Pith. sign in

REVIEW 4 major objections 6 minor 47 references

From Synthesis to Removal: Physics-Grounded Reflection Simulation and Diffusion-Based Video Dereflection

T0 review · 4 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read The paper claims video reflection removal can be solved by a closed loop of physics-grounded synthetic pairs, a one-step diffusion removal model, and a dedicated benchmark.

desk verdict Real systems contribution with a serious evaluation gap: the benchmark does not test independent reflection motion, so the SOTA video claim needs softening before acceptance. read the letter →

arxiv 2608.11562 v1 pith:HY5SHR7F submitted 2026-08-12 cs.CV cs.AIeess.IV

Zepeng Wang , Jiagao Hu , Fuhao Li , Yuxuan Chen , Fei Wang , Daiguo Zhou This is my paper · ORCID
classification cs.CVcs.AIeess.IV
keywords videoreflectionremovaldiffusionmodelssynthesisphysics-groundedaugmentationlineartstructurespaceone-stepdenoisingbenchmarktemporalcoherence
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

This paper tries to establish that video reflection removal can be solved end-to-end as a closed loop: a physics-grounded synthesis pipeline generates large-scale, pixel-aligned reflected/clean video pairs; a diffusion-based removal model learns to undo the reflections; and a purpose-built benchmark measures fidelity and human perception. The synthesis side replaces naive RGB blending with compositing of lineart structure conditions through a learned video renderer, under clip-level control of six glass effects: roughness, thickness, reflectance, partial coverage, static sources, and planar geometry. The removal side adapts a pretrained video diffusion model in two stages, first teaching reflection-aware latent adaptation from a residual-derived intensity map and then one-step pixel-geometric refinement, so the final model recovers the clean transmission in a single denoising step at 87.09 ms per frame, faster than non-diffusion baselines. A sympathetic reader would care because the paper targets the two known bottlenecks—missing paired video data and temporal coherence—and claims to remove both.

What carries the argument

The load-bearing object is the structure-space synthesis identity that replaces the linear layer model: $E_F = \operatorname{clip}(E_T + \mathcal{A}(E_R;\theta_g), 0, 1)$ and $I = G(T, E_F)$, where $\mathcal{A}$ is Physics-Grounded Augmentation (PGA)—roughness as Gaussian blur, thickness as a shifted weighted copy, reflectance as affine modulation, plus partial, static, and planar modes—and $G$ is a learned video renderer. This identity turns physical glass parameters into clip-level structural controls, so synthesized pairs stay temporally coherent. On the removal side, the engine is the two-stage objective $\mathcal{L}_{\mathrm{stage1}} = \lambda_{\mathrm{diff}}\mathcal{L}_{\mathrm{diff}} + \lambda_{\mathrm{int}}\mathcal{L}_{\mathrm{int}}$ followed by $\mathcal{L}_{\mathrm{stage2}} = \lambda_{\mathrm{rec}}\mathcal{L}_{\mathrm{rec}} + \lambda_{\mathrm{ssim}}\mathcal{L}_{\mathrm{ssim}} + \lambda_{\mathrm{dep}}\mathcal{L}_{\mathrm{depth}}$, with the emergent one-step denoising update $\hat{z}_0 = D^{(1)}_\theta(z_\tau, \tau, z_I)$ converting latent diffusion into a fast, geometry-anchored restoration.

What would settle it

Take a known clean video, synthesize reflections with S2R-Synthesis at the extremes of the PGA parameter ranges (largest roughness, largest thickness offset, and strongest reflectance), then decode the renderer's output and measure the pixel residual against the clean video in regions the reflection mask leaves untouched: if that residual is not explained by the encoded reflection layer, or if a real-glass test with a known transmission layer shows the one-step model's output drifting from the true transmission outside the training parameter range, the central synthesis-to-removal claim fails.

