cause, and the improvement instead comes from the shared lookup or the extra parameters.

Watch

Extended reading notes

Core claim

The central claim is that the orthogonal view's distortion pattern is complementary to the primitive ERP view's — a 90-degree rotation about the x-axis sends the primitive view's most distorted content (the poles) to the equator, so the rotated view has minimal distortion exactly where the original is worst — and that this complementarity can be converted into a working training signal. Two mechanisms do the conversion. The Dual-Cost Collaborative Lookup (DCCL) operator takes the current flow estimate, maps it onto the sphere, and retrieves correlation scores from both the primitive and the orthogonal cost volumes at the same spherical locations, so noisy polar-region cues in one volume are countered by clean cues in the other. The Ortho-Driven Distortion Compensation (ODDC) module estimates a per-pixel confidence for each branch as the group-wise correlation between frame-1 features and frame-2 features warped by that branch's flow, converts the orthogonal flow back into primitive coordinates, and feeds confidences together with both correlation cues into a confidence-guided ConvGRU, so the primitive flow is corrected adaptively exactly where its own evidence is weak. Trained end-to-end with a spherical-area-weighted loss on both branches, the framework reaches state-of-the-art endpoint error on both public benchmarks, including a 39.7% polar-region improvement over PanoFlow on FlowScape.

Load-bearing premise

ODDC decides how much to trust each branch by measuring group-wise correlation in primitive-view pixel coordinates only, so the whole polar-region gain rests on the assumption that a flow that is accurate on the sphere still looks accurate after being re-expressed in the primitive ERP grid — if that re-projection breaks the correspondence, the fusion weights would be systematically wrong.

Editorial extensions

If this is right

  • Because DCCL and ODDC are grafted onto the iterative refinement loop, the framework transfers: RAFT, GMA, and SKFlow backbones all improve on the EFT scene of MPFDataset, with EPE gains from 15.8% to 23.7% at full iterations.
  • The accuracy gain concentrates in the poles: on FlowScape the polar-region EPE is 39.7% better than PanoFlow, while the equatorial EPE is comparable (slightly worse), so the orthogonal prior is what buys the headline reduction.
  • The orthogonal prior accelerates convergence: PriOr-RAFT with only 4 iterations already beats the 12-iteration RAFT baseline by 13.4% EPE, and 3 iterations beat RAFT at the same count.
  • The 90-degree x-axis rotation is the load-bearing viewpoint choice: ablations show a y-axis rotation splits the polar region into a discontinuity, and 45-degree x-rotation leaves parts of the poles uncompensated, both performing worse.

Reading between the lines

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

  • Extension: the complementary-distortion trick is not specific to flow; stereo disparity, monocular depth, and video super-resolution on ERP input all suffer pole errors and could plausibly use the same rotate-and-fuse design.
  • Part of the gain may be a two-view data effect: the dual-branch design supervises two views of the same motion with shared lookups, effectively doubling the training signal. An ablation that trains the primitive branch alone on the same frames, with the orthogonal branch present at inference but unsupervised, would separate the low-distortion-prior contribution from the extra-supervision contribut
  • ODDC's confidences are computed in primitive pixel coordinates only; computing confidence in each branch's own coordinates and fusing on the sphere might close the small equatorial-region gap the authors report against PanoFlow.
  • Real-world support is currently qualitative on OmniPhotos and ODVista; a quantitative test with pseudo-ground-truth from a depth sensor or a multi-camera rig would show whether the synthetic-dataset gains survive real stitching and lens artifacts.
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 / 3 minor

Summary. The paper proposes PriOr-Flow, a dual-branch framework for panoramic optical flow estimation. The primitive branch operates on the original equirectangular view, while a second branch operates on an orthogonal view obtained by a 90-degree spherical rotation, which has low distortion near the poles. Two components are introduced: the Dual-Cost Collaborative Lookup (DCCL), which retrieves correlation cues from both cost volumes, and the Ortho-Driven Distortion Compensation (ODDC) module, which fuses motion features from both branches using confidence maps. The method is integrated with RAFT, GMA, and SKFlow. Experiments on MPFDataset and FlowScape report state-of-the-art EPE and SEPE, with particularly large gains in polar regions, and the code is publicly available.

