Pith. sign in

REVIEW 3 major objections 6 minor 45 references

HDR video can be rebuilt without warping frames if a continuous medium-exposure stream is treated as the structural backbone.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

Dual-stream HDR video can be reconstructed without fragile cross-exposure warping by backbone-guided collaborative attention plus sequence-level residual refinement.

T0 review reviewed 2026-07-14 challenge →

load-bearing objection Solid dual-stream HDR video method with a clear backbone-plus-routing reformulation; SOTA claims are real under that capture setup, but the abstract oversells “alignment-free” as a general fix for alternating-exposure pipelines. the 3 major comments →

arxiv 2603.14377 v2 pith:UIU2TBP2 submitted 2026-03-15 cs.CV

LoCAtion: Long-time Collaborative Attention Framework for High Dynamic Range Video Reconstruction

Qianyu Zhang , Bolun Zheng , Lingyu Zhu , Aiai Huang , Zongpeng Li , Shiqi Wang This is my paper
classification cs.CV
keywords HDR video reconstructiondual-stream capturealignment-free fusioncollaborative attentiontemporal consistencymedium-exposure backbonecross-exposure routing
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

Most HDR video methods try to warp differently exposed frames into alignment and then fuse them. That works when motion is mild, but large motion, occlusion, and exposure-driven appearance changes turn registration errors into ghosts and flicker. This paper argues that dual-stream capture changes the problem: a continuous medium-exposure stream already supplies a coherent scene structure, while sparse low- and high-exposure frames only need to supply missing dynamic-range cues. LoCAtion therefore skips optical-flow warping of those anchors. It routes reliable low-frequency cues into the medium-exposure backbone, then uses bidirectional and long-range temporal modeling to correct residual inconsistencies across the whole sequence. The result is higher reconstruction quality and more stable temporal profiles on hard dynamic scenes, with moderate model size and latency.

Core claim

On dual-stream inputs, HDR video reconstruction is better cast as backbone-guided cross-exposure compensation plus sequence-level residual refinement than as reference-centered alignment-and-fusion. Anchoring on the continuous medium-exposure stream, selectively injecting unaligned low/high exposure cues via reliability-guided collaborative attention, and propagating corrections with a global sequence solver yields stronger quality and temporal stability—especially on fast-motion and complex scenes—without explicit optical-flow warping of the exposure anchors.

What carries the argument

LoCAtion: Collaborative Feature Attention (CFA) estimates time-dependent reliability maps on low-frequency wavelet features of sparse low/high anchors and injects only reliable cues into the medium-exposure backbone (Temporal Prior Aggregator + Content-Faithful Fusion), then Global Sequence Consistency (GSC) predicts a residual via bidirectional context aggregation and long-range RWKV modeling to enforce whole-sequence coherence.

Load-bearing premise

The method assumes dual-stream capture already provides a continuous medium-exposure backbone and that low-frequency reliability maps alone can decide which unaligned exposure cues are safe to inject, leaving residual motion errors for later sequence correction.

What would settle it

On dual-stream sequences with large motion and saturation, measure whether removing CFA reliability routing (or replacing it with optical-flow warping of anchors) raises ghosting and temporal STD, and whether increasing the medium-exposure segment length beyond five frames fails to improve t-PSNR/t-SSIM while memory grows—if quality and stability do not track those ablations, the backbone-plus-routing claim fails.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes LoCAtion for dual-stream HDR video reconstruction. It treats the continuous medium-exposure stream as a structural backbone and sparse low/high-exposure frames as dynamic-range anchors, replacing explicit optical-flow warping of anchors with Collaborative Feature Attention (CFA: TPA reliability maps on low-frequency DWT features plus Content-Faithful Fusion) and refining intermediate estimates with Global Sequence Consistency (GSC: bidirectional residual propagation plus RWKV long-range modeling). Training uses tone-mapped L1, temporal-difference, and anchor-consistency losses. On a 16-scene synthetic dual-stream benchmark and real dual-camera sequences, the method reports leading spatial and temporal metrics, especially on dynamic/complex scenes, with competitive latency versus Transformer dual-stream baselines.

