Pith. sign in

REVIEW 3 major objections 5 minor 54 references

StableMotion: One-Step Motion Estimation with Diffusion Prior

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read StableMotion repurposes Stable Diffusion into a one-step motion estimator that beats task-specific diffusion models with 200x speedup.

desk verdict One-step repurposing of Stable Diffusion for motion estimation is a genuinely new idea, but the reported SOTA quality is undercut by an unexplained 2.85 dB discrepancy between the main table and the loss ablation. read the letter →

arxiv 2505.06668 v2 pith:G5GIOE5X submitted 2025-05-10 cs.CV cs.LGeess.IV

classification cs.CVcs.LGeess.IV
keywords diffusionpriormotionestimationstitchedimagerectanglingrollingshuttercorrectionone-stepinferencesamplingstepsdisasteradaptiveensemblestrategyStablerepurposing
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

StableMotion claims that a pretrained text-to-image diffusion model, Stable Diffusion, can be repurposed as a one-step image-to-motion estimator, and that this beats task-specific diffusion models trained from scratch on two single-image rectification tasks. The authors report state-of-the-art PSNR, SSIM, LPIPS, and FID numbers on the DIR-D stitched-image rectangling benchmark and the RS-Real rolling-shutter benchmark, with roughly 200x faster inference than prior diffusion approaches. The framework maps the input (and, for rectangling, a mask) to a per-pixel flow field that warps the input into the corrected image, using the pretrained VAE as a flow refiner and the adapted UNet as the motion estimator. To justify one-step inference, the paper introduces Sampling Steps Disaster (SSD): with multiple training losses, the model's learned conditional distribution lies between the pseudo-label and ground-truth distributions, so multi-step sampling feeds the model inputs off its training distribution and errors compound. An optional Adaptive Ensemble Strategy (AES) averages several one-step outputs to reduce generative inconsistency and boundary artifacts. If correct, the framework would make large diffusion priors a general, cheap source of geometric supervision for dense prediction tasks.

What carries the argument

The load-bearing object is the predicted flow field $\hat{F}=\theta(C,\epsilon)$ produced by a repurposed Stable Diffusion backbone, with the VAE re-used as a flow refiner (encoding homogeneous normalized flows $f'=\mathrm{cat}(F/\gamma,1)$ and decoding predicted flow latents) and the UNet re-initialized to accept $4(N+1)$ input channels: the latent condition(s) plus the noisy flow latent. The argument also rides on the one-step DDIM sampling recipe, whose justification is the SSD error expansion of Eq. (17): the composite mapping $p=s\circ\theta$ carries the training-time distribution $PD_t$ to $GD_{t-1}$, and any gap $\Delta_t=x_t-y_t$ introduced at one step is multiplied by Jacobians in later steps. AES, the third component, is an adaptive minimum-median filter whose mask is the product of a warped mask $M_{\mathrm{warp}}=W(M,\hat{F}_{0|t})$ and a fixed-edge mask $M_{\mathrm{edge}}$, used to ensemble multiple one-step outputs for the rectangling task.

What would settle it

Train the same two-loss recipe on a third synthetic flow dataset with ground-truth flows, then run DDIM with $T=1,4,16,64,256$ steps and compare both final PSNR and the per-step latent distribution shift $\Delta_t=\mathbb{E}[x_t-y_t]$ (for example, via MMD or FID between the model's intermediate latents and training-time latents). If quality does not degrade with step count, or if the per-step shift does not track the error growth in Eq. (17), SSD as stated is falsified. A simpler disconfirmation would be a task where the same training losses produce better multi-step results, contradicting the claimed universality.

