Pith. sign in

REVIEW 5 major objections 7 minor 41 references

A training-free diffusion method injects noise-perturbed conditions and refines the sampling path to improve style transfer, super-resolution, and deblurring without task-specific architectures.

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 ↗

T0 review · grok-4.5

2026-07-11 23:09 UTC pith:SJWVQGZ6

load-bearing objection Solid engineering combo of multi-step condition injection + contrastive margin that beats the usual baselines on three tasks, but the “unified / no task-specific design” claim is overstated because M and the schedules are still hand-tuned per task. the 5 major comments →

arxiv 2607.03899 v1 pith:SJWVQGZ6 submitted 2026-07-04 cs.CV

DICT: Data Injection and Contrastive Trajectory Refinement for Conditional Image Generation with Diffusion Models

Chunnan Shang , Xin Zhang , Zhizhong Wang , Hongwei Wang This is my paper
classification cs.CV
keywords conditional image generationdiffusion modelstraining-free inferencedata injectioncontrastive trajectory refinementstyle transfersuper-resolutiondeblurring
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.

Conditional image generation with diffusion models usually forces a choice: either build specialized modules for each task, or guide sampling with a scalar loss that discards spatial detail and lets errors accumulate. This paper claims a third route is possible. DICT is a training-free inference procedure that treats the condition as data rather than as a loss. It injects a noise-perturbed version of the condition into the early reverse-diffusion steps so the model can distill task-salient structure while keeping high-dimensional cues, then enforces a contrastive objective that makes each successive prediction closer to the condition than the previous one. Across style transfer, 4 imes super-resolution, and Gaussian deblurring the same formulation produces higher fidelity and better perceptual scores than both specialized pipelines and prior loss-guided methods. The practical payoff is a single, architecture-agnostic recipe that improves condition alignment without retraining or redesigning the backbone.

Core claim

The paper establishes that Data Injection of noise-perturbed conditions in early denoising stages, combined with Contrastive Trajectory Refinement between adjacent steps, yields a unified training-free inference method that consistently improves fidelity and perceptual quality over task-specific and loss-guided baselines on style transfer, super-resolution, and deblurring.

What carries the argument

Data Injection and Contrastive Trajectory Refinement (DICT): noise-perturbed condition latents are blended into early reverse steps and denoised under guidance so the model adaptively extracts spatial cues; a pairwise contrastive objective then forces each predicted clean latent to be closer to the condition than its predecessor, acting as a discrete stabilizer against error accumulation.

Load-bearing premise

A simple, hand-chosen preprocessing step plus a few task-specific scalar weights are enough to turn the raw condition into a signal the diffusion prior can distill without systematic bias or content leakage.

What would settle it

On a held-out style-transfer or restoration set, replace the hand-chosen preprocessor M with the identity map (or a deliberately mismatched filter) while keeping all other DICT weights fixed; if fidelity and style/content metrics collapse relative to the reported baselines, the central claim fails.

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

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

5 major / 7 minor

Summary. The paper proposes DICT, a training-free inference procedure for conditional diffusion generation that combines (i) Data Injection—mixing noise-perturbed condition latents into early reverse steps and denoising them under guidance—and (ii) Contrastive Trajectory Refinement—a margin-based objective that encourages successive clean predictions to improve relative to the condition. The authors cast style transfer, super-resolution, and deblurring under a shared posterior/MAP view (Eqs. 5–12), implement DICT on U-Net (SD v1.4 / LDM) and briefly on DiT (PixArt), and report quantitative and qualitative gains over task-specific and loss-guided baselines (Table 2, Fig. 3), with component ablations (Fig. 4).