Significance. If the gains hold under the dual-stream setting of [14], the work is a solid systems contribution: it converts hardware continuity into an algorithmic prior, avoids fragile cross-exposure warping of anchors, and couples reliability-gated fusion with sequence-level residual refinement. Strengths include multi-metric evaluation (PSNRμ/SSIMμ, t-PSNR/t-SSIM, FovVideoVDP, STD, real no-reference temporal metrics), ablations of TPA/BCA/LRTM/DWT and design alternatives (flow/local/cross attention), temporal-length and anchor-ratio studies, and qualitative temporal profiles. The contribution is most meaningful for dual-stream HDR pipelines rather than as a general replacement for alternating-exposure alignment-and-fusion; within that scope it is useful and carefully engineered.

major comments (3)
  1. Abstract and §I frame the contribution as escaping the fragile alignment-and-fusion paradigm of prevailing HDR video methods, but the method and all main results (§III-A, Tables I–II) assume dual-stream capture with a continuous medium-exposure backbone plus sparse anchors from [14]. Table I therefore compares AE frame-centric methods (single alternating stream) to DS methods (continuous medium stream + anchors). The largest gains appear on the 10 dynamic/complex scenes and real no-reference metrics, where structural continuity is already supplied by the medium stream. The abstract and claims should be scoped to dual-stream reconstruction, or the paper should add a same-input comparison that forces AE video methods onto dual-stream inputs (or LoCAtion onto pure alternating inputs) so the reformulation claim is not confounded with the capture prior.
  2. Table III and Fig. 5 ablate CFA/GSC only under the fixed dual-stream protocol; they show that TPA, DWT-domain fusion, BCA, and LRTM help relative to flow/local/cross-attention alternatives on the same DS input, but they do not isolate whether CFA reliability maps on low-frequency DWT components (Eqs. 1–3) plus GSC residual refinement would still beat strong AE video methods when both sides share identical inputs. Without that isolation, the load-bearing claim that alignment-free collaborative routing (rather than the dual-stream backbone itself) solves fragile alignment remains only partially supported. A controlled same-input experiment or a clear limitation statement is needed.
  3. §III-B and the weakest design assumption: reliability is estimated only on low-frequency DWT features and used as soft gates, with high-frequency content taken from the medium backbone and residual motion/occlusion left to GSC. The paper does not quantify failure modes when low-frequency differences mix motion and exposure change (large occlusion, saturation boundaries, or when anchors are temporally far). Table IV shows sensitivity to T and medium-to-anchor ratio, but there is no stress evaluation of TPA mis-gating. A short failure analysis or quantitative breakdown on motion/occlusion subsets would make the alignment-free claim more credible.
minor comments (6)
  1. Eqs. (2)–(3) and (6)–(8): architectural details of G→/G←, B(k), and the number of RWKV blocks K are underspecified (channels, kernel sizes, K). A short architecture table would aid reproducibility.
  2. §III-D: λ_temp and λ_anc are introduced without reported values or sensitivity; please state the training weights used for the main results.
  3. Fig. 2 caption and body: CFF is described at a high level (soft gates + inverse DWT) without an explicit fusion equation; adding one would match the precision of TPA/GSC.
  4. Table I header and text: clarify that AE and DS columns use different input streams generated from the same source clips, so absolute cross-protocol ranking is not apples-to-apples.
  5. Real evaluation (Table II) is no-reference only; stating this limitation more prominently in the abstract/conclusion would set expectations.
  6. Minor polish: arXiv-style citations such as EAFNet (arxiv’26) and AFUNet (ICCV’25) should be checked for final venue status; a few figure labels (e.g., temporal profiles) would benefit from larger fonts.

Circularity Check

1 steps flagged

No equation-level or prediction-by-construction circularity; only minor self-citation of the dual-stream capture setting that enables (but does not tautologically force) the claimed architecture.

specific steps
  1. self citation load bearing [§III-A Motivation and Problem Formulation; also Related Work and [14]]
    "We build LoCAtion upon the dual-stream capture setup of [14]. For a temporal segment, the input consists of a continuous medium-exposure stream Y2={y2,t}Tt=1 and a pair of low-/high-exposure anchors {y1,y3}. ... Instead of explicitly warping the exposure anchors to each medium-exposure frame, we formulate the task as backbone-guided cross-exposure compensation followed by sequence-level temporal refinement."