Watch

Extended reading notes

Core claim

This paper claims that a video diffusion prior can be turned into a practical reflection remover if the training pipeline is closed-loop. S2R-Synthesis generates paired reflected/clean videos by fusing transmission and reflection lineart conditions through a learned renderer, with Physics-Grounded Augmentation controlling roughness, thickness, reflectance, partial coverage, static sources, and planar geometry; S2R-Removal then adapts a pretrained video diffusion backbone in two stages—latent adaptation with residual-derived reflection-intensity supervision, followed by one-step pixel-geometric refinement using reconstruction, SSIM, and depth-consistency losses. The result, the paper argues, is the first diffusion-based video reflection removal model, recovering the clean transmission in a single denoising step, reaching 28.84 dB PSNR on the new S2R-Ref benchmark and 0.787/0.980 human-scored removal/preservation on real-world videos, while running at 87.09 ms per frame.

Load-bearing premise

The whole loop depends on the renderer $G$ adding reflections without changing the clean transmission video: if the lineart fusion or the learned renderer alters $T$ rather than layering a reflection on top, the removal model trains on artifacts and the benchmark gains may not transfer to real glass.

Editorial extensions

If this is right

  • Paired video reflection data stops being a bottleneck: the S2R-Synthesis pipeline can generate large, aligned, and physically varied reflected/clean pairs on demand, including effects that are difficult to capture in reality.
  • Diffusion-based dereflection becomes practical for video: because inference is one step, the model runs at 87.09 ms per frame and is about 1.67 times faster than the next-best baseline despite using a diffusion prior.
  • Standard image benchmarks no longer set the ceiling: the video-trained model also reaches state-of-the-art PSNR and SSIM on public image benchmarks through zoom-based pseudo-video inference.
  • Dereflection pays off downstream: driving-area segmentation and vehicle detection on reflection-contaminated driving videos improve after removal, suggesting preprocessing gains for perception systems.
  • A common evaluation protocol becomes possible: S2R-Bench supplies 60 paired full-reference videos and 50 real-world videos with human scoring, filling the missing benchmark gap for the task.

Reading between the lines

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

  • The PGA parameterization suggests a modular recipe for other degradations: the same structure-space augmentation with a learned renderer could synthesize rain, snow, dirt, or glare overlays, and the two-stage removal objective would transfer wherever the target is a preserved underlying layer.
  • The reflection-intensity head effectively produces a dense reflection-strength map as a byproduct; this could serve as weak supervision for reflection segmentation or as an interpretability signal for where the diffusion model decides to edit.
  • If the emergent one-step capability is a general property of latent diffusion adaptation, then many video restoration tasks currently confined to slow multi-step diffusion inference could adopt the same two-stage latent-adaptation-then-pixel-geometric-refinement schedule.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes a closed-loop framework for video reflection removal. S2R-Synthesis generates paired reflected/clean videos by combining a structure-space physics-grounded augmentation (PGA) of reflection lineart conditions with a Wan2.1-based video diffusion renderer trained on FLUX-generated pseudo reflection videos. S2R-Removal adapts a pretrained video diffusion prior via two-stage training: Stage I performs reflection-aware latent adaptation with an intensity map loss derived from |I-T|, and Stage II applies one-step pixel-geometric refinement with reconstruction, SSIM, and depth-consistency losses. The authors introduce S2R-Bench, comprising S2R-Ref (60 full-reference clips built from DRR image pairs with identical Ken Burns motions applied to both layers) and S2R-Real (50 in-the-wild videos with human evaluation). Experiments on S2R-Bench and image benchmarks report state-of-the-art PSNR/SSIM/TC and human scores, with 87.09 ms/frame inference, faster than non-diffusion baselines.

Significance. If validated, the framework addresses a real bottleneck—the lack of scalable, paired, temporally coherent video dereflection data—and demonstrates that a video diffusion prior can be adapted to a restoration task with efficient one-step inference. The paper's strengths include the physically motivated augmentation primitives with explicit derivations (roughness, thickness, reflectance, partial, static), a self-contained synthesis pipeline, a data ablation (Table 2) showing consistent gains from each PGA component, downstream task improvements in segmentation and detection (Tables 10-11), and the first dedicated video reflection removal benchmark. The internal consistency of the ablations and the clear reporting of training configurations are commendable. However, the full-reference benchmark's construction and several unquantified assumptions limit the strength of the central 'temporally coherent, state-of-the-art video dereflection' claim.