Watch

Extended reading notes

Core claim

The paper's central claim is that the architecture and weights of Stable Diffusion can be adapted, with minimal input-channel surgery, into an image-to-motion model: the UNet takes the concatenation of latent condition features and a noised latent flow, and predicts a denoised flow feature; the VAE decodes it into a normalized homogeneous flow $f'=\operatorname{cat}(F/\gamma,1)$, which after denormalization warps the condition image via $W(I_{\mathrm{cond}},\hat{F})$ to produce the rectified output. Training combines the diffusion reconstruction loss with a pixel-space warp-condition loss and a perceptual loss, which together steer the model toward a conditional distribution intermediate between the pseudo-label and ground-truth distributions. Inference is a single DDIM step from pure noise, a choice justified by the paper's SSD analysis: the first-order Taylor expansion of the error between a multi-step chain and a corrected chain yields an error that sums products of the distribution shifts $\Delta_t$ and Jacobians, growing with step count and vanishing when $T=1$. The empirical backbone is a reported state-of-the-art result on DIR-D and RS-Real, plus an ablation showing the same architecture trained from scratch on one dataset fails to converge, and that an image-to-image variant of SD converges to an inferior PSNR. This combination—pretrained prior, auxiliary-loss training, and one-step sampling—is what the authors say accounts for both the quality and the approximately 200x speedup over prior diffusion methods.

Load-bearing premise

The whole justification for one-step inference rests on the assumption that training with both reconstruction and pixel-space losses puts the learned conditional distribution strictly between the pseudo-label and ground-truth distributions at every noise level, so that multi-step sampling feeds the model inputs it never saw in training; if that in-between property or the first-order Taylor picture fails, one-step may still work empirically but the paper's stated mechanism would collapse.

Editorial extensions

If this is right

  • On DIR-D and RS-Real, a pretrained Stable Diffusion backbone outperforms diffusion models trained from scratch for the same tasks, suggesting that task-specific diffusion training may no longer be the default for geometric correction.
  • One-step DDIM sampling brings per-image inference to about 32 ms on an H100, roughly 200x faster than prior diffusion-based motion estimators even with AES post-processing.
  • The SSD phenomenon implies that when a diffusion model is trained with auxiliary losses, sampling steps should be tuned downward rather than upward; the model's best output can come from a single step.
  • The same weights transfer without fine-tuning to unseen stitched-image and real rolling-shutter data, indicating the pretrained semantic and geometric prior is what generalizes.
  • The framework could in principle be re-targeted to other single-image correction tasks by changing the condition channel count and retraining on their flow datasets.

Reading between the lines

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

  • If SSD is a general property, then many conditional diffusion models trained with perceptual, adversarial, or task losses—not just motion estimators—should also show non-monotonic step-count behavior, so practitioners should sweep step counts rather than assume more steps help.
  • One-step inference plus a random noise draw effectively turns the adapted diffusion model into a near-deterministic flow regressor with stochastic perturbation; ensembling over runs, as AES does, averages out that perturbation and may be essential for stable outputs.
  • The SSD argument's Taylor expansion assumes the composite map $p$ is smooth enough for first-order propagation; if the denoiser is highly non-Lipschitz near the data manifold, multi-step errors could grow faster than the formula suggests, which is a testable prediction about the shape of the error curve.
  • A natural testable extension is to apply the same two-loss recipe to other dense prediction tasks such as optical flow or depth refinement; the paper's ablation predicts that without pretrained priors the architecture will fail to converge on small datasets.
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

3 major / 5 minor

Summary. StableMotion repurposes Stable Diffusion 2.0 into an image-to-motion estimator for two single-image rectification tasks: stitched image rectangling (SIR) and rolling shutter correction (RSC). The VAE is adapted as a flow refiner and the UNet as a motion estimator, with the UNet's first layer replicated and scaled to accept the concatenated condition and flow latents. Training combines diffusion reconstruction loss, pixel-space condition loss, and perceptual loss. Inference uses a one-step DDIM schedule, optionally followed by an Adaptive Ensemble Strategy (AES). The paper reports state-of-the-art PSNR, SSIM, LPIPS, and FID on DIR-D and RS-Real, attributes a substantial speedup (about 200x) to one-step sampling, and introduces Sampling Steps Disaster (SSD) to explain why increasing sampling steps degrades performance. Generalization to unseen OOD data is demonstrated qualitatively and with no-reference evaluations.