Significance. If the gains are genuinely attributable to the two proposed mechanisms rather than to per-task condition engineering and schedule tuning, DICT would be a useful practical contribution: a single inference recipe that improves fidelity and perceptual quality across heterogeneous condition types without retraining or architectural redesign. Strengths include a clear algorithmic description (Alg. 1), public code, multi-task evaluation spanning semantic and inverse-problem settings, ablations that show both modules matter, and a DiT transfer check. The work is incremental relative to SDEdit, FreeDoM/TFG-style guidance, and consistency-based inverse solvers, but a well-validated training-free recipe with cross-task transfer would still be of interest to the diffusion community.

major comments (5)
  1. The central “unified / no task-dependent architectures” claim is overstated relative to the method as specified. Sec. 3.3 and Appendix B introduce a task-dependent operator M (identity for style, bicubic upsampling for SR, Wiener deblur for deblurring) and Table 1 / Appendix C prescribe substantially different injection windows T1 (1–18 / 1–8 / 1–50), η1 multi-stage schedules, γ_data, and N_iter. These choices are not derived from the unified posterior (Eqs. 5–12) and are selected by visual/empirical balancing. Without an experiment that freezes M and the schedule family across tasks (or replaces M by a single generic preprocessor), it remains unclear how much of Table 2 is due to Data Injection + Contrastive Trajectory Refinement versus careful condition engineering that ordinary loss-guided baselines also perform via task losses.
  2. The ablations in Fig. 4 remove Data Injection or Contrastive Trajectory Refinement but retain the same M and task-specific schedules. They therefore do not isolate the contribution of the claimed mechanisms from the contribution of the hand-chosen condition preprocessing. A load-bearing control would be: (a) DICT with M = identity (or a shared generic M) on SR/deblurring, and (b) a strong loss-guided baseline given the same M and the same number of gradient steps / iterations. Without these, the attribution of “consistent gains” to DI+CTR is incomplete.
  3. Table 2 reports point estimates with no error bars, seeds, or statistical tests, despite many free parameters (α_data, γ_data, T1, η1/η2, N_iter1/2, α_margin, α_loss) and multi-stage schedules. Several “wins” are narrow or mixed (e.g., style Text Score second to TFG; deblurring PSNR below DCDP*; SR SSIM below FPS-SMC*). Given the amount of per-task tuning documented in Appendix C, the claim of consistent superiority needs either multi-seed reporting or a sensitivity study showing that the ranking is stable under modest hyperparameter perturbation.
  4. Appendix A presents Data Injection as a “non-gradient score proxy” via an empirical Dirac construction and Contrastive Trajectory Refinement as a “discrete Lyapunov stabilizer” / soft manifold projection. These are labeled intuitions, yet the main text (Sec. 3.3) and abstract lean on them to justify why DICT avoids information bottlenecks and error accumulation. There is no formal derivation linking Eq. (15) to a score or likelihood maximizer, nor any measurement of local truncation error or monotonic energy decay under L2. Either demote these claims to pure intuition in the main narrative, or add a minimal validation (e.g., trajectory loss curves, LTE proxy, or a controlled toy inverse problem).
  5. Computational cost is a first-order property of training-free guidance methods, yet the paper only discusses overhead qualitatively in Future Work (E.2). DICT runs N_iter1 and N_iter2 gradient updates per step over long trajectories (50–200 steps), plus dual U-Net evaluations in the injection window (Eq. 15). Without wall-clock or NFE comparisons against TFG, FreeDoM, PSLD, FlowDPS, etc., under matched quality, it is hard to assess whether the reported gains are practical. Please report NFEs / runtime for the settings in Table 1.
minor comments (7)
  1. Notation for the condition target is inconsistent: L1 is written with x in Eq. (17) and with y in Eq. (19); Alg. 1 uses x throughout. Unify symbols for the condition vs. the generated target.
  2. Fig. 1 and Fig. 2 use mixed notation (y, ŷ, c_task, etc.) that is hard to map onto Eqs. (14)–(20). A single symbol table would help.
  3. In Table 2, mark which methods operate in latent vs. pixel space more systematically in the table header (currently only via asterisks in the caption/figure).
  4. Related Work omits several recent training-free / posterior-sampling baselines that are close in spirit (e.g., DPS, ΠGDM, Red-diff, MPGD); a short positioning paragraph would clarify novelty vs. scalar guidance.
  5. Appendix C’s qualitative sweeps for T1, N_iter, η, α_data, γ_data, α_margin are useful but only for style transfer; at least one SR/deblurring sensitivity figure would support the claim that the same design principles transfer.
  6. Typos / wording: “Toaddresstheseissues”, “condition-specificdesigns”, “DataInjectionandContrastiveTrajectoryRefinement” (missing spaces in several places); “the finale of a fireworks display” prompt is fine but figure captions sometimes run together.
  7. Eq. (3) cites PLMS with k=4 historical steps, but later updates rewrite z_t from optimized z_0|t; clarify whether the multistep PLMS history remains valid after these latent edits.