major comments (4)
  1. [Section D.1, Table 1] The S2R-Ref benchmark applies identical Ken Burns transforms (Pan or Zoom) to both the reflected video and the clean ground truth, so the reflection layer never moves independently of the transmission layer and viewpoint-dependent parallax is absent. This does not exercise the defining video-specific phenomenon that motivates the paper in Section 1 ('reflections vary over time, move independently from the transmission layer, and interact with camera motion'). Appendix I explicitly concedes that the synthesis pipeline does not simulate the coupled change between camera motion and reflection geometry. Consequently, the PSNR/SSIM/TC numbers in Table 1 measure a static reflection overlaid on a globally transformed scene, and the claim of 'temporally coherent state-of-the-art video dereflection' is not supported in the regime where reflections move or parallax occurs. Please either construct an additional full-reference or per-subset benchmark with independent reflection motion, or clearly scope the claims to the tested setting.
  2. [Tables 1 and 3] No error bars, confidence intervals, or significance tests are reported for any metric. The S2R-Ref subset contains only 60 clips, the human scores in S2R-Real come from only 15 participants, and several cross-benchmark differences are small (e.g., Nature PSNR 27.89 for Ours vs. 28.08 for GenSIRR). Without variance or per-clip statistics, the state-of-the-art claims on both video and image benchmarks are not yet established. Please report standard deviations across clips or across multiple training runs, and inter-rater agreement for the human evaluation.
  3. [Eq. (6), Section 4.2] The paper's headline efficiency claim rests on the one-step deterministic denoising update in Eq. (6), yet no ablation compares one-step inference against multi-step sampling in Stage II. The text asserts that Stage I instils a 'meaningful one-step denoising capability' and cites observation in [6], but the paper does not show that the one-step mapping is stable or that it does not sacrifice quality relative to more sampling steps. Please provide an ablation over the number of denoising steps at inference, and, if the one-step design is retained, report the quality/efficiency trade-off explicitly.
  4. [Section 3.2, Eq. (2)] The paper claims 'exact paired supervision' for the synthesized pairs (I, T), but this holds only if the learned renderer G preserves the clean video T and adds reflections without otherwise editing it. Since G is trained on FLUX pseudo-reflection videos without an explicit T-preservation constraint, and since the fused condition E_F is clipped in Eq. (14), the renderer could introduce appearance or geometry changes to T that the removal model then learns to reverse. Please quantify how much G modifies T in non-reflection regions (e.g., report a consistency metric between G(T, E_F) and T in areas where the reflection condition is zero), or add a preservation loss during renderer training to substantiate the 'exact paired supervision' claim.
minor comments (6)
  1. [Section 2.2] The sentence 'which no prior benchmark provides' should be qualified to 'no prior benchmark for video reflection removal provides', since image-level reflection removal benchmarks do exist.
  2. [Appendix A] The appendix mixes two numbering schemes: sections are sometimes labeled 'A1', 'A2', etc. and sometimes 'A.1', 'A.2'. Please standardize the section labels.
  3. [Table 2] The table caption does not define 'Trad. [4]' until the text in Section 5.3; please explicitly state that it refers to RGB-space layer blending under Eq. (1).
  4. [Table 9] The zoom-out inference mode consistently outperforms zoom-in, but no explanation is given for this asymmetry. Since the removal model is trained on videos with camera motion, the direction dependence is surprising and deserves a brief discussion.
  5. [Figure 10] The attention-map visualization is informative, but the paper should clarify whether the 'DiT Map w/ head' is the same as the predicted intensity map M-hat or a separate attention response; the terminology is currently ambiguous.
  6. [Section 5.4, Table 3] The ablation only reports results on S2R-Ref. Reporting the same ablation on an image benchmark or on a subset of S2R-Real with dynamic motion would strengthen the claim that the depth-consistency loss specifically benefits video geometry rather than merely improving static-frame metrics.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central removal and benchmark claims are evaluated against captured DRR pairs and external image benchmarks; self-citations are non-load-bearing, and the admitted lack of parallax simulation is a coverage limitation, not a circular reduction.