Significance. If the empirical results hold, the paper makes a valuable contribution: it shows that a pretrained text-to-image diffusion backbone can be adapted to dense geometric motion estimation with one-step inference, yielding both quality and efficiency gains over task-specific diffusion models trained from scratch. The from-scratch ablation in Table 3 supports the claim that diffusion priors are essential for convergence on a single task dataset, and the one-step inference is validated not only on the proposed model but also on the external MDM model in Table 6. The code/weights release and the reported 32 ms inference time are concrete strengths. However, the central quality claim is not yet statistically secured: the reported margins over baselines are small, no error bars or significance tests are provided, and one baseline LPIPS value in Table 1 is implausible and may invalidate the perceptual SOTA claim on DIR-D.

major comments (3)
  1. [Table 1] The LPIPS value reported for RecDiffusion (0.789) is implausible for a method whose PSNR is 22.21 dB and SSIM is 0.773; every other entry in that table lies in the 0.14-0.15 range. This strongly suggests a different LPIPS variant, a typo, or an inconsistent evaluation protocol. Because the DIR-D perceptual-quality SOTA claim (LPIPS 0.136 vs 0.789) depends on this number, the authors must report the exact LPIPS configuration (backbone network, spatial normalization, preprocessing), re-evaluate RecDiffusion with the same code, or correct the table; otherwise the perceptual SOTA margin may shrink or even invert.
  2. [Section 4.2, Tables 1-2] The reported SOTA margins are small and no error bars, repeated-run variance, seeds, test-set sizes, or statistical significance tests are provided. On RS-Real, StableMotion is actually worse than RS-Diffusion on LPIPS (0.068 vs 0.067) and its SSIM lead is only 0.02; on DIR-D, the PSNR gain over RecDiffusion is 0.85 dB. Without variance estimates, the claim of state-of-the-art quality is not statistically supported. The authors should add bootstrap confidence intervals across test images or multiple training seeds and state the evaluation protocol explicitly.
  3. [Section 3.3, Eq. (17), and Appendix] The SSD argument is not a rigorous derivation. Definition 2 defines p as a mapping from PD_t to GD_{t-1}, but Eq. (15) composes p with itself without specifying the time-indexed domains or showing that composition is well-defined; the Taylor expansion in Eq. (17) assumes differentiability and small perturbations, and the conclusion that the error rises exponentially is not established from the sum in Eq. (25). The key premise that joint training with reconstruction and condition losses yields a distribution intermediate between pseudo-label and ground truth is illustrated in Fig. 4 but never verified. The one-step design is empirically supported by Table 6 and Fig. 1, so the paper's core empirical claim does not collapse, but the SSD explanation should be reframed as a hypothesis or made mathematically precise.
minor comments (5)
  1. [Abstract] The phrase 'a novel framework leverages knowledge' is missing the relative pronoun 'that'; please revise to 'a novel framework that leverages knowledge'.
  2. [Section 1] There is a typo in 'DM-basd methods'; it should read 'DM-based methods'.
  3. [Section 4.6, Table 4] The table caption reads 'Ablaiton on loss items'; this should be 'Ablation on loss items'.
  4. [Section 4.1] The implementation details report 32 ms inference on an NVIDIA H100, but the speedup comparison in the abstract and introduction does not state the hardware used for the baseline methods; please specify the hardware and measurement protocol.
  5. [Section 3.2] The sentence 'For SIR, the conditions C include stitched images alone with its masks' should read 'along with its masks'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: core results are externally benchmarked, and the SSD theory is a post-hoc explanation rather than a fitted input or self-defined prediction.

full rationale

The paper's central claim—repurposing Stable Diffusion priors yields state-of-the-art stitched-image rectangling and rolling-shutter correction at roughly 200x speedup—is validated on external test sets (DIR-D and RS-Real) against both non-self baselines (He et al., Nie et al., CoupledTPS, Yan et al.) and diffusion-based baselines (RecDiffusion, RS-Diffusion). Training uses standard supervised losses (diffusion reconstruction, pixel-space warping condition loss, and VGG perceptual loss) on held-out test data; reporting PSNR/SSIM/LPIPS/FID on the same target is conventional evaluation, not a fitted parameter renamed as a prediction. The one-step inference design is an empirical choice supported by the sampling-step curves in Fig. 1 and Fig. 8, with SSD offered as a post-hoc theoretical account. SSD assumes an intermediate learned distribution and derives error accumulation via the Taylor expansion in Eq. 17; that derivation is heuristic and even appears to have an indexing inconsistency at T=1, but this is a rigor/correctness concern, not a circular reduction, because the one-step result is independently measured and not defined into existence by the equation. The cited prior works from the same group (RecDiffusion/MDM [52], RS-Diffusion [46], DMHomo [22]) are used as benchmarks, as prior art for conditional-loss training, and in Appendix Table 6 as an external observation that another model also degrades with more sampling steps; these are empirical and falsifiable outside the present model's fitted values, so they do not make the argument load-bearing on self-citation. The implausible RecDiffusion LPIPS in Table 1 is a baseline-reporting concern for the skeptic, but it does not indicate that StableMotion's derivation reduces to its own inputs.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a small number of hand-set hyperparameters and several unproven modeling assumptions about how pretrained diffusion features transfer to flow prediction and how the learned conditional distribution behaves. No new physical or external entities are introduced.