Circularity Check

0 steps flagged

No load-bearing circular derivation; DICT is an empirical training-free procedure whose reported gains are measured on external metrics, not quantities forced by construction from its free parameters.

full rationale

The paper presents an inference-time algorithm (Data Injection of noise-perturbed conditions via Eqs. 13-18 plus Contrastive Trajectory Refinement via the margin loss L2 in Eq. 19) and evaluates it with standard external metrics (Text Score, Style/CLIP Loss, PSNR/SSIM/LPIPS) on held-out image sets. Hyperparameters (M, α_data, γ_data, T1 ranges, η schedules, N_iter) are chosen by hand per task (Table 1, Appendix B/C) and the method therefore re-introduces task-specific engineering, but those choices do not make the tabulated numbers tautological rewritings of the free parameters themselves. There is no self-definitional equation, no fitted scalar re-labeled as a prediction, no uniqueness theorem imported from the authors, and no self-citation that is load-bearing for the central claim. Appendix A supplies only post-hoc intuitions (score-proxy, discrete Lyapunov). The derivation chain is therefore self-contained against external benchmarks; the residual score of 1 reflects only the ordinary presence of author-chosen schedules that are not themselves circular.

Axiom & Free-Parameter Ledger

6 free parameters · 4 axioms · 2 invented entities

The central empirical claim rests on a pretrained diffusion prior, a small collection of hand-chosen scalar hyperparameters that differ by task, lightweight task-specific preprocessing operators M, and the modeling assumption that noise-perturbed condition latents act as a non-gradient score proxy while the contrastive margin acts as a discrete Lyapunov stabilizer. No new physical entities are postulated; the free parameters and domain assumptions listed below are what a re-implementer must accept or retune.