full rationale

The derivation chain is not circular at any load-bearing point. The physics-grounded augmentation parameters are hand-sampled clip-level values (Section C, Table 5), not fitted to the removal model's outputs, and the learned renderer G is trained on FLUX-generated pseudo reflections rather than on the removal model's predictions. S2R-Removal itself is assessed on S2R-Ref, which Section D.1 states is 'constructed from the DRR dataset [5]' with clean captured ground truth, as well as on the external image benchmarks Real, SIR2, Nature, and OpenRR-1k; these are independent of the S2R-Synthesis data used for training, so the reported PSNR/SSIM gains are not forced by construction. The paper's self-citations are transparent and non-load-bearing: DAI [5] is explicitly excluded from S2R-Ref to avoid data leakage, and the one-step denoising capability is directly ablated in Table 3 rather than being imported solely from the cited observation in [6]. Appendix I does concede an important limitation, namely that the synthesis pipeline 'does not explicitly simulate the coupled change between camera motion and reflection geometry, such as viewpoint-dependent reflection parallax,' and Section D.1 applies identical virtual camera motions to both layers of S2R-Ref, so the full-reference benchmark does not exercise independent reflection motion. That is a genuine limitation on the breadth of the temporal-coherence claim, but it is an evaluation-coverage and correctness-risk issue, not a circular derivation: the method's outputs are still compared against captured and external ground truth rather than against its own fitted inputs. Overall, the paper is self-contained against external benchmarks and does not reduce any prediction to its input by definition.

Assumptions & free parameters 5 free parameters · 7 assumptions · 0 invented entities

The central empirical claims rest on a standard linear layer-formation model and on hand-designed augmentation controls; no new physical entity is introduced. The main debts are the assumed validity of FLUX pseudo-labels, the unproven one-step denoising capability, the assumption that the renderer preserves the clean transmission layer exactly, and benchmark design choices. These are reasonable engineering assumptions for a systems paper, but they are not independently certified.

free parameters (5)
  • PGA augmentation parameter ranges = Not fitted; hand-set ranges (Appendix C, Table 7)
    Hand-set ranges define the diversity of synthesized reflections; they are motivated by optics but not calibrated to measured glass behavior.
  • Stage training loss weights = lambda_diff=1.0, lambda_int=0.1, lambda_rec=1.0, lambda_ssim=0.2, lambda_dep=0.5
    Hand-tuned coefficients in Eqs. (5) and (10); ablations show the loss components matter, but the weights are not swept.
  • Reflection mask threshold tau_m = 12 (pixel intensity units)
    Hand-set threshold in Eq. (9) for binarizing reflection-corrupted regions in the depth consistency loss.
  • Training budgets = 8k renderer steps; 30k steps per removal stage; 38k pairs
    Arbitrary training budgets; no convergence or overfitting analysis is provided.
  • S2R-Ref motion parameters = 10 fps, 81 frames, Ken Burns scales 1.2 to 2.0
    Benchmark construction choices create synthetic camera motion on static DRR pairs; reflection dynamics are not independently modeled.
assumptions (7)
  • domain assumption Linear additive layer formation I = alpha_t T + alpha_r R (Eq. 1)
    Used throughout synthesis and removal; real glass can be nonlinear, dispersive, and multi-interface.
  • domain assumption The learned renderer G preserves the clean transmission T exactly
    Eq. (2) and Section 3.2; pair validity depends on this.
  • domain assumption Lineart representations E_T and E_R capture structure and suppress appearance, and are closed under layer fusion
    Eq. (2) and Section 3.1; no proof that lineart space is equivalent to physical layer composition.
  • ad hoc to paper FLUX pseudo-reflection videos provide valid reflection structure despite temporal flicker
    Section 3.1 and Appendix A.1; pseudo labels inherit FLUX prompt biases.
  • domain assumption Frozen LeReS depth estimates are reliable on both reflected and clean frames
    Section 4.2, Eq. (9); reflection overlays can distort monocular depth estimates.
  • ad hoc to paper One-step deterministic denoising from pure noise is a valid restoration mapping
    Section 4.2, Eq. (6); the emergent one-step capability is asserted rather than derived, and inference noise is unspecified.
  • standard math GGX microfacet, paraxial Snell, and Fresnel formulas used as motivation
    Section 3.3 and Appendix C; standard reference formulas, not numerically fitted.