free parameters (3)
  • gamma (flow normalization scale) = not reported (estimated as max absolute flow)
    Introduced in Eq. 6 to scale raw flows into the VAE input range; if mis-set, all predicted flows scale incorrectly and warped outputs distort.
  • loss weights (MSE:cond:perceptual) = 1:1:0.01
    Set by hand in Section 4.1; no sensitivity analysis is provided, yet these weights control the balance between diffusion reconstruction and pixel-space flow supervision.
  • learning rate = 2e-5
    Chosen as a standard fine-tuning value; no ablation or schedule details, and the final result may depend on it.
assumptions (5)
  • domain assumption Stable Diffusion's pretrained weights encode geometric and content priors that transfer to dense motion estimation for image rectification.
    Central motivation; partially supported by the from-scratch ablation in Table 3, but the specific prior content that helps is not isolated.
  • ad hoc to paper Replicating the UNet input layer N times and scaling by 1/(N+1) preserves the pretrained behavior while accepting 4(N+1) input channels.
    Section 3.2 introduces this adaptation without an ablation comparing other channel-integration strategies.
  • ad hoc to paper A diffusion model trained with both diffusion reconstruction and pixel-space condition losses learns a conditional distribution that is an interpolation between the pseudo-label and ground-truth distributions.
    Core premise of SSD in Section 3.3, visualized in Figure 4 but not formally proven or measured.
  • ad hoc to paper The first-order Taylor expansion of the composite mapping p in Eq. 17 remains valid over the entire sampling chain.
    Used to claim exponential error growth; no differentiability or Lipschitz conditions are stated, and the expansion is applied iteratively without controlling remainder terms.
  • domain assumption The pseudo labels in DIR-D and RS-Real are accurate enough that flow supervision learns the intended motion.
    The paper notes pseudo-label inaccuracies in Sec. 3.3 but assumes the condition loss can compensate; no error analysis of label noise is given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of StableMotion: One-Step Motion Estimation with Diffusion Prior." pith.science (2026). https://pith.science/paper/G5GIOE5X

@misc{pith2026250506668,
  author       = {Pith},
  title        = {Pith review of: StableMotion: One-Step Motion Estimation with Diffusion Prior},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G5GIOE5X}},
  note         = {Machine review of arXiv:2505.06668}
}
abstract

We present StableMotion, a novel framework that leverages geometric and content priors from pretrained large-scale image diffusion models for motion estimation in single-image rectification tasks such as Stitched Image Rectangling (SIR) and Rolling Shutter Correction (RSC). Specifically, StableMotion takes a text-to-image Stable Diffusion (SD) model as its backbone and repurposes it as an image-to-motion estimator. To mitigate inconsistent outputs produced by diffusion models, we propose Adaptive Ensemble Strategy (AES), which consolidates multiple outputs into a cohesive, high-fidelity result. Additionally, we present Sampling Steps Disaster (SSD), a counterintuitive phenomenon in which increasing the number of sampling steps can lead to poorer outcomes, motivating our one-step inference design. StableMotion is evaluated on two image rectification tasks and delivers state-of-the-art performance on both, while also showing promising transferability through qualitative examples and no-reference evaluations on unseen SIR-OOD and real-captured RSC benchmarks. Supported by SSD, StableMotion achieves efficient one-step inference, offering over 100$\times$ speedup compared to previous diffusion model-based methods even when combined with the optional AES post-processing. Code and weights are available at https://github.com/ivowang/StableMotion.