free parameters (6)
  • α_data (latent mixing weight) = 0.88
    Fixed at 0.88 for all three tasks; controls how strongly the noise-perturbed condition is blended into the current latent.
  • γ_data (noise-prediction mixing weight) = 0.2 / 0.1
    Set to 0.2 for style transfer and 0.1 for SR/deblurring; balances condition-injected versus unconditional noise prediction.
  • T1 (Data Injection step window) = task-dependent ranges
    Task-specific early-step ranges (1-18 style, 1-8 SR, 1-50 deblur) chosen by qualitative inspection of content leakage versus under-stylization.
  • η1, η2 (gradient step sizes) and N_iter1/2 = task-specific schedules
    Multi-stage or decaying schedules and iteration counts (typically 3) selected per task to keep updates inside the data manifold; listed in Table 1 and Appendix C.
  • α_margin (contrastive margin) = 1.0
    Fixed at 1.0; enforces a minimum improvement of the loss between adjacent clean predictions.
  • α_loss (VGG vs CLIP style weight) = 0.02
    Set to 0.02 for style transfer to balance mid-level texture and high-level semantic style losses.
axioms (4)
  • domain assumption A pretrained latent diffusion model (SD v1.4 or LDM) already encodes a sufficiently rich prior that can be steered by injected condition latents without further training.
    Invoked throughout Sec. 3; the method never updates network weights.
  • ad hoc to paper Noise-perturbed condition latents preserve task-salient high-dimensional structure while disrupting irrelevant content, acting as a non-gradient score proxy.
    Stated in Sec. 3.3 and elaborated as an informal hypothesis in Appendix A.1.
  • ad hoc to paper Enforcing a positive margin between successive clean-prediction losses yields a discrete Lyapunov-like stabilizer that reduces local truncation error accumulation.
    Claimed in Sec. 3.3 and Appendix A.2; no formal proof is supplied.
  • domain assumption Task-specific lightweight preprocessing M (identity / bicubic / Wiener) is sufficient to place the condition on a manifold compatible with the diffusion prior.
    Defined in Eq. (13) and detailed in Appendix B for each task.
invented entities (2)
  • Data Injection mechanism (noise-perturbed condition latent ĉ_t mixed into early reverse steps) no independent evidence
    purpose: Preserve spatial richness of the condition that scalar losses discard and allow adaptive distillation of task-salient cues.
    Core algorithmic construct of the paper; no independent physical existence outside the proposed sampler.
  • Contrastive Trajectory Refinement objective L2 no independent evidence
    purpose: Force monotonic improvement of condition alignment between adjacent denoising states and thereby limit error accumulation.
    Defined in Eq. (19); functions as a soft manifold-projection operator only inside the paper’s sampling loop.

pith-pipeline@v1.1.0-grok45 · 27728 in / 3305 out tokens · 24161 ms · 2026-07-11T23:09:44.015814+00:00 · methodology

0 comments
read the original abstract

Diffusion models have become a dominant paradigm for conditional image generation, yet existing approaches generally follow two directions: task-specific designs that can improve performance but limit generalization, and training-free loss guidance that compresses rich conditions into scalar objectives and applies stepwise guidance, leading to information bottlenecks and error accumulation along the sampling trajectory. Given the urgent need for an effective unified framework across diverse conditional image generation tasks, we propose Data Injection and Contrastive Trajectory Refinement (DICT), a training-free inference method that enhances conditional image generation without introducing task-dependent architectures. DICT introduces Data Injection, where noise-perturbed conditional signals are integrated into early denoising stages; by performing guided denoising on these injected signals, DICT adaptively selects and distills task-salient information from the raw condition, effectively preserving spatial richness and ensuring precise condition-to-generation alignment. Furthermore, DICT applies Contrastive Trajectory Refinement across adjacent denoising states, enabling pairwise comparisons that progressively improve sample quality. These designs keep inference simple while improving cross-task transfer under a unified diffusion formulation. Extensive experiments on conditional image generation tasks (e.g., style transfer, image super-resolution, and image deblurring) show consistent gains in fidelity and perceptual quality over representative task-specific and loss-guided baselines.

Figures

Figures reproduced from arXiv: 2607.03899 by Chunnan Shang, Hongwei Wang, Xin Zhang, Zhizhong Wang.

