Pith. sign in

REVIEW 3 major objections 5 minor 50 references

The paper claims that extremely low-light RAW restoration should treat motion blur as a first-class degradation, and that a one-step latent MeanFlow with a domain-conditioned tokenizer beats prior methods on clean and motion-degraded extrem

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 ↗

An extreme low-light RAW enhancement pipeline combining a domain-conditioned tokenizer, one-step MeanFlow transport, and training-only Retinex loss, evaluated on a new synthetic motion-blur benchmark where it reports 6 to 7 dB gains.

T0 review reviewed 2026-08-04 challenge →

load-bearing objection Worth a serious referee for the SIED results and the MeanFlow adaptation, but the SIDED benchmark—and its 6–7 dB margins—cannot be evaluated until the missing blur recipe appears and the physics are corrected. the 3 major comments →

arxiv 2608.01720 v1 pith:TS54ETO7 submitted 2026-08-03 cs.CV cs.AI

When Extreme Darkness Meets Motion Blur: MeanFlow for Unified RAW Restoration

Zepu Wang , Jingze Liang , Weijie Xiao , Kexin Chen This is my paper
classification cs.CV cs.AI
keywords extremely low-light RAW enhancementmotion deblurringMeanFlowone-step generative modeldomain-adaptive tokenizerRetinex guidanceRAW restorationSIDED dataset
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

Extremely low-light RAW enhancement usually corrects exposure and noise but ignores the camera shake that naturally accompanies handheld dark shots. This paper claims that motion blur should be treated as a first-class degradation and that a one-step latent-flow pipeline can restore both darkness and blur directly from sensor data. To test this, it builds SIDED, a motion-degraded version of the SIED dataset, and reports PSNR gains of 6.06 to 7.42 dB over the strongest compared methods there, along with leading PSNR and SSIM on the clean SIED benchmark. If the claim holds, the framework makes practical handheld extremely-dark RAW restoration feasible in a single network evaluation.

Core claim

The paper's central discovery is that the extreme-dark-to-well-exposed RAW transformation can be carried out in a shared latent space by a one-step MeanFlow, and that the same tokenizer supports a separate blind-latent-deblurring route when motion degradation is present. On the Sony subset of SIED the method reports PSNR 26.03/25.73/25.04 dB at 0.01/0.001/0.0001 lux, the best among compared methods; on SIDED it reports 24.62/24.65/24.23 dB, 6.06–7.42 dB above the strongest competitor. A training-only Retinex-guided refinement stage improves SSIM and SAM at a small PSNR cost, and the paper shows that the motion-aware route remains stable as illumination drops.

What carries the argument

The framework rests on three coupled pieces. A unified RAW tokenizer with domain-adaptive affine modulation (DAAM) maps scaled low-light and well-exposed RAW measurements into one shared latent space, using zero-initialized per-domain scale and shift so pretrained mappings survive. The task-aware MeanFlow group—MeanFlow being a one-step generative model that fits an average velocity field between distributions—transports low-light latents to well-exposed latents in one function evaluation, and optionally runs a blind latent deblurring MeanFlow, initialized on GT-domain motion pairs and cascade-adapted to the enhancement output, without needing a blur kernel at inference. A frozen Retinex net

Load-bearing premise

The load-bearing premise is that SIDED's controlled, channel-wise motion blur applied to noisy Bayer RAW values faithfully mimics real handheld camera shake in extreme darkness; if that realism fails, the reported margins may not transfer to real captures.

What would settle it

Collect real paired sharp/blurry extremely-dark RAW images (e.g., tripod reference plus handheld shots, or IMU-measured shake) and compare the released model against the strongest baselines. A drop from the reported 6–7 dB margin to near parity would falsify the transfer claim; alternatively, re-running SIDED with several realistic measured blur kernels and checking whether the gains persist would test the benchmark's difficulty.