how reviews work

0 comments
Cite this review

Pith. "Pith review of From Synthesis to Removal: Physics-Grounded Reflection Simulation and Diffusion-Based Video Dereflection." pith.science (2026). https://pith.science/paper/HY5SHR7F

@misc{pith2026260811562,
  author       = {Pith},
  title        = {Pith review of: From Synthesis to Removal: Physics-Grounded Reflection Simulation and Diffusion-Based Video Dereflection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HY5SHR7F}},
  note         = {Machine review of arXiv:2608.11562}
}
read the original abstract

Videos captured through glass often contain reflections that degrade visual quality and interfere with downstream vision tasks. Although single-image reflection removal has been extensively studied, video reflection removal remains largely underexplored due to the lack of paired video data, temporally coherent removal models, and dedicated evaluation benchmarks. We present a closed-loop framework that unifies physics-grounded reflection simulation, diffusion-based video dereflection, and benchmark evaluation. Our S2R-Synthesis pipeline generates paired reflected and reflection-free videos by performing physics-grounded augmentation in the structure space and rendering realistic reflected videos with a trained video diffusion renderer; the augmentation models key glass-related effects including roughness-induced blur, thickness-induced ghosting, and reflectance variation. Based on the synthesized data, we introduce S2R-Removal, the first diffusion-based video reflection removal model, which adapts a pretrained video diffusion prior through reflection-aware latent adaptation and one-step pixel-geometric refinement, recovering the clean transmission in a single denoising step. We further build S2R-Bench, the first benchmark for video reflection removal, supporting both full-reference evaluation and real-world human perceptual assessment. Experiments on S2R-Bench and multiple public image benchmarks demonstrate state-of-the-art performance and faster inference than even non-diffusion baselines, and validate the effectiveness of S2R-Synthesis. Project page: https://codingwzp.github.io/VideoDereflection_S2R.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 36 canonical work pages

  1. [6]

    Rectifying latent space for generative single-image reflection removal,

    M. Li, J. Hu, H. Wang, Q. Hu, J. Wang, and X. Guo, “Rectifying latent space for generative single-image reflection removal,”arXiv preprint arXiv:2512.06358, 2025

    arXiv 2025

  2. [1]

    Separating reflections and lighting using independent compo- nents analysis,

    H. Farid and E. H. Adelson, “Separating reflections and lighting using independent compo- nents analysis,” inProceedings. 1999 IEEE computer society conference on computer vision and pattern recognition (Cat. No PR00149), vol. 1. IEEE, 1999, pp. 262–267

    work page 1999

  3. [2]

    Robust optical flow estimation of double-layer images under transparency or reflection,

    J. Yang, H. Li, Y. Dai, and R. T. Tan, “Robust optical flow estimation of double-layer images under transparency or reflection,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016, pp. 1410–1419

    work page 2016

  4. [3]

    Single image reflection separation with perceptual losses,

    X. Zhang, R. Ng, and Q. Chen, “Single image reflection separation with perceptual losses,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 4786–4794

    work page 2018

  5. [4]

    Single image reflection separation via component synergy,

    Q. Hu and X. Guo, “Single image reflection separation via component synergy,” inProceedings of the IEEE/CVF international conference on computer vision, 2023, pp. 13138–13147

    work page 2023

  6. [5]

    Dereflection any image with diffusion priors and diversified data,

    J. Hu, C. Yang, Z. Zhou, J. Fang, Q. Tian, and W. Shen, “Dereflection any image with diffusion priors and diversified data,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 40, no. 6, 2026, pp. 4860–4868. 11 Physics-Grounded Reflection Simulation and Diffusion-Based Video Dereflection

    work page 2026

  7. [7]