Figures

Figures reproduced from arXiv: 2505.06668 by the authors.

Figure 1
Figure 1. Two applications of the StableMotion framework, as well as the Sampling Steps Disaster (SSD). [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Repurposing from SD. Taking image rectangling as example. At each timestep, the predicted flow feature is decoded and [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Overview of the inference scheme. Taking image rectangling as example. Note that the sampling process is one-step. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Explanation of Sampling Steps Disaster (SSD). With [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Comparing image rectangling (above the dotted line) and rolling shutter correction (below the dotted line) with previous methods. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Calculation of Maes. 3.4. Adaptive Ensemble Strategy Diffusion models can produce inconsistent results with their generative nature. Besides, tasks like image rectangling in￾volves warping images using a motion field, which makes addressing boundary issues challenging.…
Figure 7
Figure 7. Figure 7: Generalization experiments. StableMotion excels in addressing irregular boundaries and resolving rolling shutter distortions. [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 9
Figure 9. Figure 9: Distortions and boundaries are further repaired by AES. [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: Compare StableMotion to image-to-image framework. [PITH_FULL_IMAGE:figures/full_fig_p008_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 38 canonical work pages

  1. [1]

    Seam carving for content- aware image resizing

    Shai Avidan and Ariel Shamir. Seam carving for content- aware image resizing. InSeminal Graphics Papers: Pushing the Boundaries, Volume 2, pages 609–617. 2023. 3

  2. [2]

    Images that sound: Composing images and sounds on a single canvas

    Ziyang Chen, Daniel Geng, and Andrew Owens. Images that sound: Composing images and sounds on a single canvas. arXiv preprint arXiv:2405.12221, 2024. 2

  3. [3]

    Diffusion models beat gans on image synthesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis. InNeurIPS, pages 8780–8794,

  4. [4]

    Geowiz- ard: Unleashing the diffusion priors for 3d geometry estima- tion from a single image.arXiv preprint arXiv:2403.12013,

    Xiao Fu, Wei Yin, Mu Hu, Kaixuan Wang, Yuexin Ma, Ping Tan, Shaojie Shen, Dahua Lin, and Xiaoxiao Long. Geowiz- ard: Unleashing the diffusion priors for 3d geometry estima- tion from a single image.arXiv preprint arXiv:2403.12013,

  5. [5]

    Factor- ized diffusion: Perceptual illusions by noise decomposition

    Daniel Geng, Inbum Park, and Andrew Owens. Factor- ized diffusion: Perceptual illusions by noise decomposition. arXiv preprint arXiv:2404.11615, 2024. 2

  6. [6]

    Visual ana- grams: Generating multi-view optical illusions with diffu- sion models

    Daniel Geng, Inbum Park, and Andrew Owens. Visual ana- grams: Generating multi-view optical illusions with diffu- sion models. InCVPR, pages 24154–24163, 2024. 2

  7. [7]

    Rectangling panoramic images via warping.ACM TOG, 32(4):1–10,

    Kaiming He, Huiwen Chang, and Jian Sun. Rectangling panoramic images via warping.ACM TOG, 32(4):1–10,

  8. [8]

    Unsupervised semantic correspondence using stable diffu- sion.NeurIPS, 36:8266–8279, 2024

    Eric Hedlin, Gopal Sharma, Shweta Mahajan, Hossam Isack, Abhishek Kar, Andrea Tagliasacchi, and Kwang Moo Yi. Unsupervised semantic correspondence using stable diffu- sion.NeurIPS, 36:8266–8279, 2024. 2

Show all 54 references
  1. [9]

    Classifier-free diffusion guidance

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. InNeurIPS 2021 Workshop on Deep Generative Models and Downstream Applications, 2021. 2, 3

  2. [10]

    Denoising dif- fusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. InNeurIPS, pages 6840–6851,

  3. [11]

    Sim- ple diffusion: End-to-end diffusion for high resolution im- ages

    Emiel Hoogeboom, Jonathan Heek, and Tim Salimans. Sim- ple diffusion: End-to-end diffusion for high resolution im- ages. InICML, pages 13213–13232, 2023. 2