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

If this is right

  • Extreme low-light enhancement can be performed in a single function evaluation instead of iterative diffusion, while still modeling the full dark-to-light distribution.
  • Motion blur and extreme darkness can be handled jointly in the RAW domain, preserving sensor noise statistics rather than pre-ISP information.
  • SIDED provides a controlled benchmark for motion-degraded extreme low-light RAW restoration, enabling direct comparison of future methods in this setting.
  • Because the Retinex guidance is training-only, the improved structure and color fidelity it brings costs no extra inference time.
  • The widening PSNR advantage as illumination falls suggests that exposure-and-noise-only methods are most vulnerable exactly when blur removes scarce structural cues.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Inference: if SIDED's synthetic blur approximates real handheld shake, the same latent route should transfer to real blurry RAW captures; a field test with gyro-measured or tripod-vs-handheld pairs would verify this.
  • Inference: because blur is applied to already-noisy Bayer values, the deblurring network may be exploiting noise correlations specific to the synthetic pipeline, so the 6–7 dB margin could shrink on true noise-motion coupling.
  • Inference: the DAAM domain-calibration idea is not limited to darkness and could be reused for other RAW domain gaps, such as sensor, ISO, or exposure differences, within one tokenizer.
  • Inference: the Stage III PSNR-versus-perception trade-off suggests that for motion-degraded extreme darkness, structural and color-direction metrics may be more informative than pixel error alone.
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 / 5 minor

Summary. The paper proposes SIDED, a synthetic motion-blurred extension of SIED for extremely low-light RAW enhancement, and a framework combining (i) a domain-adaptive RAW tokenizer with dataset-adaptive scaling and Domain-Adaptive Affine Modulation (DAAM), (ii) a one-step 'MeanFlow' enhancement model in latent space, (iii) an optional blind latent deblurring MeanFlow for motion-degraded inputs, and (iv) a training-only Retinex-based refinement. On SIED, the method reports the best PSNR/SSIM at all three illumination levels (26.03/25.73/25.04 dB PSNR). On the proposed SIDED, it reports PSNR gains of 6.06–7.42 dB over compared methods. The core enhancement objective is the MeanFlow velocity target from Geng et al. (2025), applied in the tokenizer's latent space.

Significance. If SIDED's degradation model is physically faithful, the dataset would fill a real gap: existing RAW low-light benchmarks largely ignore accompanying motion blur, and the paper is the first to propose a controlled RAW benchmark for this coupled degradation. The unified tokenizer with DAAM and the one-step latent MeanFlow design are interesting and potentially useful, and the training-only physics guidance is a clean way to add supervision without inference cost. The paper ships a code link, which is a plus. However, the current evidence for the central claim is substantially weakened by the unverifiable construction of SIDED and by the comparison protocol, which gives the proposed method an extra deblurring module unavailable to baselines. The empirical claims therefore need additional support before the conclusions can be accepted.