Figure 1
Figure 1. Figure 1: Overview of DICT. Data Injection is integrated into the early reverse diffu￾sion stages to retain conditional richness, leveraging guided denoising to adaptively extract task-salient information from the perturbed condition. Simultaneously, Con￾trastive Trajectory Refinement spans the entire trajectory to ensure coherence and refine the generative path for superior quality. Task-specific conditions, where … view at source ↗
Figure 2
Figure 2. Figure 2: The detailed schematics illustrate the Data Injection and Contrastive Trajec￾tory Refinement within the DICT framework. common in fixed-initialization methods. (iii) Adaptive Selectivity. SDEdit lacks a mechanism to distinguish between relevant and irrelevant information—it at￾tempts to reconstruct the entire noisy input. DICT, however, leverages the noise prediction, i.e., ϵti = ϵθ(t) in Eq. 15 as a natur… view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparisons. An asterisk (∗) denotes models in the pixel space; those unmarked operate in the latent space. causing tangential drift. Our L2 acts as a discrete Lyapunov constraint enforc￾ing monotonic energy decay (∆L1 ≤ −αmargin). Functioning as a soft Manifold Projection Operator, it generates a normal force to mitigate orthogonal LTE, promoting coherence. Details are in Sec. A of the Supplem… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative and quantitative ablation results. “w/o DI” and “w/o CT” denote without Data Injection and Contrastive Trajectory Refinement, respectively. improvement in Text Score, it results in a substantial surge in Style Loss and CLIP Loss. This suggests that while the model may follow the text prompt slightly more easily without the Data Injection, it fails to capture the essential style, leading to a si… view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative DiT-based stylization results. suffer from inaccurate local details, particularly around the teeth region; for de￾blurring, the recovery of fine textures is significantly constrained, as seen in the blurred eye regions in the second row of [PITH_FULL_IMAGE:figures/full_fig_p015_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative results showing the effect of parameter T1 on the style transfer task. For style transfer, an optimal T1 (e.g., T1 = 18 in [PITH_FULL_IMAGE:figures/full_fig_p021_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative results of style transfer with varying parameter Niter1 and Niter2. Niter2: The parameter represents the number of iterative updates applied to the predicted data for the next timestep during the reverse diffusion process. The selection of Niter2 depends on the effectiveness of the updates and whether the updated data distribution remains within the model’s valid range. The pri￾mary goal of thi… view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative results of style transfer with varying parameter η1 and η2. η1: The parameter controls the speed of iterative updates to the predicted data at the current timestep during the reverse diffusion process. The magnitude of the learning rate depends on the required optimization granularity and the stability of the optimization for a specific task. For instance, in style transfer, since we are concer… view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative results showing the effect of parameter αdata for style transfer. Text Style γdata = 0 γdata = 0.2 γdata = 0.4 γdata = 1 “a photo of a phone from the 20s” [PITH_FULL_IMAGE:figures/full_fig_p024_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative results showing the effect of parameter γdata for style transfer. theoretical insights and empirical observations, we utilize task-specific schedules for η1 to ensure both stylistic expression and structural fidelity. η2: The parameter controls the speed of iterative updates for the predicted data at the next timestep during the reverse diffusion process. The learning rate is determined by the… view at source ↗
Figure 11
Figure 11. Figure 11: Qualitative results showing the effect of parameter αmargin for style transfer. data. Specifically, for style transfer tasks, our focus is on the style of the image rather than its semantic content. Consequently, we increase αdata to suppress the influence of the reference image’s content, thereby mitigating the risk of content leakage. In contrast, for super-resolution and deblurring, the choice depends … view at source ↗
Figure 12
Figure 12. Figure 12: Qualitative DiT-based stylization results. γdata = 0.2 to be the optimal value, striking a balance between achieving the desired degree of stylization and maintaining the integrity of the original content. αmargin: This parameter controls the maximum permissible reduction in the loss between the predicted data and the conditional image at the current timestep relative to the previous one. The selection of… view at source ↗
Figure 13
Figure 13. Figure 13: Comprehensive comparisons between DICT, the base model (w/o DICT), and SDEdit. Both visual visualizations and numerical metrics are provided for (a) style transfer, (b) super-resolution, and (c) deblurring. in [PITH_FULL_IMAGE:figures/full_fig_p027_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Qualitative results on the style transfer task. For each group, the top row shows the content and style references, and the bottom row is the resulting stylized output [PITH_FULL_IMAGE:figures/full_fig_p030_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Qualitative results on the style transfer task. For each group, the top row shows the content and style references, and the bottom row is the resulting stylized output [PITH_FULL_IMAGE:figures/full_fig_p031_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Qualitative results on the image super-resolution task [PITH_FULL_IMAGE:figures/full_fig_p032_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Qualitative results on the image deblurring task [PITH_FULL_IMAGE:figures/full_fig_p033_17.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