    The reformulation from alignment-and-fusion to backbone-guided routing is only possible because of the continuous medium-exposure stream introduced in the authors’ own prior dual-stream work [14] (same core authors). Real evaluation data and the EAFNet baseline also come from [14]. This is a mild self-citation dependence for the enabling input prior, not a reduction of the CFA/GSC equations or the reported metrics to a tautology; gains over other dual-stream methods remain empirical.

full rationale

LoCAtion is an empirical methods paper: it proposes CFA (TPA reliability maps on low-frequency DWT components + CFF) and GSC (BCA + LRTM residual) for dual-stream HDR video, trains with standard L1 + temporal + anchor-consistency losses, and reports PSNR/SSIM/t-metrics/FovVideoVDP/STD against synthetic ground truth derived from external Vimeo-90K and Cinematic clips plus no-reference metrics on real sequences. There are no first-principles derivations, uniqueness theorems, fitted parameters renamed as independent predictions, or equations that reduce to their own inputs by construction (Eqs. 1–12 define modules and losses, not tautologies). Self-citation to the authors’ dual-stream setup and EAFNet baseline [14] is present and supplies the continuous medium-exposure backbone that the reformulation exploits, but this is ordinary prior-work dependence for an input protocol; the architectural gains and SOTA numbers are measured against other methods under fixed protocols and ablations (Tables I–IV), not forced by the citation. Score 1 reflects that single non-load-bearing self-citation of the enabling capture setting; the central empirical claims remain independently testable.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 2 invented entities

The central claim rests on the dual-stream capture model, the sufficiency of low-frequency reliability routing without warping, hand-chosen temporal window and loss weights, and architectural modules (TPA, CFF, BCA, LRTM/RWKV) introduced for this pipeline. No new physical entities; free parameters are engineering knobs and training losses rather than fitted physical constants.

free parameters (5)
  • medium-exposure segment length T
    Chosen as T=5 for main results after ablation over T in {1,5,7,10,15,20}; directly controls temporal context and memory and affects reported metrics.
  • medium-to-anchor sampling ratio
    Main setting 5:1 selected after comparing 1:1 through 55:1; changes temporal stability and is not derived.
  • loss weights λ_temp and λ_anc
    Appear in L_total but numerical values are not stated; they control the trade-off among reconstruction, flicker, and anchor consistency.
  • tone-mapping constant κ=5000
    Fixed in τ(·) for supervision and metrics; standard in HDR literature but still a chosen scale for the objective.
  • number of RWKV blocks K and BCA passes
    Architectural depth/pass count for LRTM/BCA chosen by design; not derived from first principles.
axioms (4)
  • domain assumption Dual-stream capture provides a continuous medium-exposure structural backbone plus sparse complementary low/high anchors.
    Problem formulation §III-A and hardware setting of [14]; without this input the backbone-guided claim does not apply.
  • ad hoc to paper Cross-exposure reliability can be estimated on low-frequency DWT features and used as soft gates without explicit optical-flow warping of anchors.
    Eqs. (1)–(3) and CFA design; high-frequency medium-exposure content is preserved by construction.
  • domain assumption Sequence-level residual refinement with bidirectional recurrence and RWKV long-range modeling improves temporal coherence of intermediate HDR estimates.
    GSC stage §III-C; standard video-restoration prior specialized to dual-stream HDR.
  • domain assumption Synthetic multi-exposure generation from Vimeo-90K and cinematic HDR clips is a valid proxy for training/evaluation of dual-stream HDR video.
    §IV-A dataset construction; real tests lack HDR ground truth.
invented entities (2)
  • Collaborative Feature Attention (CFA) with Temporal Prior Aggregator and Content-Faithful Fusion no independent evidence
    purpose: Route reliable low/high-exposure cues into the medium-exposure backbone without warping.
    Paper-specific module stack; evidence is internal ablations and metrics, not an independent physical measurement.
  • Global Sequence Consistency (GSC) residual solver (BCA + LRTM/RWKV) no independent evidence
    purpose: Propagate bidirectional and long-range corrections across the reconstruction segment.
    Architectural invention for this task; validated only within the paper’s experiments.