Significance. If the reported results are correct, the paper makes a substantial empirical contribution: it is the first panoramic optical flow method that explicitly targets polar-region distortion through an orthogonal-view branch, and it demonstrates large, consistent improvements over prior SOTA on two public benchmarks. The proposed modules are architecture-agnostic, and the ablations in Tables 3 and 5 suggest that the method is broadly applicable and even effective with fewer iterations. The public code release is a strength, and the paper is honest about the equatorial trade-off in Table 7. However, the central mechanism depends on a coordinate transformation for flow fields that is not defined in the paper; this must be resolved before the significance of the empirical gains can be assessed.

major comments (3)
  1. [Section 3.1, 3.4, 3.5 (Eqs. 3, 12, 18)]
  2. [Section 3.2, Eq. (7)]
  3. [Section 4.3, Tables 1, 2, 4, 5]
minor comments (3)
  1. [Section 3.1, Eq. (3)]
  2. [Section 3.4, Eq. (13)]
  3. [Tables 1-7]

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: PriOr-Flow's orthogonal-view modules are trained and measured on held-out test data, with no equation reducing to a fitted target.

full rationale

The derivation is self-contained. The orthogonal view is a fixed spherical rotation and ERP re-projection (Eqs. 2-3), not a parameter fitted to the benchmark. DCCL and ODDC are learned modules evaluated on official test splits of MPFDataset and FlowScape; the reported EPE/SEPE reductions are therefore measurements, not identities by construction. The loss in Eq. 18 supervises each branch with ground truth transformed by the same spherical geometry, and although the paper reuses the image transform T for flow fields (Eqs. 12 and 18), that is a geometric correctness or implementation question, not a circular definition: the output flow is not defined as the fitted quantity. The citations to overlapping-author works [10,11,31,47,56] appear as background for related low-distortion or prior ideas, and the load-bearing low-distortion claim is independently justified by spherical projection geometry and external references [40,46]; no uniqueness theorem or author-imported constraint is used to force the design. No fitted parameter is renamed as a prediction, and no benchmark number reduces by construction. Thus there is no significant circularity.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The method rests on standard geometric formulas (ERP projection, spherical rotation) and on domain assumptions about the complementary distortion pattern and the reliability of group-wise correlation as a confidence signal. Free parameters are limited to a handful of training hyperparameters and the rotation angle/axis chosen by ablation. No new physical entities are introduced.

free parameters (4)
  • rotation_axis_and_angle = x-axis, 90 degrees
    Selected by ablation on the EFT test split (Table 4); the complementary-distortion prior is defined by this choice.
  • number_of_refinement_iterations = 12
    Set following RAFT and validated by ablation (Table 5); fewer iterations trade accuracy for speed.
  • loss_schedule_gamma = 0.8
    Hyperparameter from RAFT [41]; weights iteration losses in Eq. 18.
  • learning_rate = 1e-4, one-cycle
    Standard RAFT fine-tuning setting from Section 4.2.
assumptions (5)
  • standard math ERP projection is a linear map from spherical coordinates to image coordinates (Eq. 1).
    Used to define pixel positions and to generate the orthogonal view via spherical rotation (Eq. 2-3).
  • domain assumption Rotating the sphere by 90 degrees about the x-axis yields an ERP view whose distortion is complementary, i.e., original polar regions become equatorial.
    The central prior exploited by the method; Fig. 3 shows the complementary distortion maps.
  • domain assumption Group-wise correlation (Eq. 13) computed in the primitive view is a valid confidence estimate for both primitive and re-projected orthogonal flows.
    ODDC fusion weights rely on this; the paper validates it only empirically (Table 2).
  • domain assumption The DCCL local neighborhood in the primitive grid, after spherical rotation to the orthogonal grid, is a meaningful correlation neighborhood in the orthogonal cost volume.
    Eq. 6 assumes that a Euclidean neighborhood around a warped pixel remains a useful neighborhood after rotation and reprojection, despite grid resampling.
  • domain assumption Pre-trained RAFT weights on FlyingThings transfer to panoramic optical flow after fine-tuning.
    All initialized models and baselines rely on this transfer (Section 4.2).