major comments (3)
  1. [Experimental Setup, SIDED construction; Table 2] The central 6.06–7.42 dB PSNR gains on SIDED are measured on a dataset whose generation is described only as applying 'channel-wise motion degradation' to both Bayer measurements, with 'complete generation details are provided in the supplementary material'; the preprint, however, has no supplementary. As described, blur is applied after the SIED noise is already present. This is not the physical order of real camera shake: blur occurs during exposure, so shot noise is integrated along the motion trajectory, and read noise is added afterward. Convolving an already noisy RAW with a unit-sum kernel smooths the noise and reduces its high-frequency variance, which can make the task artificially easy and may disproportionately favor a learned deblurring module. The kernel family, sizes, spatial variability, and exact pipeline must be specified, and the experiments should be repeated with blur
  2. [Results on SIDED / Compared Methods] All seven baselines (SID, LLPackNet, RRT, MCR, DNF, RAWMamba, SIED) are extreme-low-light RAW enhancement methods without any deblurring component, while the proposed method, on motion-degraded inputs, additionally activates a dedicated Blind Latent Deblurring MeanFlow (Methodology, 'Blind Latent Deblurring MeanFlow'). The reported 6.06–7.42 dB advantage therefore conflates the advantage of having a deblurring module with the quality of the proposed architecture. A fairer assessment would require RAW-domain joint enhancement+deblurring baselines, an 'Ours without deblurring module' variant, or a standard deblurring module cascaded with a strong enhancement baseline. As it stands, the large SIDED margin is not an apples-to-apples comparison.
  3. [Inference] In the motion-degraded route, the output depends on a fresh sample ε∼N(0,I) with no averaging or seed control described (Methodology, 'Inference'). Since the SIDED results in Table 2 are point estimates, the stochasticity of this route may make the reported margins non-reproducible. Please report mean±std over multiple samples or use a fixed seed/averaging scheme, and state the evaluation protocol explicitly.
minor comments (5)
  1. [Abstract / Typos] The abstract contains 'focusonillumination' (missing space). Table 2 lists LLPackNet PSNR as '15.5' at 0.0001 lux while other entries have two decimals; this should be '15.50' for consistency.
  2. [Supplementary references] The paper repeatedly refers to the supplementary material for SIDED construction details and for 'full objectives and settings' of cascade adaptation, but no supplementary is appended to the preprint. This needs to be included in the submission.
  3. [Ablation Studies, Table 3] The standalone tokenizer ablation reports PSNR >47 dB in rendered RGB space. This is far outside the range of Tables 1–2. Please state explicitly that these numbers are not comparable to the full-pipeline results because no enhancement or deblurring is applied.
  4. [Stage III Ablation, Table 4] The paper explains the PSNR drop by arguing that MSE 'intrinsically favors overly smoothed estimates.' This is plausible, but it would be more convincing to add a perceptual metric (e.g., LPIPS) or a small user study to support the claim that the Stage III output is perceptually better despite lower PSNR.
  5. [Figures 2 and 3] The qualitative figures would benefit from zoomed crops with a magnified inset; the current global views make it hard for readers to verify the stated differences in thin structures, fence bars, and seams.

Circularity Check

0 steps flagged

No significant circularity: the pipeline is a standard supervised restoration system; the main weakness is missing SIDED generation details, which is a reproducibility/validity issue rather than a circular derivation.

full rationale

The paper's derivation chain is a conventional supervised RAW-enhancement pipeline. The tokenizer is trained with reconstruction losses (L_tok) against scaled low-light and ground-truth inputs; the Enhancement MeanFlow is trained with a flow-matching objective (L_mf) using paired latent endpoints (z_l, z_g); and the Stage III refinement uses supervised losses (L_ret, L_mse, L_sam) computed against ground-truth decompositions and RGB targets. None of these objectives is defined in terms of the method's own output masquerading as an external prediction. The per-dataset scale s_D is a preprocessing normalization estimated from training patches, not a fitted parameter later reported as a prediction. The MeanFlow formulation is taken from the cited external work (Geng et al. 2025). There are no load-bearing self-citations: SIED is a third-party benchmark, the Retinex network is external (Cai et al. 2023), and the baselines are independent. The SIED evaluation (Table 1) therefore provides genuinely external support. The SIDED half is scientifically weaker: its construction details are deferred to a missing supplementary ('Complete generation details are provided in the supplementary material'), and the benchmark is authored by the same group, with the method's deblurring component trained and evaluated on the same synthetic degradation distribution. That makes the 6–7 dB margins difficult to interpret as evidence of real-world robustness, but it is not a circular derivation: the SIDED losses remain supervised, and testing on the same distribution used for training is standard empirical practice, not a reduction of the result to its inputs. Under the strict standard requiring an exhibited equation-level reduction or a fitted-parameter-renamed-as-prediction, no such step is present.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 1 invented entities

The central claim rests on the MeanFlow self-consistency objective borrowed from Geng et al. 2025, the stability of the DAAM-conditioned latent space, and the validity of a frozen Retinex network and differentiable ISP at extreme darkness. The main free inputs are the SIDED blur statistics, the dataset scale s_D with target intensity tau, and the loss weights, none of which are fully specified in the main text. The only invented artifact is the SIDED dataset itself, which lacks external validation.