reviewed 2026-07-14 · how reviews work

0 comments
Cite this review

Pith. "Pith review of LoCAtion: Long-time Collaborative Attention Framework for High Dynamic Range Video Reconstruction." pith.science (2026). https://pith.science/paper/UIU2TBP2

@misc{pith2026260314377,
  author       = {Pith},
  title        = {Pith review of: LoCAtion: Long-time Collaborative Attention Framework for High Dynamic Range Video Reconstruction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UIU2TBP2}},
  note         = {Machine review of arXiv:2603.14377}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Prevailing High Dynamic Range (HDR) video reconstruction methods are fundamentally trapped in a fragile alignment-and-fusion paradigm. While explicit spatial alignment can successfully recover fine details in controlled environments, it becomes a severe bottleneck in unconstrained dynamic scenes. By forcing rigid alignment across unpredictable motions and varying exposures, these methods inevitably translate registration errors into severe ghosting artifacts and temporal flickering. In this paper, we rethink this conventional prerequisite. Recognizing that explicit alignment is inherently vulnerable to real-world complexities, we propose LoCAtion, a Long-time Collaborative Attention framework that reformulates HDR video generation from a fragile spatial warping task into a robust, alignment-free collaborative feature routing problem. Guided by this new formulation, our architecture explicitly decouples the highly entangled reconstruction task. Rather than struggling to rigidly warp neighboring frames, we anchor the scene on a continuous medium-exposure backbone and utilize collaborative attention to dynamically harvest and inject reliable irradiance cues from unaligned exposures. Furthermore, we introduce a learned global sequence solver. By leveraging bidirectional context and long-range temporal modeling, it propagates corrective signals and structural features across the entire sequence, inherently enforcing whole-video coherence and eliminating jitter. Extensive experiments demonstrate that LoCAtion achieves state-of-the-art visual quality and temporal stability, offering a highly competitive balance between accuracy and computational efficiency.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

45 extracted references · 4 linked inside Pith

  1. [1]

    Recovering high dynamic range radiance maps from photographs,

    P. E. Debevec and J. Malik, “Recovering high dynamic range radiance maps from photographs,” inACM SIGGRAPH 2008 classes, 2008, pp. 1–10

    2008

  2. [2]

    Bidirectional translation between uhd-hdr and hd-sdr videos,

    M. Yao, D. He, X. Li, Z. Pan, and Z. Xiong, “Bidirectional translation between uhd-hdr and hd-sdr videos,”IEEE Transactions on Multimedia, vol. 25, pp. 8672–8686, 2023

    2023

  3. [3]

    Hdr video reconstruction with tri- exposure quad-bayer sensors,

    Y . Jiang, I. Choi, J. Jiang, and J. Gu, “Hdr video reconstruction with tri- exposure quad-bayer sensors,”arXiv preprint arXiv:2103.10982, 2021

    Pith/arXiv arXiv 2021

  4. [4]

    Neuromorphic camera guided high dynamic range imaging,

    J. Han, C. Zhou, P. Duan, Y . Tang, C. Xu, C. Xu, T. Huang, and B. Shi, “Neuromorphic camera guided high dynamic range imaging,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 1730–1739

    2020

  5. [5]

    High dynamic range video,

    S. B. Kang, M. Uyttendaele, S. Winder, and R. Szeliski, “High dynamic range video,”ACM Transactions on Graphics (TOG), vol. 22, no. 3, pp. 319–325, 2003

    2003

  6. [6]

    Deep hdr video from sequences with alternating exposures,

    N. K. Kalantari and R. Ramamoorthi, “Deep hdr video from sequences with alternating exposures,” inComputer graphics forum, vol. 38, no. 2. Wiley Online Library, 2019, pp. 193–205

    2019

  7. [7]

    Two exposure fusion using prior- aware generative adversarial network,

    J.-L. Yin, B.-H. Chen, and Y .-T. Peng, “Two exposure fusion using prior- aware generative adversarial network,”IEEE Transactions on Multime- dia, vol. 24, pp. 2841–2851, 2021

    2021

  8. [8]