how reviews work

0 comments
Cite this review

Pith. "Pith review of PriOr-Flow: Enhancing Primitive Panoramic Optical Flow with Orthogonal View." pith.science (2026). https://pith.science/paper/6D5Q7FCJ

@misc{pith2026250623897,
  author       = {Pith},
  title        = {Pith review of: PriOr-Flow: Enhancing Primitive Panoramic Optical Flow with Orthogonal View},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6D5Q7FCJ}},
  note         = {Machine review of arXiv:2506.23897}
}
read the original abstract

Panoramic optical flow enables a comprehensive understanding of temporal dynamics across wide fields of view. However, severe distortions caused by sphere-to-plane projections, such as the equirectangular projection (ERP), significantly degrade the performance of conventional perspective-based optical flow methods, especially in polar regions. To address this challenge, we propose PriOr-Flow, a novel dual-branch framework that leverages the low-distortion nature of the orthogonal view to enhance optical flow estimation in these regions. Specifically, we introduce the Dual-Cost Collaborative Lookup (DCCL) operator, which jointly retrieves correlation information from both the primitive and orthogonal cost volumes, effectively mitigating distortion noise during cost volume construction. Furthermore, our Ortho-Driven Distortion Compensation (ODDC) module iteratively refines motion features from both branches, further suppressing polar distortions. Extensive experiments demonstrate that PriOr-Flow is compatible with various perspective-based iterative optical flow methods and consistently achieves state-of-the-art performance on publicly available panoramic optical flow datasets, setting a new benchmark for wide-field motion estimation. The code is publicly available at: https://github.com/longliangLiu/PriOr-Flow.

Figures

Figures reproduced from arXiv: 2506.23897 by the authors.

Figure 1
Figure 1. Row 1: Comparisons with state-of-the-art panoramic optical flow estimation methods on MPFDataset [28] and FlowS￾cape [35]. Row 2: Visual comparison on the virtual dataset FlowScape [35]. Row 3: Visual comparison on the real dataset ODVista [42]. Our method performs well in the polar regions. cess on well-established benchmarks. However, the increas￾ing availability of panoramic cameras has created a press￾ing demand… view at source ↗
Figure 3
Figure 3. Distortion maps of the primitive view and the orthogonal [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figure 2
Figure 2. The orthogonal view is obtained by rotating the spheri [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Overview of Our proposed PriOr-Flow (PriOr-RAFT version). The Dual-Cost Collaborative Lookup (DCCL) operator uses the current optical flow to jointly retrieve information from both the primitive and orthogonal cost volumes. In each iteration, the Ortho￾Driven Distortio…
Figure 5
Figure 5. Figure 5: Schematic diagram of the Dual-Cost Collaborative [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 6
Figure 6. Figure 6: Qualitative results on the test set of MPFDataset. Our PriOr-RAFT outperforms SLOF in polar regions. [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: Qualitative results on the test set of FlowScape. Our PriOr-RAFT outperforms PanoFlow in the polar regions as well as in [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 8
Figure 8. Figure 8: Qualitative results on real-world scenarios. The left two columns present the results on the OmniPhotos dataset, while the right [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

57 extracted references · 50 canonical work pages

  1. [1]

    A survey of represen- tation learning, optimization strategies, and applications for omnidirectional vision

    Hao Ai, Zidong Cao, and Lin Wang. A survey of represen- tation learning, optimization strategies, and applications for omnidirectional vision. arXiv preprint arXiv:2502.10444 ,

    arXiv

  2. [2]

    Omniflownet: a perspective neural network adaptation for optical flow estimation in om- nidirectional images

    Charles-Olivier Artizzu, Haozhou Zhang, Guillaume Allib- ert, and C ´edric Demonceaux. Omniflownet: a perspective neural network adaptation for optical flow estimation in om- nidirectional images. In 2020 25th International Conference on Pattern Recognition (ICPR) , pages 2657–2662. IEEE,

    work page 2020

  3. [3]