41 extracted references · 13 linked inside Pith

  1. [1]

    arXiv preprint arXiv:2410.04479 (2025)

    Alkhouri, I., Liang, S., Huang, C.H., Dai, J., Qu, Q., Ravishankar, S., Wang, R.: Sitcom: Step-wise triple-consistent diffusion sampling for inverse problems. arXiv preprint arXiv:2410.04479 (2025)

    Pith/arXiv arXiv 2025

  2. [2]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Castillo, A., Kohler, J., Pérez, J.C., Pérez, J.P., Pumarola, A., Ghanem, B., Ar- beláez, P., Thabet, A.: Adaptive guidance: Training-free acceleration of conditional diffusion models. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 39, pp. 1962–1970 (2025)

    1962

  3. [3]

    In: International conference on learning representations

    Chen, J., Yu, J., Ge, C., Yao, L., Xie, E., Wang, Z., Kwok, J., Luo, P., Lu, H., Li, Z.: PixArt-α: Fast training of diffusion transformer for photorealistic text-to-image synthesis. In: International conference on learning representations. vol. 2024, pp. 57611–57640 (2024)

    2024

  4. [4]

    arXiv preprint arXiv:2209.14687 (2022)

    Chung, H., Kim, J., Mccann, M.T., Klasky, M.L., Ye, J.C.: Diffusion posterior sam- pling for general noisy inverse problems. arXiv preprint arXiv:2209.14687 (2022)

    Pith/arXiv arXiv 2022

  5. [5]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Deng, Y., He, X., Tang, F., Dong, W.: Z*: Zero-shot style transfer via attention reweighting. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 6934–6944 (2024)

    2024

  6. [6]

    In: The Twelfth International Conference on Learning Representations (2024)

    Dou, Z., Song, Y.: Diffusion posterior sampling for linear inverse problem solv- ing: A filtering perspective. In: The Twelfth International Conference on Learning Representations (2024)

    2024

  7. [7]

    IEEE Transactions on Pattern Analysis and Machine Intelligence (2025)

    Gao, J., Sun, Y., Liu, Y., Tang, Y., Zeng, Y., Qi, D., Chen, K., Zhao, C.: Styleshot: A snapshot on any style. IEEE Transactions on Pattern Analysis and Machine Intelligence (2025)

    2025

  8. [8]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Gatys, L.A., Ecker, A.S., Bethge, M.: Image style transfer using convolutional neural networks. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 2414–2423 (2016)

    2016

  9. [9]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Hertz, A., Voynov, A., Fruchter, S., Cohen-Or, D.: Style aligned image generation via shared attention. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 4775–4785 (2024)

    2024

  10. [10]

    Advances in neural information processing systems33, 6840–6851 (2020)

    Ho, J., Jain, A., Abbeel, P.: Denoising diffusion probabilistic models. Advances in neural information processing systems33, 6840–6851 (2020)

    2020

  11. [11]

    In: Proceedings of the IEEE international conference on computer vision

    Huang, X., Belongie, S.: Arbitrary style transfer in real-time with adaptive instance normalization. In: Proceedings of the IEEE international conference on computer vision. pp. 1501–1510 (2017)

    2017

  12. [12]

    2017 IEEE International Conference on Computer Vision (ICCV) pp

    Huang, X., Belongie, S.J.: Arbitrary style transfer in real-time with adaptive in- stance normalization. 2017 IEEE International Conference on Computer Vision (ICCV) pp. 1510–1519 (2017),https://api.semanticscholar.org/CorpusID: 6576859

    2017

  13. [13]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Karras, T., Laine, S., Aila, T.: A style-based generator architecture for generative adversarial networks. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 4401–4410 (2019)

    2019

  14. [14]

    arXiv preprint arXiv:2503.08136 (2025)

    Kim, J., Kim, B.S., Ye, J.C.: Flowdps: Flow-driven posterior sampling for inverse problems. arXiv preprint arXiv:2503.08136 (2025)