    Hdr video reconstruction: A coarse-to-fine network and a real-world benchmark dataset,

    G. Chen, C. Chen, S. Guo, Z. Liang, K.-Y . K. Wong, and L. Zhang, “Hdr video reconstruction: A coarse-to-fine network and a real-world benchmark dataset,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 2502–2511

    2021

  9. [9]

    Exposure completing for temporally consistent neural high dynamic range video rendering,

    J. Cui, W. Jiang, Z. Peng, Z. Pan, and Z. Cao, “Exposure completing for temporally consistent neural high dynamic range video rendering,” in Proceedings of the 32nd ACM International Conference on Multimedia, 2024, pp. 10 027–10 035

    2024

  10. [10]

    Hdrflow: Real-time hdr video reconstruction with large motions,

    G. Xu, Y . Wang, J. Gu, T. Xue, and X. Yang, “Hdrflow: Real-time hdr video reconstruction with large motions,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 24 851–24 860

    2024

  11. [11]

    Ug- hdr: Uncertainty-guided hdr imaging under large motion and extreme exposure,

    J. Huang, L. Gao, R. Xiao, Y . Wang, L. Wang, C. Tang, and J. Lv, “Ug- hdr: Uncertainty-guided hdr imaging under large motion and extreme exposure,”IEEE Transactions on Multimedia, 2026

    2026

  12. [12]

    Deep sr-hdr: Joint learning of super-resolution and high dynamic range imaging for dynamic scenes,

    X. Tan, H. Chen, K. Xu, Y . Jin, and C. Zhu, “Deep sr-hdr: Joint learning of super-resolution and high dynamic range imaging for dynamic scenes,”IEEE Transactions on Multimedia, vol. 25, pp. 750–763, 2021

    2021

  13. [13]

    Lan-hdr: Luminance-based alignment network for high dynamic range video reconstruction,

    H. Chung and N. I. Cho, “Lan-hdr: Luminance-based alignment network for high dynamic range video reconstruction,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 12 760–12 769

    2023

  14. [14]

    Capturing stable hdr videos using a dual-camera system,

    Q. Zhang, B. Zheng, L. Zhu, H. Pan, Z. Zhu, Z. Li, and S. Wang, “Capturing stable hdr videos using a dual-camera system,”arXiv preprint arXiv:2507.06593, 2025

    arXiv 2025

  15. [15]

    Deep conditional hdri: Inverse tone mapping via dual encoder-decoder conditioning method,

    Y . Nam, J. Kim, J.-h. Shim, and S.-J. Kang, “Deep conditional hdri: Inverse tone mapping via dual encoder-decoder conditioning method,” IEEE Transactions on Multimedia, vol. 26, pp. 8504–8515, 2024

    2024

  16. [16]

    Extending dynamic range of monochrome and color im- ages through fusion,

    L. Bogoni, “Extending dynamic range of monochrome and color im- ages through fusion,” inProceedings 15th International Conference on Pattern Recognition. ICPR-2000, vol. 3. IEEE, 2000, pp. 7–12

    2000

  17. [17]

    Robust patch-based hdr reconstruction of dynamic scenes

    P. Sen, N. K. Kalantari, M. Yaesoubi, S. Darabi, D. B. Goldman, and E. Shechtman, “Robust patch-based hdr reconstruction of dynamic scenes.”ACM Trans. Graph., vol. 31, no. 6, pp. 203–1, 2012

    2012

  18. [18]

    Hdr deghosting: How to deal with saturation?

    J. Hu, O. Gallo, K. Pulli, and X. Sun, “Hdr deghosting: How to deal with saturation?” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2013, pp. 1163–1170

    2013

  19. [19]

    Fast and robust high dynamic range image generation with camera and object movement,

    T. Groschet al., “Fast and robust high dynamic range image generation with camera and object movement,”Vision, Modeling and Visualization, RWTH Aachen, vol. 277284, no. 3, p. 2, 2006

    2006

  20. [20]

    Automatic high-dynamic range image generation for dynamic scenes,

    K. Jacobs, C. Loscos, and G. Ward, “Automatic high-dynamic range image generation for dynamic scenes,”IEEE Computer Graphics and Applications, vol. 28, no. 2, pp. 84–93, 2008