  4. [12]

    LoRA: Low-rank adaptation of large language models

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen- Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. LoRA: Low-rank adaptation of large language models. In ICLR, 2022

  5. [13]

    Low-light image enhancement with wavelet-based diffusion models.ACM TOG, 42(6), 2023

    Hai Jiang, Ao Luo, Haoqiang Fan, Songchen Han, and Shuaicheng Liu. Low-light image enhancement with wavelet-based diffusion models.ACM TOG, 42(6), 2023. 2

  6. [14]

    Towards robust image stitching: An adap- tive resistance learning against compatible attacks

    Zhiying Jiang, Xingyuan Li, Jinyuan Liu, Xin Fan, and Risheng Liu. Towards robust image stitching: An adap- tive resistance learning against compatible attacks. InAAAI, pages 2589–2597, 2024. 1

  7. [15]

    Perceptual losses for real-time style transfer and super-resolution

    Justin Johnson, Alexandre Alahi, and Li Fei-Fei. Perceptual losses for real-time style transfer and super-resolution. In ECCV, pages 694–711. Springer, 2016. 4

  8. [16]

    Elucidating the design space of diffusion-based generative models

    Tero Karras, Miika Aittala, Timo Aila, and Samuli Laine. Elucidating the design space of diffusion-based generative models. InNeurIPS, pages 26565–26577, 2022. 2

  9. [17]

    Repurpos- ing diffusion-based image generators for monocular depth estimation

    Bingxin Ke, Anton Obukhov, Shengyu Huang, Nando Met- zger, Rodrigo Caye Daudt, and Konrad Schindler. Repurpos- ing diffusion-based image generators for monocular depth estimation. InCVPR, pages 9492–9502, 2024. 2

  10. [18]

    Auto-encoding varia- tional bayes

    Diederik P Kingma and Max Welling. Auto-encoding varia- tional bayes. InICLR, 2014. 3

  11. [19]

    A robust method for strong rolling shutter effects correction using lines with automatic feature selection

    Yizhen Lao and Omar Ait-Aider. A robust method for strong rolling shutter effects correction using lines with automatic feature selection. InICCV, pages 4795–4803, 2018. 3

  12. [20]

    Warping residual based image stitching for large parallax

    Kyu-Yul Lee and Jae-Young Sim. Warping residual based image stitching for large parallax. InCVPR, pages 8198– 8206, 2020. 1

  13. [21]

    Your diffusion model is secretly a zero-shot classifier

    Alexander C Li, Mihir Prabhudesai, Shivam Duggal, Ellis Brown, and Deepak Pathak. Your diffusion model is secretly a zero-shot classifier. InICCV, pages 2206–2217, 2023. 2

  14. [22]

    Dmhomo: Learning ho- mography with diffusion models.ACM TOG, 2024

    Haipeng Li, Hai Jiang, Ao Luo, Ping Tan, Haoqiang Fan, Bing Zeng, and Shuaicheng Liu. Dmhomo: Learning ho- mography with diffusion models.ACM TOG, 2024. 2, 5

  15. [23]

    More control for free! im- age synthesis with semantic diffusion guidance

    Xihui Liu, Dong Huk Park, Samaneh Azadi, Gong Zhang, Arman Chopikyan, Yuxiao Hu, Humphrey Shi, Anna Rohrbach, and Trevor Darrell. More control for free! im- age synthesis with semantic diffusion guidance. InWACV, pages 289–299, 2023. 2

  16. [24]

    DPM-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps

    Cheng Lu, Yuhao Zhou, Fan Bao, Jianfei Chen, Chongxuan Li, and Jun Zhu. DPM-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps. InNeurIPS, pages 5775–5787, 2022. 2

  17. [25]

    Diffusion hyperfeatures: Search- ing through time and space for semantic correspondence

    Grace Luo, Lisa Dunlap, Dong Huk Park, Aleksander Holyn- ski, and Trevor Darrell. Diffusion hyperfeatures: Search- ing through time and space for semantic correspondence. NeurIPS, 36:47500–47510, 2024. 2

  18. [26]

    Skipdiff: Adaptive skip diffusion model for high-fidelity perceptual image super-resolution