free parameters (3)
  • SIDED motion blur kernel distribution = Unspecified (supplementary)
    Kernel size, trajectory, and angle distribution chosen by hand to simulate handheld shake; directly sets benchmark difficulty and the reported 6.06-7.42 dB margins. Construction details are deferred to the absent supplementary material.
  • Dataset-adaptive scale s_D = tau / median_i q99(x_l^(i)) per dataset
    s_D is estimated from the 99th percentile of training low-light RAW patches in Phase A; reversible normalization. tau, the target intensity, is a hand-chosen constant whose value is not stated.
  • Loss weights and training constants (lambda_l, lambda_g, lambda_grad, lambda_ret, lambda_mse, lambda_sam, rho, w, r) = Not reported in main text
    Stage I, II, and III objectives each carry weighting coefficients; DAAM uses a learning-rate ratio rho > 1; mean-flow parameter r and adaptive weight w appear in the Stage II objective. Values are deferred to the supplementary, so a re-implementation must guess them.
axioms (4)
  • domain assumption MeanFlow self-consistent velocity target u_tgt = v - (t - r) u_dot_theta, with the directional derivative computed by Jacobian-vector product, is a valid one-step transport objective.
    Adopted from Geng et al. 2025 and used verbatim in Stage II (Methodology > Stage II). The paper assumes this objective trains a valid one-step generative map in the tokenizer latent space; no derivation or check is given here.
  • standard math Zero-initialized DAAM affine modulation preserves the pretrained mapping at activation.
    F' = F (dot) (1 + gamma_d) + beta_d with gamma_d and beta_d initialized to zero (Methodology > Phase C); true by construction, so low risk.
  • domain assumption The frozen pretrained Retinex network and the frozen differentiable ISP remain valid at extreme darkness down to 0.0001 lux.
    Stage III computes (L_g, R_g) = R(y_g) for restored RGB and compares with GT decomposition (Methodology > Stage III). Cai et al. 2023 is cited, but the stability of Retinex decomposition on extremely weak, noisy inputs is not discussed.
  • domain assumption A shared latent space trained with L1 reconstruction and DAAM preserves the information needed for one-step latent transport between dark and well-exposed domains.
    The pipeline assumes z_l and z_g live in a common space where linear interpolation is a meaningful MeanFlow path. Tokenizer ablations (Table 3) support reconstruction quality but not the sufficiency of the latent for one-step transport.
invented entities (1)
  • SIDED benchmark (motion-degraded SIED quadruplets x_l, x_g, x^m_l, x^m_g) no independent evidence
    purpose: Benchmark for extremely low-light RAW enhancement with accompanying motion blur; the platform for the headline 6-7 dB gains.
    The dataset is constructed by the authors from SIED with a synthetic blur process that is not fully specified in the text. No external validation against real low-light shaky captures is provided, so it cannot be independently checked from the paper.

reviewed 2026-08-04 · how reviews work

0 comments
Cite this review

Pith. "Pith review of When Extreme Darkness Meets Motion Blur: MeanFlow for Unified RAW Restoration." pith.science (2026). https://pith.science/paper/TS54ETO7

@misc{pith2026260801720,
  author       = {Pith},
  title        = {Pith review of: When Extreme Darkness Meets Motion Blur: MeanFlow for Unified RAW Restoration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TS54ETO7}},
  note         = {Machine review of arXiv:2608.01720}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Extremely low-light RAW enhancement aims to recover severely attenuated sensor signals, yet existing methods often focus on illumination and noise while overlooking the motion-induced degradations inherent in practical low-light imaging. We present a framework for robust extremely low-light RAW enhancement under realistic acquisition degradations. First, we introduce See in the Degraded Extremely Dark (SIDED), a new dataset that applies controlled motion degradation to extremely low-light RAW pairs while retaining their original sensor noise. Second, we propose a unified RAW tokenizer equipped with explicit domain-conditioned representation calibration to align extremely low-light and well-exposed RAW data, followed by a MeanFlow that performs enhancement in a single function evaluation. To our knowledge, this is the first work to formulate extremely low-light RAW enhancement under realistic motion-degraded acquisition and address it with MeanFlow. We further introduce a physics-guided refinement model to strengthen illumination--reflectance consistency, pixel fidelity, and color preservation without incurring additional inference cost. Extensive experiments demonstrate that our framework achieves state-of-the-art performance in extremely low-light RAW enhancement, and robustly handles coupled motion and noise degradations.