    Video reflection removal through spatio-temporal optimization,

    A. Nandoriya, M. Elgharib, C. Kim, M. Hefeeda, and W. Matusik, “Video reflection removal through spatio-temporal optimization,” inProceedings of the IEEE International Conference on Computer Vision, 2017, pp. 2411–2419

    work page 2017

  8. [8]

    Single image reflection removal be- yond linearity,

    Q. Wen, Y. Tan, J. Qin, W. Liu, G. Han, and S. He, “Single image reflection removal be- yond linearity,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, pp. 3771–3779

    work page 2019

Show all 47 references
  1. [9]

    FLUX.2: Frontier Visual Intelligence,

    B. F. Labs, “FLUX.2: Frontier Visual Intelligence,” https://bfl.ai/blog/flux-2, 2025

    2025

  2. [10]

    Wan: Open and advanced large-scale video generative models,

    T.Wan, A.Wang, B.Ai, B.Wen,C.Mao,C.-W.Xie,D.Chen,F.Yu,H.Zhao, J.Yangetal.,“Wan: Open and advanced large-scale video generative models,”arXiv preprint arXiv:2503.20314, 2025

    2025 arXiv

  3. [11]

    User assisted separation of reflections from a single image using a sparsity prior,

    A. Levin and Y. Weiss, “User assisted separation of reflections from a single image using a sparsity prior,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 29, no. 9, pp. 1647–1654, 2007

    2007

  4. [12]

    Robust separation of reflection from multiple images,

    X. Guo, X. Cao, and Y. Ma, “Robust separation of reflection from multiple images,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2014, pp. 2187–2194

    2014

  5. [13]

    A generic deep architecture for single image reflection removal and image smoothing,

    Q. Fan, J. Yang, G. Hua, B. Chen, and D. Wipf, “A generic deep architecture for single image reflection removal and image smoothing,” inProceedings of the IEEE international conference on computer vision, 2017, pp. 3238–3247

    2017

  6. [14]

    Seeing deeply and bidirectionally: A deep learning approach for single image reflection removal,

    J. Yang, D. Gong, L. Liu, and Q. Shi, “Seeing deeply and bidirectionally: A deep learning approach for single image reflection removal,” inProceedings of the european conference on computer vision (ECCV), 2018, pp. 654–669

    2018

  7. [15]

    Crrn: Multi-scale guided concurrent reflection removal network,

    R. Wan, B. Shi, L.-Y. Duan, A.-H. Tan, and A. C. Kot, “Crrn: Multi-scale guided concurrent reflection removal network,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018, pp. 4777–4785

    2018

  8. [16]

    Polarized reflection removal with perfect alignment in the wild,

    C. Lei, X. Huang, M. Zhang, Q. Yan, W. Sun, and Q. Chen, “Polarized reflection removal with perfect alignment in the wild,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 1750–1758

    2020

  9. [17]

    Location-aware single image reflection removal,

    Z. Dong, K. Xu, Y. Yang, H. Bao, W. Xu, and R. W. Lau, “Location-aware single image reflection removal,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 5017–5026

    2021

  10. [18]

    Reversible decoupling network for single image reflection removal,

    H. Zhao, M. Li, Q. Hu, and X. Guo, “Reversible decoupling network for single image reflection removal,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 26430–26439

    2025

  11. [19]

    Revisiting single image reflection removal in the wild,

    Y. Zhu, X. Fu, P.-T. Jiang, H. Zhang, Q. Sun, J. Chen, Z.-J. Zha, and B. Li, “Revisiting single image reflection removal in the wild,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 25468–25478

    2024

  12. [20]

    User-assisted video reflection removal,

    A. Ahmed, S. Kim, M. Elgharib, and M. Hefeeda, “User-assisted video reflection removal,” in Proceedings of the 12th ACM Multimedia Systems Conference, 2021, pp. 122–131. 12 Physics-Grounded Reflection Simulation and Diffusion-Based Video Dereflection

    2021

  13. [21]

    Single image reflection removal exploiting misaligned training data and network enhancements,

    K. Wei, J. Yang, Y. Fu, D. Wipf, and H. Huang, “Single image reflection removal exploiting misaligned training data and network enhancements,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, pp. 8178–8187

    2019

  14. [22]