    Xiaotong Luo, Yuan Xie, Yanyun Qu, and Yun Fu. Skipdiff: Adaptive skip diffusion model for high-fidelity perceptual image super-resolution. InAAAI, pages 4017–4025, 2024. 2

  19. [27]

    Taming diffusion models for im- age restoration: A review.arXiv preprint arXiv:2409.10353,

    Ziwei Luo, Fredrik K Gustafsson, Zheng Zhao, Jens Sj¨olund, and Thomas B Sch ¨on. Taming diffusion models for im- age restoration: A review.arXiv preprint arXiv:2409.10353,

  20. [28]

    Deep rectangling for image stitching: A learning base- line

    Lang Nie, Chunyu Lin, Kang Liao, Shuaicheng Liu, and Yao Zhao. Deep rectangling for image stitching: A learning base- line. InCVPR, pages 5740–5748, 2022. 1, 3, 7, 8, 11

  21. [29]

    Semi-supervised coupled thin-plate spline model for rotation correction and beyond.IEEE TPAMI, pages 1–13,

    Lang Nie, Chunyu Lin, Kang Liao, Shuaicheng Liu, and Yao Zhao. Semi-supervised coupled thin-plate spline model for rotation correction and beyond.IEEE TPAMI, pages 1–13,

  22. [30]

    Barron, and Ben Milden- hall

    Ben Poole, Ajay Jain, Jonathan T. Barron, and Ben Milden- hall. Dreamfusion: Text-to-3d using 2d diffusion. InICLR,

  23. [31]

    Vijay Rengarajan, A. N. Rajagopalan, and R. Aravind. From bows to arrows: Rolling shutter rectification of urban scenes. InCVPR, pages 2773–2781, 2016. 3

  24. [32]

    Rajagopalan

    Vijay Rengarajan, Yogesh Balaji, and A.N. Rajagopalan. Unrolling the shutter: Cnn to correct motion distortions. In CVPR, 2017. 2, 3

  25. [33]

    High-resolution image syn- 9 thesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj¨orn Ommer. High-resolution image syn- 9 thesis with latent diffusion models. InCVPR, pages 10684– 10695, 2022. 2, 3

  26. [34]

    Photorealistic text-to-image diffusion models with deep language understanding.NeurIPS, 35:36479–36494, 2022

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. Photorealistic text-to-image diffusion models with deep language understanding.NeurIPS, 35:36479–36494, 2022. 2

  27. [35]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. InICML, pages 2256– 2265, 2015. 2

  28. [36]

    Denois- ing diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denois- ing diffusion implicit models. InICLR, 2021. 8, 11

  29. [37]

    Generative modeling by es- timating gradients of the data distribution

    Yang Song and Stefano Ermon. Generative modeling by es- timating gradients of the data distribution. InNeurIPS, pages 1–9, 2019. 2

  30. [38]

    Score-based generative modeling through stochastic differential equa- tions

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equa- tions. InICLR, 2021. 2

  31. [39]

    Image alignment and stitching: A tutorial.Foundations and Trends® in Computer Graphics and Vision, 2(1):1–104, 2007

    Richard Szeliski et al. Image alignment and stitching: A tutorial.Foundations and Trends® in Computer Graphics and Vision, 2(1):1–104, 2007. 1

  32. [40]

    Emergent correspondence from image diffusion

    Luming Tang, Menglin Jia, Qianqian Wang, Cheng Perng Phoo, and Bharath Hariharan. Emergent correspondence from image diffusion. InNeurIPS, pages 1363–1389, 2023. 2

  33. [41]

    Diffusion model is secretly a training-free open vocabulary semantic segmenter

    Jinglong Wang, Xiawei Li, Jing Zhang, Qingyuan Xu, Qin Zhou, Qian Yu, Lu Sheng, and Dong Xu. Diffusion model is secretly a training-free open vocabulary semantic segmenter. arXiv preprint arXiv:2309.02773, 2023. 2

  34. [42]

    Zero-shot image restoration using denoising diffusion null-space model

    Yinhuai Wang, Jiwen Yu, and Jian Zhang. Zero-shot image restoration using denoising diffusion null-space model. In ICLR, 2023. 2

  35. [43]