Figures

Figures reproduced from arXiv: 2608.01720 by Jingze Liang, Kexin Chen, Weijie Xiao, Zepu Wang.

Figure 1
Figure 1. Figure 1: Overview of the proposed framework. A unified RAW tokenizer establishes a shared latent space, in which a task-aware [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Qualitative comparison on the Sony subset of SIED without motion blur at [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparison on the Sony subset of SIDED with motion blur at [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Local Stage III ablation. Stage III recovers the [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

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

50 extracted references · 38 canonical work pages

  1. [1]

    Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education

    Clancey, William J. Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education. Proceedings of the Eighth International Joint Conference on Artificial Intelligence (IJCAI-83)

  2. [2]

    Classification Problem Solving

    Clancey, William J. Classification Problem Solving. Proceedings of the Fourth National Conference on Artificial Intelligence

  3. [3]

    , title =

    Robinson, Arthur L. , title =. 1980 , doi =. https://science.sciencemag.org/content/208/4447/1019.full.pdf , journal =

    1980

  4. [4]

    New Ways to Make Microcircuits Smaller---Duplicate Entry

    Robinson, Arthur L. New Ways to Make Microcircuits Smaller---Duplicate Entry. Science

  5. [5]

    Clancey and Glenn Rennels , abstract =

    Diane Warner Hasling and William J. Clancey and Glenn Rennels , abstract =. Strategic explanations for a diagnostic consultation system , journal =. 1984 , issn =. doi:https://doi.org/10.1016/S0020-7373(84)80003-6 , url =

    doi:10.1016/s0020-7373(84)80003-6 1984

  6. [6]

    and Rennels, Glenn R

    Hasling, Diane Warner and Clancey, William J. and Rennels, Glenn R. and Test, Thomas. Strategic Explanations in Consultation---Duplicate. The International Journal of Man-Machine Studies

  7. [7]

    Poligon: A System for Parallel Problem Solving

    Rice, James. Poligon: A System for Parallel Problem Solving

  8. [8]

    Transfer of Rule-Based Expertise through a Tutorial Dialogue

    Clancey, William J. Transfer of Rule-Based Expertise through a Tutorial Dialogue

  9. [9]

    The Engineering of Qualitative Models

    Clancey, William J. The Engineering of Qualitative Models

  10. [10]

    2023 , eprint=

    Attention Is All You Need , author=. 2023 , eprint=

    2023

  11. [11]

    Pluto: The 'Other' Red Planet

    NASA. Pluto: The 'Other' Red Planet

  12. [12]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =

    Cai, Yuanhao and Bian, Hao and Lin, Jing and Wang, Haoqian and Timofte, Radu and Zhang, Yulun , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =. 2023 , pages =

    work page 2023

  13. [13]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Guo, Chunle and Li, Chongyi and Guo, Jichang and Loy, Chen Change and Hou, Junhui and Kwong, Sam and Cong, Runmin , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    work page

  14. [14]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Xu, Xiaogang and Wang, Ruixing and Fu, Chi-Wing and Jia, Jiaya , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2022 , pages =

    work page 2022

  15. [15]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume =

    Wang, Tao and Zhang, Kaihao and Shen, Tianrun and Luo, Wenhan and Stenger, Bjorn and Lu, Tong , title =. Proceedings of the AAAI Conference on Artificial Intelligence , volume =. 2023 , month =

    work page 2023

  16. [16]

    Proceedings of the 27th ACM International Conference on Multimedia , year =

    Zhang, Yonghua and Zhang, Jiawan and Guo, Xiaojie , title =. Proceedings of the 27th ACM International Conference on Multimedia , year =

    work page

  17. [17]

    Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Chen, Chen and Chen, Qifeng and Xu, Jia and Koltun, Vladlen , title =. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    work page

  18. [18]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Wei, Kaixuan and Fu, Ying and Yang, Jiaolong and Huang, Hua , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    work page