    2008

  21. [21]

    Gradient-directed multiexposure compo- sition,

    W. Zhang and W.-K. Cham, “Gradient-directed multiexposure compo- sition,”IEEE Transactions on Image Processing, vol. 21, no. 4, pp. 2318–2323, 2011

    2011

  22. [22]

    La-hdr: Light adaptive hdr reconstruction framework for single ldr image considering varied light conditions,

    X. Hu, L. Shen, M. Jiang, R. Ma, and P. An, “La-hdr: Light adaptive hdr reconstruction framework for single ldr image considering varied light conditions,”IEEE Transactions on Multimedia, vol. 25, pp. 4814–4829, 2022

    2022

  23. [23]

    Deep arbitrary hdri: Inverse tone mapping with controllable exposure changes,

    S. Y . Jo, S. Lee, N. Ahn, and S.-J. Kang, “Deep arbitrary hdri: Inverse tone mapping with controllable exposure changes,”IEEE Transactions on Multimedia, vol. 24, pp. 2713–2726, 2021

    2021

  24. [24]

    Deep high dynamic range imaging of dynamic scenes

    N. K. Kalantari, R. Ramamoorthiet al., “Deep high dynamic range imaging of dynamic scenes.”ACM Trans. Graph., vol. 36, no. 4, pp. 144–1, 2017

    2017

  25. [25]

    Deep hdr reconstruction of dynamic scenes,

    F. Peng, M. Zhang, S. Lai, H. Tan, and S. Yan, “Deep hdr reconstruction of dynamic scenes,” in2018 IEEE 3rd International Conference on Image, Vision and Computing (ICIVC). IEEE, 2018, pp. 347–351

    2018

  26. [26]

    A fast, scalable, and reliable deghosting method for extreme exposure fusion,

    K. R. Prabhakar, R. Arora, A. Swaminathan, K. P. Singh, and R. V . Babu, “A fast, scalable, and reliable deghosting method for extreme exposure fusion,” in2019 IEEE International Conference on Computational Photography (ICCP). IEEE, 2019, pp. 1–8

    2019

  27. [27]

    Safnet: Se- lective alignment fusion network for efficient hdr imaging,

    L. Kong, B. Li, Y . Xiong, H. Zhang, H. Gu, and J. Chen, “Safnet: Se- lective alignment fusion network for efficient hdr imaging,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 256–273

    2024

  28. [28]

    Attention-guided network for ghost-free high dynamic range imaging,

    Q. Yan, D. Gong, Q. Shi, A. v. d. Hengel, C. Shen, I. Reid, and Y . Zhang, “Attention-guided network for ghost-free high dynamic range imaging,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, pp. 1751–1760

    2019

  29. [29]

    Adnet: Attention-guided deformable convolutional network for high dynamic range imaging,

    Z. Liu, W. Lin, X. Li, Q. Rao, T. Jiang, M. Han, H. Fan, J. Sun, and S. Liu, “Adnet: Attention-guided deformable convolutional network for high dynamic range imaging,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 463– 470

    2021

  30. [30]

    Ghost-free high dynamic range imaging with context-aware transformer,

    Z. Liu, Y . Wang, B. Zeng, and S. Liu, “Ghost-free high dynamic range imaging with context-aware transformer,” inEuropean Conference on Computer Vision. Springer, 2022, pp. 344–360

    2022

  31. [31]

    A unified hdr imaging method with pixel and patch level,

    Q. Yan, W. Chen, S. Zhang, Y . Zhu, J. Sun, and Y . Zhang, “A unified hdr imaging method with pixel and patch level,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 22 211–22 220

    2023

  32. [32]

    High dynamic range imaging: Spatially varying pixel exposures,

    S. K. Nayar and T. Mitsunaga, “High dynamic range imaging: Spatially varying pixel exposures,” inProceedings IEEE Conference on Computer Vision and Pattern Recognition. CVPR 2000 (Cat. No. PR00662), vol. 1. IEEE, 2000, pp. 472–479

    2000

  33. [33]