    Pith/arXiv arXiv 2025

  15. [15]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Lei, M., Song, X., Zhu, B., Wang, H., Zhang, C.: Stylestudio: Text-driven style transfer with selective control of style elements. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 23443–23452 (2025)

    2025

  16. [16]

    Li,H.,Pereira,M.:Solvinginverseproblemsviadiffusionoptimalcontrol.Advances in Neural Information Processing Systems37, 73549–73571 (2024) DICT 35

    2024

  17. [17]

    arXiv preprint arXiv:2403.06054 (2024)

    Li, X., Kwon, S.M., Liang, S., Alkhouri, I.R., Ravishankar, S., Qu, Q.: Decoupled data consistency with diffusion purification for image restoration. arXiv preprint arXiv:2403.06054 (2024)

    Pith/arXiv arXiv 2024

  18. [18]

    ACM Transactions on Graphics (TOG)43(6), 1–10 (2024)

    Liu, G., Xia, M., Zhang, Y., Chen, H., Xing, J., Wang, Y., Wang, X., Shan, Y., Yang, Y.: Stylecrafter: Taming artistic video diffusion with reference-augmented adapter learning. ACM Transactions on Graphics (TOG)43(6), 1–10 (2024)

    2024

  19. [19]

    arXiv preprint arXiv:2202.09778 (2022)

    Liu, L., Ren, Y., Lin, Z., Zhao, Z.: Pseudo numerical methods for diffusion models on manifolds. arXiv preprint arXiv:2202.09778 (2022)

    Pith/arXiv arXiv 2022

  20. [20]

    arXiv preprint arXiv:1711.05101 (2017)

    Loshchilov, I., Hutter, F.: Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101 (2017)

    Pith/arXiv arXiv 2017

  21. [21]

    arXiv preprint arXiv:2108.01073 (2021)

    Meng, C., He, Y., Song, Y., Song, J., Wu, J., Zhu, J.Y., Ermon, S.: Sdedit: Guided image synthesis and editing with stochastic differential equations. arXiv preprint arXiv:2108.01073 (2021)

    Pith/arXiv arXiv 2021

  22. [22]

    Paszke, A., Gross, S., Chintala, S., Chanan, G., Yang, E., DeVito, Z., Lin, Z., Desmaison, A., Antiga, L., Lerer, A.: Automatic differentiation in pytorch (2017)

    2017

  23. [23]

    arXiv preprint arXiv:2412.00100 (2024)

    Patel, M., Wen, S., Metaxas, D.N., Yang, Y.: Steering rectified flow models in the vector field for controlled image generation. arXiv preprint arXiv:2412.00100 (2024)

    Pith/arXiv arXiv 2024

  24. [24]

    Peebles,W.,Xie,S.:Scalablediffusionmodelswithtransformers.In:Proceedingsof the IEEE/CVF international conference on computer vision. pp. 4195–4205 (2023)

    2023

  25. [25]

    Issues in Accounting Education26(3), 593–608 (2011)

    Phillips, F., Mackintosh, B.: Wiki art gallery, inc.: A case for critical thinking. Issues in Accounting Education26(3), 593–608 (2011)

    2011

  26. [26]

    In: Proceed- ings of the IEEE/CVF conference on computer vision and pattern recognition

    Qi, T., Fang, S., Wu, Y., Xie, H., Liu, J., Chen, L., He, Q., Zhang, Y.: Deadiff: An efficient stylization diffusion model with disentangled representations. In: Proceed- ings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 8693–8702 (2024)

    2024

  27. [27]

    In: International conference on machine learning

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al.: Learning transferable visual models from natural language supervision. In: International conference on machine learning. pp. 8748–8763. PmLR (2021)

    2021

  28. [28]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 10684–10695 (2022)

    2022

  29. [29]

    Advances in Neural Information Processing Systems36, 49960–49990 (2023)