  19. [19]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =

    Jiang, Hai and Guan, Binhao and Liu, Zhen and Liu, Xiaohong and Yu, Jian and Liu, Zheng and Han, Songchen and Liu, Shuaicheng , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =. 2025 , pages =

    work page 2025

  20. [20]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =

    Cho, Sung-Jin and Ji, Seo-Won and Hong, Jun-Pyo and Jung, Seung-Won and Ko, Sung-Jea , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =. 2021 , pages =

    work page 2021

  21. [21]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Zamir, Syed Waqas and Arora, Aditya and Khan, Salman and Hayat, Munawar and Khan, Fahad Shahbaz and Yang, Ming-Hsuan , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2022 , pages =

    work page 2022

  22. [22]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Kong, Lingshun and Dong, Jiangxin and Ge, Jianjun and Li, Mingqiang and Pan, Jinshan , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2023 , pages =

    work page 2023

  23. [23]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Liu, Chengxu and Wang, Xuan and Xu, Xiangyu and Tian, Ruhao and Li, Shuai and Qian, Xueming and Yang, Ming-Hsuan , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2024 , pages =

    work page 2024

  24. [24]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Jin, Xin and Han, Ling-Hao and Li, Zhen and Guo, Chun-Le and Chai, Zhi and Li, Chongyi , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2023 , pages =

    work page 2023

  25. [25]

    IEEE Transactions on Circuits and Systems for Video Technology , volume =

    Chen, Xianmin and Han, Longfei and Huang, Peiliang and Feng, Xiaoxu and Zhang, Dingwen and Han, Junwei , title =. IEEE Transactions on Circuits and Systems for Video Technology , volume =. 2026 , month =. doi:10.1109/TCSVT.2025.3589476 , issn =

    work page arXiv 2026

  26. [26]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year =

    Yan, Qingsen and Feng, Yixu and Zhang, Cheng and Pang, Guansong and Shi, Kangbiao and Wu, Peng and Dong, Wei and Sun, Jinqiu and Zhang, Yanning , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year =

    work page

  27. [27]

    2018 , eprint =

    Wei, Chen and Wang, Wenjing and Yang, Wenhan and Liu, Jiaying , title =. 2018 , eprint =

    work page 2018

  28. [28]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Wu, Wenhui and Weng, Jian and Zhang, Pingping and Wang, Xu and Yang, Wenhan and Jiang, Jianmin , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2022 , pages =

    work page 2022

  29. [29]

    Advances in Neural Information Processing Systems , volume =

    Kawar, Bahjat and Elad, Michael and Ermon, Stefano and Song, Jiaming , title =. Advances in Neural Information Processing Systems , volume =. 2022 , pages =

    work page 2022

  30. [30]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops , month =

    Zhu, Yuanzhi and Zhang, Kai and Liang, Jingyun and Cao, Jiezhang and Wen, Bihan and Timofte, Radu and Van Gool, Luc , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops , month =. 2023 , pages =

    2023

  31. [31]

    Computer Vision -- ECCV 2024 , year =

    Lin, Xinqi and He, Jingwen and Chen, Ziyan and Lyu, Zhaoyang and Dai, Bo and Yu, Fanghua and Qiao, Yu and Ouyang, Wanli and Dong, Chao , title =. Computer Vision -- ECCV 2024 , year =

    work page 2024

  32. [32]

    High-Resolution Image Synthesis With Latent Diffusion Models , booktitle =

    Rombach, Robin and Blattmann, Andreas and Lorenz, Dominik and Esser, Patrick and Ommer, Bj. High-Resolution Image Synthesis With Latent Diffusion Models , booktitle =. 2022 , pages =

    work page 2022

  33. [33]

    Computer Vision -- ECCV 2024 , year =

    Jiang, Yitong and Zhang, Zhaoyang and Xue, Tianfan and Gu, Jinwei , title =. Computer Vision -- ECCV 2024 , year =

    work page 2024

  34. [34]

    2023 , eprint =

    Flow Matching for Generative Modeling , author =. 2023 , eprint =

    work page 2023

  35. [35]