    ReconFusion: 3d reconstruction with diffusion priors

    Rundi Wu, Ben Mildenhall, Philipp Henzler, Keunhong Park, Ruiqi Gao, Daniel Watson, Pratul P Srinivasan, Dor Verbin, Jonathan T Barron, Ben Poole, et al. ReconFusion: 3d reconstruction with diffusion priors. InCVPR, pages 21551–21561, 2024. 2

  36. [44]

    From text to mask: Localizing entities using the attention of text-to-image diffusion models.arXiv preprint arXiv:2309.04109, 2023

    Changming Xiao, Qi Yang, Feng Zhou, and Changshui Zhang. From text to mask: Localizing entities using the attention of text-to-image diffusion models.arXiv preprint arXiv:2309.04109, 2023. 2

  37. [45]

    Tan, Bing Zeng, and Shuaicheng Liu

    Weilong Yan, Robby T. Tan, Bing Zeng, and Shuaicheng Liu. Deep homography mixture for single image rolling shutter correction. InICCV, pages 9868–9877, 2023. 2, 3, 7, 11

  38. [46]

    Single image rolling shutter removal with diffusion models.arXiv preprint arXiv:2407.02906, 2024

    Zhanglei Yang, Haipeng Li, Mingbo Hong, Bing Zeng, and Shuaicheng Liu. Single image rolling shutter removal with diffusion models.arXiv preprint arXiv:2407.02906, 2024. 2, 3, 5, 7, 8, 11

  39. [47]

    Efficient conditional diffu- sion model with probability flow sampling for image super- resolution

    Yutao Yuan and Chun Yuan. Efficient conditional diffu- sion model with probability flow sampling for image super- resolution. InAAAI, pages 6862–6870, 2024. 2

  40. [48]

    As-projective-as-possible image stitching with mov- ing dlt

    Julio Zaragoza, Tat-Jun Chin, Michael S Brown, and David Suter. As-projective-as-possible image stitching with mov- ing dlt. InCVPR, pages 2339–2346, 2013. 1, 7

  41. [49]

    A tale of two features: Stable diffusion complements dino for zero-shot semantic correspondence.NeurIPS, 36: 45533–45547, 2024

    Junyi Zhang, Charles Herrmann, Junhwa Hur, Luisa Pola- nia Cabrera, Varun Jampani, Deqing Sun, and Ming-Hsuan Yang. A tale of two features: Stable diffusion complements dino for zero-shot semantic correspondence.NeurIPS, 36: 45533–45547, 2024. 2

  42. [50]

    Adding conditional control to text-to-image diffusion models

    Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In ICCV, pages 3836–3847, 2023. 2

  43. [51]

    Recstitchnet: Learning to stitch images with rectangular boundaries.Computational Visual Media, 2024

    Yun Zhang, Yukun Lai, Nie Lang, Fang-Lue Zhang, and Lin Xu. Recstitchnet: Learning to stitch images with rectangular boundaries.Computational Visual Media, 2024. 3

  44. [52]

    Recdiffu- sion: Rectangling for image stitching with diffusion models

    Tianhao Zhou, Haipeng Li, Ziyi Wang, Ao Luo, ChenLin Zhang, Jiajun Li, Bing Zeng, and Shuaicheng Liu. Recdiffu- sion: Rectangling for image stitching with diffusion models. InCVPR, pages 1–10, 2023. 1, 2, 3, 4, 5, 7, 11

  45. [53]

    Semi-supervised wide-angle portraits correction by multi-scale transformer

    Fushun Zhu, Shan Zhao, Peng Wang, Hao Wang, Hua Yan, and Shuaicheng Liu. Semi-supervised wide-angle portraits correction by multi-scale transformer. InCVPR, pages 19689–19698, 2022. 1

  46. [54]

    Learning structure- and-motion-aware rolling shutter correction

    Bingbing Zhuang, Quoc-Huy Tran, Pan Ji, Loong-Fah Cheong, and Manmohan Chandraker. Learning structure- and-motion-aware rolling shutter correction. InCVPR, 2019. 2, 3 10 Appendix Sampling Steps Disaster We propose Sampling Steps Disaster (SSD), which refers to the errors that ...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.