    Vi- sual distortions in 360° videos

    Roberto G de A Azevedo, Neil Birkbeck, Francesca De Si- mone, Ivan Janatra, Balu Adsumilli, and Pascal Frossard. Vi- sual distortions in 360° videos. IEEE Transactions on Cir- cuits and Systems for Video Technology , 30(8):2524–2537,

    work page

  4. [4]

    Omniphotos: casual 360 vr photography

    Tobias Bertel, Mingze Yuan, Reuben Lindroos, and Christian Richardt. Omniphotos: casual 360 vr photography. ACM Transactions on Graphics (TOG), 39(6):1–12, 2020. 6, 8

    work page 2020

  5. [5]

    Revisiting optical flow estimation in 360 videos

    Keshav Bhandari, Ziliang Zong, and Yan Yan. Revisiting optical flow estimation in 360 videos. In 2020 25th Inter- national Conference on Pattern Recognition (ICPR) , pages 8196–8203. IEEE, 2021. 1, 3

    work page 2020

  6. [6]

    Learning omnidirectional flow in 360 video via siamese representation

    Keshav Bhandari, Bin Duan, Gaowen Liu, Hugo Latapie, Ziliang Zong, and Yan Yan. Learning omnidirectional flow in 360 video via siamese representation. In European Con- ference on Computer Vision, pages 557–574. Springer, 2022. 2, 3, 8

    work page 2022

  7. [7]

    Exploring simple siamese rep- resentation learning

    Xinlei Chen and Kaiming He. Exploring simple siamese rep- resentation learning. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 15750–15758, 2021. 2, 3

    work page 2021

  8. [8]

    Re- gion separable stereo matching

    Junda Cheng, Xin Yang, Yuechuan Pu, and Peng Guo. Re- gion separable stereo matching. IEEE Transactions on Mul- timedia, 25:4880–4893, 2022. 1

    work page 2022

Show all 57 references
  1. [9]

    Coa- trsnet: Fully exploiting convolution and attention for stereo matching by region separation

    Junda Cheng, Gangwei Xu, Peng Guo, and Xin Yang. Coa- trsnet: Fully exploiting convolution and attention for stereo matching by region separation. International Journal of Computer Vision, 132(1):56–73, 2024. 1

    2024

  2. [10]

    Adaptive fusion of single-view and multi-view depth for autonomous driving

    Junda Cheng, Wei Yin, Kaixuan Wang, Xiaozhi Chen, Shijie Wang, and Xin Yang. Adaptive fusion of single-view and multi-view depth for autonomous driving. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10138–10147, 2024. 2

    2024

  3. [11]

    Monster: Marry monodepth to stereo unleashes power

    Junda Cheng, Longliang Liu, Gangwei Xu, Xianqi Wang, Zhaoxing Zhang, Yong Deng, Jinliang Zang, Yurui Chen, Zhipeng Cai, and Xin Yang. Monster: Marry monodepth to stereo unleashes power. arXiv preprint arXiv:2501.08643 ,

  4. [12]

    Spherenet: Learning spherical representations for detection and classification in omnidirectional images

    Benjamin Coors, Alexandru Paul Condurache, and Andreas Geiger. Spherenet: Learning spherical representations for detection and classification in omnidirectional images. In Proceedings of the European conference on computer vision (ECCV), pages 518–533, 2018. 8

    2018

  5. [13]

    Deformable convolutional networks

    Jifeng Dai, Haozhi Qi, Yuwen Xiong, Yi Li, Guodong Zhang, Han Hu, and Yichen Wei. Deformable convolutional networks. In Proceedings of the IEEE international confer- ence on computer vision, pages 764–773, 2017. 2, 3

    2017

  6. [14]

    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. In Pro- ceedings of the IEEE international conference on computer...