    Advances in Neural Information Processing Systems , volume =

    Geng, Zhengyang and Deng, Mingyang and Bai, Xingjian and Kolter, Zico and He, Kaiming , title =. Advances in Neural Information Processing Systems , volume =. 2025 , pages =

    work page 2025

  36. [36]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Chang, Woong-Gi and You, Tackgeun and Seo, Seonguk and Kwak, Suha and Han, Bohyung , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    work page

  37. [37]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume =

    Perez, Ethan and Strub, Florian and de Vries, Harm and Dumoulin, Vincent and Courville, Aaron , title =. Proceedings of the AAAI Conference on Artificial Intelligence , volume =. 2018 , month =

    work page 2018

  38. [38]

    ACM Transactions on Graphics , volume =

    Jiang, Hai and Luo, Ao and Fan, Haoqiang and Han, Songchen and Liu, Shuaicheng , title =. ACM Transactions on Graphics , volume =. 2023 , month =. doi:10.1145/3618373 , url =

    work page doi:10.1145/3618373 2023

  39. [39]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =

    Yi, Xunpeng and Xu, Han and Zhang, Hao and Tang, Linfeng and Ma, Jiayi , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =. 2023 , pages =

    work page 2023

  40. [40]

    and Wen, Bihan , title =

    Wang, Yufei and Yu, Yi and Yang, Wenhan and Guo, Lanqing and Chau, Lap-Pui and Kot, Alex C. and Wen, Bihan , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , month =. 2023 , pages =

    work page 2023

  41. [41]

    Computer Vision -- ECCV 2022 , year =

    Zhou, Shangchen and Li, Chongyi and Loy, Chen Change , title =. Computer Vision -- ECCV 2022 , year =

    work page 2022

  42. [42]

    Computer Vision -- ECCV 2022 , year =

    Zhao, Yuzhi and Xu, Yongzhe and Yan, Qiong and Yang, Dingdong and Wang, Xuehui and Po, Lai-Man , title =. Computer Vision -- ECCV 2022 , year =

    work page 2022

  43. [43]

    IEEE Transactions on Multimedia , year =

    Li, Yuezhou and Xu, Rui and Niu, Yuzhen and Guo, Wenzhong and Zhao, Tiesong , title =. IEEE Transactions on Multimedia , year =

    work page

  44. [44]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Lv, Xiaoqian and Zhang, Shengping and Wang, Chenyang and Zheng, Yichen and Zhong, Bineng and Li, Chongyi and Nie, Liqiang , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2024 , pages =

    work page 2024

  45. [45]

    , title =

    Liu, Mingyu and Cui, Yuning and Ren, Wenqi and Zhou, Juxiang and Knoll, Alois C. , title =. IEEE Transactions on Circuits and Systems for Video Technology , year =

    work page

  46. [46]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Dong, Xingbo and Xu, Wanyan and Miao, Zhihui and Ma, Lan and Zhang, Chao and Yang, Jiewen and Jin, Zhe and Teoh, Andrew Beng Jin and Shen, Jiajun , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2022 , pages =

    work page 2022

  47. [47]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Lamba, Mohit and Mitra, Kaushik , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2021 , pages =

    work page 2021

  48. [48]

    2020 , eprint =

    Towards Fast and Light-Weight Restoration of Dark Images , author =. 2020 , eprint =

    work page 2020

  49. [49]

    An Adversarial Training Framework Based on Unsupervised Feature Reconstruction Constraints for Crystalline Silicon Solar Cells Anomaly Detection , year =

    Zhu, Ning and Wang, Jing and Zhang, Ying and Wang, Huan and Han, Te , journal =. An Adversarial Training Framework Based on Unsupervised Feature Reconstruction Constraints for Crystalline Silicon Solar Cells Anomaly Detection , year =

    work page

  50. [50]

    High-Order Flow Matching: Unified Framework and Sharp Statistical Rates , url =

    Su, Maojiang and Hu, Jerry Yao-Chieh and Lee, Yi-Chen and Zhu, Ning and Chung, Jui-Hui and Wu, Shang and Song, Zhao and Chen, Minshuo and Liu, Han , booktitle =. High-Order Flow Matching: Unified Framework and Sharp Statistical Rates , url =

    work page

This paper was first reviewed by deepseek-v4-flash on August 4, 2026.