    Rout, L., Raoof, N., Daras, G., Caramanis, C., Dimakis, A., Shakkottai, S.: Solving linear inverse problems provably via posterior sampling with latent diffusion mod- els. Advances in Neural Information Processing Systems36, 49960–49990 (2023)

    2023

  30. [30]

    arXiv preprint arXiv:1409.1556 (2014)

    Simonyan, K., Zisserman, A.: Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556 (2014)

    Pith/arXiv arXiv 2014

  31. [31]

    arXiv preprint arXiv:2306.00983 (2023)

    Sohn, K., Ruiz, N., Lee, K., Chin, D.C., Blok, I., Chang, H., Barber, J., Jiang, L., Entis, G., Li, Y., et al.: Styledrop: Text-to-image generation in any style. arXiv preprint arXiv:2306.00983 (2023)

    Pith/arXiv arXiv 2023

  32. [32]

    arXiv preprint arXiv:2404.02733 (2024)

    Wang, H., Spinelli, M., Wang, Q., Bai, X., Qin, Z., Chen, A.: Instantstyle: Free lunch towards style-preserving in text-to-image generation. arXiv preprint arXiv:2404.02733 (2024)

    Pith/arXiv arXiv 2024

  33. [33]

    The MIT press (1964)

    Wiener, N.: Extrapolation, interpolation, and smoothing of stationary time series. The MIT press (1964)

    1964

  34. [34]

    In: Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition

    Wu, R., Yang, T., Sun, L., Zhang, Z., Li, S., Zhang, L.: Seesr: Towards semantics- aware real-world image super-resolution. In: Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition. pp. 25456–25467 (2024) 36 C. Shang et al

    2024

  35. [35]

    arXiv preprint arXiv:2408.16766 (2024)

    Xing, P., Wang, H., Sun, Y., Wang, Q., Bai, X., Ai, H., Huang, R., Li, Z.: Csgo: Content-style composition in text-to-image generation. arXiv preprint arXiv:2408.16766 (2024)

    Pith/arXiv arXiv 2024

  36. [36]

    arXiv preprint arXiv:2501.18913 (2025)

    Xu, T., Cai, X., Zhang, X., Ge, X., He, D., Sun, M., Liu, J., Zhang, Y.Q., Li, J., Wang, Y.: Rethinking diffusion posterior sampling: From conditional score estima- tor to maximizing a posterior. arXiv preprint arXiv:2501.18913 (2025)

    Pith/arXiv arXiv 2025

  37. [37]

    Advances in Neural Information Processing Systems37, 22370–22417 (2024)

    Ye, H., Lin, H., Han, J., Xu, M., Liu, S., Liang, Y., Ma, J., Zou, J.Y., Ermon, S.: Tfg: Unified training-free guidance for diffusion models. Advances in Neural Information Processing Systems37, 22370–22417 (2024)

    2024

  38. [38]

    Yu, J., Wang, Y., Zhao, C., Ghanem, B., Zhang, J.: Freedom: Training-free energy- guidedconditionaldiffusionmodel.In:ProceedingsoftheIEEE/CVFInternational Conference on Computer Vision. pp. 23174–23184 (2023)

    2023

  39. [39]

    In: Proceedings of the Computer Vision and Pattern Recognition Con- ference

    Yue, Z., Liao, K., Loy, C.C.: Arbitrary-steps image super-resolution via diffusion inversion. In: Proceedings of the Computer Vision and Pattern Recognition Con- ference. pp. 23153–23163 (2025)

    2025

  40. [40]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Zhang, B., Chu, W., Berner, J., Meng, C., Anandkumar, A., Song, Y.: Improving diffusion inverse problem solving with decoupled noise annealing. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 20895–20905 (2025)

    2025

  41. [41]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Zhang, L., Rao, A., Agrawala, M.: Adding conditional control to text-to-image diffusion models. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 3836–3847 (2023)

    2023