    2015

  7. [15]

    Corners for layout: End-to-end layout recovery from 360 images

    Clara Fernandez-Labrador, Jose M Facil, Alejandro Perez- Yus, C´edric Demonceaux, Javier Civera, and Jose J Guer- rero. Corners for layout: End-to-end layout recovery from 360 images. IEEE Robotics and Automation Letters , 5(2): 1255–1262, 2020. 3, 8

    2020

  8. [16]

    Flow-edge guided video completion

    Chen Gao, Ayush Saraf, Jia-Bin Huang, and Johannes Kopf. Flow-edge guided video completion. In Computer Vision– ECCV 2020: 16th European Conference, Glasgow, UK, Au- gust 23–28, 2020, Proceedings, Part XII 16, pages 713–729. Springer, 2020. 1

    2020

  9. [17]

    Are we ready for autonomous driving? the kitti vision benchmark suite

    Andreas Geiger, Philip Lenz, and Raquel Urtasun. Are we ready for autonomous driving? the kitti vision benchmark suite. In 2012 IEEE conference on computer vision and pat- tern recognition, pages 3354–3361. IEEE, 2012. 1

    2012

  10. [18]

    Group-wise correlation stereo network

    Xiaoyang Guo, Kai Yang, Wukui Yang, Xiaogang Wang, and Hongsheng Li. Group-wise correlation stereo network. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 3273–3282, 2019. 5

    2019

  11. [19]

    Gradi- ent operators for feature extraction from omnidirectional panoramic images

    Kenji Hara, Kohei Inoue, and Kiichi Urahama. Gradi- ent operators for feature extraction from omnidirectional panoramic images. Pattern Recognition Letters, 54:89–96,

  12. [20]

    Rife: real-time intermediate flow estimation for video frame interpolation 2020, 2011

    Zhewei Huang, Tianyuan Zhang, Wen Heng, Boxin Shi, and Shuchang Zhou. Rife: real-time intermediate flow estimation for video frame interpolation 2020, 2011. 1

    2020

  13. [21]

    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 European conference on computer vision, pages 668–685. Springer, 2022. 1, 3

    2022

  14. [22]

    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 Proceedings of the IEEE/CVF international conference on computer vision , pages 9772– 9781, 2021. 1, 3, 7

    2021

  15. [23]

    E-cnn: Ac- curate spherical camera rotation estimation via uniformiza- tion of distorted optical flow fields

    Dabae Kim, Sarthak Pathak, Alessandro Moro, Ren Ko- matsu, Atsushi Yamashita, and Hajime Asama. E-cnn: Ac- curate spherical camera rotation estimation via uniformiza- tion of distorted optical flow fields. In ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech ...

    2019

  16. [24]

    Deep video inpainting

    Dahun Kim, Sanghyun Woo, Joon-Young Lee, and In So Kweon. Deep video inpainting. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5792–5801, 2019. 1

    2019

  17. [25]

    Fast optical flow using dense inverse search

    Till Kroeger, Radu Timofte, Dengxin Dai, and Luc Van Gool. Fast optical flow using dense inverse search. In Computer Vision–ECCV 2016: 14th European Conference, Amster- dam, The Netherlands, October 11–14, 2016, Proceedings, Part IV 14, pages 471–488. Springer, 2016. 8

    2016

  18. [26]

    Omnis- plat: Taming feed-forward 3d gaussian splatting for omni- directional images with editable capabilities

    Suyoung Lee, Jaeyoung Chung, Kihoon Kim, Jaeyoo Huh, Gunhee Lee, Minsoo Lee, and Kyoung Mu Lee. Omnis- plat: Taming feed-forward 3d gaussian splatting for omni- directional images with editable capabilities. arXiv preprint arXiv:2412.16604, 2024. 1

    2024 arXiv

  19. [27]

    Practical stereo matching via cascaded re- current network with adaptive correlation

    Jiankun Li, Peisen Wang, Pengfei Xiong, Tao Cai, Zi- wei Yan, Lei Yang, Jiangyu Liu, Haoqiang Fan, and Shuaicheng Liu. Practical stereo matching via cascaded re- current network with adaptive correlation. In Proceedings of the IEEE/CVF conference on computer vision and pattern...