    Benchmarking single-image reflectionremovalalgorithms,

    R. Wan, B. Shi, H. Li, Y. Hong, L.-Y. Duan, and A. C. Kot, “Benchmarking single-image reflectionremovalalgorithms,”IEEETransactionsonPatternAnalysisandMachineIntelligence, vol. 45, no. 2, pp. 1424–1441, 2022

    2022

  15. [23]

    Ntire2025challengeonsingleimagereflectionremovalinthewild: Datasets, methods and results,

    K. Yang, J. Cai, L. Ouyang, F.-A. Vasluianu, R. Timofte, J. Ding, H. Sun, L. Fu, J. Li, C. M. Ho etal., “Ntire2025challengeonsingleimagereflectionremovalinthewild: Datasets, methods and results,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp...

    2025

  16. [24]

    Removing reflections from raw photos,

    E. Kee, A. Pikielny, K. Blackburn-Matzen, and M. Levoy, “Removing reflections from raw photos,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 161–171

    2025

  17. [25]

    Reflection removal through efficient adapta- tion of diffusion transformers,

    D. Zakarin, T. Wandel, A. Obukhov, and D. Dai, “Reflection removal through efficient adapta- tion of diffusion transformers,”arXiv preprint arXiv:2512.05000, 2025

    2025

  18. [26]

    Promptrr: Diffusion models as prompt generators for single image reflection removal,

    T. Wang, W. Lu, K. Zhang, T. Lu, and M.-H. Yang, “Promptrr: Diffusion models as prompt generators for single image reflection removal,”arXiv preprint arXiv:2402.02374, 2024

    2024 arXiv

  19. [27]

    Single-image reflection removal via self-supervised diffusion models: Z. lu et al

    Z. Lu, W. Wang, T. Guo, and F. Wang, “Single-image reflection removal via self-supervised diffusion models: Z. lu et al.”The Journal of Supercomputing, vol. 81, no. 1, p. 338, 2025

    2025

  20. [28]

    Fumo: Prior-modulated diffusion for single image reflection removal,

    T. Xu, C. Zhang, G. Zhai, and X. Liu, “Fumo: Prior-modulated diffusion for single image reflection removal,”arXiv preprint arXiv:2603.19036, 2026

    2026 arXiv

  21. [29]

    controlnet-aux: Controlnet auxiliary models,

    HuggingFace, “controlnet-aux: Controlnet auxiliary models,” https://github.com/huggingfa ce/controlnet_aux, 2023

    2023

  22. [30]

    Microfacet models for refraction through rough surfaces

    B. Walter, S. R. Marschner, H. Li, and K. E. Torrance, “Microfacet models for refraction through rough surfaces.”Rendering techniques, vol. 2007, p. 18th, 2007

    2007

  23. [31]

    Real shading in unreal engine 4,

    B. Karis and E. Games, “Real shading in unreal engine 4,”Proc. Physically Based Shading Theory Practice, vol. 4, no. 3, p. 1, 2013

    2013

  24. [32]

    Born and E

    M. Born and E. Wolf,Principles of optics: electromagnetic theory of propagation, interference and diffraction of light. Elsevier, 2013

    2013

  25. [33]

    Learning to recover 3d scene shape from a single image,

    W. Yin, J. Zhang, O. Wang, S. Niklaus, L. Mai, S. Chen, and C. Shen, “Learning to recover 3d scene shape from a single image,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 204–213

    2021

  26. [34]

    Autoawg: Adverse weather generation with adaptive multi-controls for automotive videos,

    J. Hu, D. Zhou, D. Fu, F. Li, Z. Wang, F. Wang, W. Liao, J. Xie, and H. Sun, “Autoawg: Adverse weather generation with adaptive multi-controls for automotive videos,” inProceedings of the 2026 International Conference on Multimedia Retrieval, 2026, pp. 835–844

    2026

  27. [35]

    Image quality metrics: Psnr vs. ssim,

    A. Hore and D. Ziou, “Image quality metrics: Psnr vs. ssim,” in2010 20th international conference on pattern recognition. IEEE, 2010, pp. 2366–2369

    2010

  28. [36]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: from error visibility to structural similarity,”IEEE transactions on image processing, vol. 13, no. 4, pp. 600–612, 2004. 13 Physics-Grounded Reflection Simulation and Diffusion-Based Video De...