    Reconstructing interlaced high- dynamic-range video using joint learning,

    I. Choi, S.-H. Baek, and M. H. Kim, “Reconstructing interlaced high- dynamic-range video using joint learning,”IEEE Transactions on Image Processing, vol. 26, no. 11, pp. 5353–5366, 2017

    2017

  34. [34]

    High dynamic range video with ghost removal,

    S. Mangiat and J. Gibson, “High dynamic range video with ghost removal,” inApplications of digital image processing XXXIII, vol. 7798. SPIE, 2010, pp. 307–314

    2010

  35. [35]

    Spatially adaptive filtering for registration artifact removal in hdr video,

    ——, “Spatially adaptive filtering for registration artifact removal in hdr video,” in2011 18th IEEE International Conference on Image Processing. IEEE, 2011, pp. 1317–1320

    2011

  36. [36]

    Motion aware exposure bracketing for hdr video,

    Y . Gryaditskaya, T. Pouli, E. Reinhard, K. Myszkowski, and H.-P. Seidel, “Motion aware exposure bracketing for hdr video,” inComputer Graphics F orum, vol. 34, no. 4. Wiley Online Library, 2015, pp. 119– 130

    2015

  37. [37]

    A maximum a posteriori estimation framework for robust high dynamic range video synthesis,

    Y . Li, C. Lee, and V . Monga, “A maximum a posteriori estimation framework for robust high dynamic range video synthesis,”IEEE Transactions on Image Processing, vol. 26, no. 3, pp. 1143–1157, 2016

    2016

  38. [38]

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

    K. C. Chan, S. Zhou, X. Xu, and C. C. Loy, “Basicvsr++: Improving video super-resolution with enhanced propagation and alignment,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 5972–5981

    2022

  39. [39]

    Vision-rwkv: Efficient and scalable visual perception with rwkv-like architectures,

    Y . Duan, W. Wang, Z. Chen, X. Zhu, L. Lu, T. Lu, Y . Qiao, H. Li, J. Dai, and W. Wang, “Vision-rwkv: Efficient and scalable visual perception with rwkv-like architectures,”arXiv preprint arXiv:2403.02308, 2024

    Pith/arXiv arXiv 2024

  40. [40]

    Domainplus: Cross transform domain learning towards high dynamic range imaging,

    B. Zheng, X. Pan, H. Zhang, X. Zhou, G. Slabaugh, C. Yan, and S. Yuan, “Domainplus: Cross transform domain learning towards high dynamic range imaging,” inProceedings of the 30th ACM International Conference on Multimedia, 2022, pp. 1954–1963

    2022

  41. [41]

    Alignment-free hdr deghosting with semantics consistent transformer,

    S. Tel, Z. Wu, Y . Zhang, B. Heyrman, C. Demonceaux, R. Timofte, and D. Ginhac, “Alignment-free hdr deghosting with semantics consistent transformer,” in2023 IEEE/CVF International Conference on Computer Vision (ICCV). IEEE Computer Society, 2023, pp. 12 790–12 799

    2023

  42. [42]

    Afunet: Cross-iterative alignment-fusion synergy for hdr reconstruction via deep unfolding paradigm,

    X. Li, Z. Ni, and W. Yang, “Afunet: Cross-iterative alignment-fusion synergy for hdr reconstruction via deep unfolding paradigm,”arXiv preprint arXiv:2506.23537, 2025

    Pith/arXiv arXiv 2025

  43. [43]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980, 2014

    Pith/arXiv arXiv 2014

  44. [44]

    Video enhance- ment with task-oriented flow,

    T. Xue, B. Chen, J. Wu, D. Wei, and W. T. Freeman, “Video enhance- ment with task-oriented flow,”International Journal of Computer Vision, vol. 127, no. 8, pp. 1106–1125, 2019

    2019

  45. [45]

    Fovvideovdp: A visible difference predictor for wide field-of-view video,

    R. K. Mantiuk, G. Denes, A. Chapiro, A. Kaplanyan, G. Rufo, R. Bachy, T. Lian, and A. Patney, “Fovvideovdp: A visible difference predictor for wide field-of-view video,”ACM Transactions on Graphics (TOG), vol. 40, no. 4, pp. 1–19, 2021

    2021

This paper was first reviewed by grok-4.5 on July 14, 2026.