    2022

  20. [28]

    Deep 360° optical flow estimation based on multi- projection fusion

    Yiheng Li, Connelly Barnes, Kun Huang, and Fang-Lue Zhang. Deep 360° optical flow estimation based on multi- projection fusion. In European Conference on Computer Vi- sion, pages 336–352. Springer, 2022. 1, 2, 3, 6, 8

    2022

  21. [29]

    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

  22. [30]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101, 2017. 6

    2017 arXiv

  23. [31]

    Elfnet: Evidential local-global fusion for stereo matching

    Jieming Lou, Weide Liu, Zhuo Chen, Fayao Liu, and Jun Cheng. Elfnet: Evidential local-global fusion for stereo matching. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 17784–17793, 2023. 2

    2023

  24. [32]

    Multiview stereo with cascaded epipolar raft

    Zeyu Ma, Zachary Teed, and Jia Deng. Multiview stereo with cascaded epipolar raft. In European Conference on Com- puter Vision, pages 734–750. Springer, 2022. 1

    2022

  25. [33]

    A large dataset to train convolutional networks for disparity, optical flow, and scene flow estimation

    Nikolaus Mayer, Eddy Ilg, Philip Hausser, Philipp Fischer, Daniel Cremers, Alexey Dosovitskiy, and Thomas Brox. A large dataset to train convolutional networks for disparity, optical flow, and scene flow estimation. In Proceedings of the IEEE conference on computer vision and ...

    2016

  26. [34]

    Csflow: Learning optical flow via cross strip cor- relation for autonomous driving

    Hao Shi, Yifan Zhou, Kailun Yang, Xiaoting Yin, and Kai- wei Wang. Csflow: Learning optical flow via cross strip cor- relation for autonomous driving. In 2022 IEEE intelligent vehicles symposium (IV), pages 1851–1858. IEEE, 2022. 8

    2022

  27. [35]

    Panoflow: Learning 360° optical flow for surrounding tem- poral understanding

    Hao Shi, Yifan Zhou, Kailun Yang, Xiaoting Yin, Ze Wang, Yaozu Ye, Zhe Yin, Shi Meng, Peng Li, and Kaiwei Wang. Panoflow: Learning 360° optical flow for surrounding tem- poral understanding. IEEE Transactions on Intelligent Transportation Systems, 24(5):5570–5585, 2023. 1, 2, 3, 6, 8

    2023

  28. [36]

    Flattening the earth: two thousand years of map projections

    John P Snyder. Flattening the earth: two thousand years of map projections. University of Chicago Press, 1997. 3

    1997

  29. [37]

    Kernel transformer net- works for compact spherical convolution

    Yu-Chuan Su and Kristen Grauman. Kernel transformer net- works for compact spherical convolution. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9442–9451, 2019. 3

    2019

  30. [38]

    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,

  31. [39]

    Skflow: Learning optical flow with super kernels

    Shangkun Sun, Yuanqi Chen, Yu Zhu, Guodong Guo, and Ge Li. Skflow: Learning optical flow with super kernels. Advances in Neural Information Processing Systems , 35: 11313–11326, 2022. 1, 3, 7

    2022

  32. [40]

    Weighted-to-spherically- uniform quality evaluation for omnidirectional video

    Yule Sun, Ang Lu, and Lu Yu. Weighted-to-spherically- uniform quality evaluation for omnidirectional video. IEEE signal processing letters, 24(9):1408–1412, 2017. 2

    2017

  33. [41]

    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

  34. [42]

    Odvista: An omnidirectional video dataset for super-resolution and quality enhancement tasks

    Ahmed Telili, Ibrahim Farhat, Wassim Hamidouche, and Hadi Amirpour. Odvista: An omnidirectional video dataset for super-resolution and quality enhancement tasks. In 2024 IEEE International Conference on Image Processing (ICIP), pages 131–136. IEEE, 2024. 1, 6, 8

    2024

  35. [43]

    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 Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8121–8130,

  36. [44]

    Deep flow-guided video inpainting

    Rui Xu, Xiaoxiao Li, Bolei Zhou, and Chen Change Loy. Deep flow-guided video inpainting. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3723–3732, 2019. 1

    2019

  37. [45]