    2004

  29. [37]

    Avid: Any-length video inpainting with diffusion model,

    Z. Zhang, B. Wu, X. Wang, Y. Luo, L. Zhang, Y. Zhao, P. Vajda, D. Metaxas, and L. Yu, “Avid: Any-length video inpainting with diffusion model,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2024, pp. 7162–7172

    2024

  30. [38]

    Single image reflection separation via dual-stream interactive transformers,

    Q. Hu, H. Wang, and X. Guo, “Single image reflection separation via dual-stream interactive transformers,”Advances in Neural Information Processing Systems, vol. 37, pp. 55228–55248, 2024

    2024

  31. [39]

    Scaling instruction-based video editing with a high-quality synthetic dataset,

    Q. Bai, Q. Wang, H. Ouyang, Y. Yu, H. Wang, W. Wang, K. L. Cheng, S. Ma, Y. Zeng, Z. Liu et al., “Scaling instruction-based video editing with a high-quality synthetic dataset,”arXiv preprint arXiv:2510.15742, 2025

    2025

  32. [40]

    Qwen3-vl technical report,

    S. Bai, Y. Cai, R. Chen, K. Chen, X. Chen, Z. Cheng, L. Deng, W. Ding, C. Gao, C. Geet al., “Qwen3-vl technical report,”arXiv preprint arXiv:2511.21631, 2025

    2025 arXiv

  33. [41]

    Openrr-5k: A large-scale benchmark for reflection removal in the wild,

    J. Cai, K. Yang, L. Ouyang, L. Fu, J. Ding, J. Shen, and Z. Meng, “Openrr-5k: A large-scale benchmark for reflection removal in the wild,” in2025 IEEE 8th International Conference on Multimedia Information Processing and Retrieval (MIPR). IEEE, 2025, pp. 14–19

    2025

  34. [42]

    Humanvid: Demystifying training data for camera-controllable human image animation,

    Z. Wang, Y. Li, Y. Zeng, Y. Fang, Y. Guo, W. Liu, J. Tan, K. Chen, T. Xue, B. Daiet al., “Humanvid: Demystifying training data for camera-controllable human image animation,” Advances in Neural Information Processing Systems, vol. 37, pp. 20111–20131, 2024

    2024

  35. [43]

    Ultravideo: High-quality uhd video dataset with comprehensive captions,

    Z. Xue, J. Zhang, T. Hu, H. He, Y. Chen, Y. Cai, Y. Wang, C. Wang, Y. Liu, X. Liet al., “Ultravideo: High-quality uhd video dataset with comprehensive captions,”arXiv preprint arXiv:2506.13691, 2025

    2025 arXiv

  36. [44]

    Kling-omni technical report,

    K. Team, J. Chen, Y. Ci, X. Du, Z. Feng, K. Gai, S. Guo, F. Han, J. He, K. Heet al., “Kling-omni technical report,”arXiv preprint arXiv:2512.16776, 2025

    2025 arXiv

  37. [45]

    Yolop: You only look once for panoptic driving perception,

    D. Wu, M.-W. Liao, W.-T. Zhang, X.-G. Wang, X. Bai, W.-Q. Cheng, and W.-Y. Liu, “Yolop: You only look once for panoptic driving perception,”Machine Intelligence Research, vol. 19, no. 6, pp. 550–562, 2022

    2022

  38. [46]

    Ultralytics yolov5,

    G. Jocher, “Ultralytics yolov5,” https://github.com/ultralytics/yolov5, 2020

    2020

  39. [47]

    Bdd100k: A diverse driving dataset for heterogeneous multitask learning,

    F. Yu, H. Chen, X. Wang, W. Xian, Y. Chen, F. Liu, V. Madhavan, and T. Darrell, “Bdd100k: A diverse driving dataset for heterogeneous multitask learning,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2020, pp. 2636–2645. 14 Physics-Groun...

    2020

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.