    Quadratic video interpolation

    Xiangyu Xu, Li Siyao, Wenxiu Sun, Qian Yin, and Ming- Hsuan Yang. Quadratic video interpolation. Advances in Neural Information Processing Systems, 32, 2019. 1

    2019

  38. [46]

    Osrt: Omnidirectional image super- resolution with distortion-aware transformer

    Fanghua Yu, Xintao Wang, Mingdeng Cao, Gen Li, Ying Shan, and Chao Dong. Osrt: Omnidirectional image super- resolution with distortion-aware transformer. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 13283–13292, 2023. 2

    2023

  39. [47]

    Frequency and spatial dual guidance for im- age dehazing

    Hu Yu, Naishan Zheng, Man Zhou, Jie Huang, Zeyu Xiao, and Feng Zhao. Frequency and spatial dual guidance for im- age dehazing. In European conference on computer vision , pages 181–198. Springer, 2022. 2

    2022

  40. [48]

    360 optical flow us- ing tangent images

    Mingze Yuan and Christian Richardt. 360 optical flow us- ing tangent images. In British Machine Vision Confer- ence:(BMVC). Christian Richardt, 2021. 2, 3, 6, 8

    2021

  41. [49]

    Se- marflow: Injecting semantics into unsupervised optical flow estimation for autonomous driving

    Shuai Yuan, Shuzhi Yu, Hannah Kim, and Carlo Tomasi. Se- marflow: Injecting semantics into unsupervised optical flow estimation for autonomous driving. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 9566–9577, 2023. 1

    2023

  42. [50]

    Sdv-loam: Semi-direct visual–lidar odometry and mapping

    Zikang Yuan, Qingjie Wang, Ken Cheng, Tianyu Hao, and Xin Yang. Sdv-loam: Semi-direct visual–lidar odometry and mapping. IEEE Transactions on Pattern Analysis and Ma- chine Intelligence, 45(9):11203–11220, 2023

    2023

  43. [51]

    Sr-livo: Lidar-inertial-visual odometry and mapping with sweep reconstruction

    Zikang Yuan, Jie Deng, Ruiye Ming, Fengtian Lang, and Xin Yang. Sr-livo: Lidar-inertial-visual odometry and mapping with sweep reconstruction. IEEE Robotics and Automation Letters, 2024

    2024

  44. [52]

    Sr-lio: Lidar-inertial odometry with sweep reconstruction

    Zikang Yuan, Fengtian Lang, Tianle Xu, and Xin Yang. Sr-lio: Lidar-inertial odometry with sweep reconstruction. In 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 7862–7869. IEEE, 2024. 1

    2024

  45. [53]

    V oxel-svio: Stereo visual-inertial odometry based on voxel map

    Zikang Yuan, Fengtian Lang, Jie Deng, Hongcheng Luo, and Xin Yang. V oxel-svio: Stereo visual-inertial odometry based on voxel map. IEEE Robotics and Automation Letters, 2025. 1

    2025

  46. [54]

    Saliency detection in 360 videos

    Ziheng Zhang, Yanyu Xu, Jingyi Yu, and Shenghua Gao. Saliency detection in 360 videos. In Proceedings of the Eu- ropean conference on computer vision (ECCV) , pages 488– 503, 2018. 6

    2018

  47. [55]

    Global matching with overlapping at- tention for optical flow estimation

    Shiyu Zhao, Long Zhao, Zhixing Zhang, Enyu Zhou, and Dimitris Metaxas. Global matching with overlapping at- tention for optical flow estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17592–17601, 2022. 1, 3

    2022

  48. [56]

    Improving scene text image super-resolution via dual prior modulation network

    Shipeng Zhu, Zuoyan Zhao, Pengfei Fang, and Hui Xue. Improving scene text image super-resolution via dual prior modulation network. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 3843–3851, 2023. 2

    2023

  49. [57]

    View synthesis with sculpted neu- ral points

    Yiming Zuo and Jia Deng. View synthesis with sculpted neu- ral points. arXiv preprint arXiv:2205.05869, 2022. 1

    2022 arXiv

Pith tools

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