Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

ASGDiffusion: Parallel High-Resolution Generation with Asynchronous Structure Guidance

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Training-free diffusion generates high-res images in parallel with stale guidance.

desk verdict Missing sync-vs-async ablation undercuts an otherwise credible speedup claim. read the letter →

arxiv 2412.06163 v1 pith:KQPEOCTX submitted 2024-12-09 cs.CV

Yuming Li , Peidong Jia , Daiwei Hong , Yueru Jia , Qi She , Rui Zhao , Ming Lu , Shanghang Zhang This is my paper · ORCID
classification cs.CV
keywords training-freehigh-resolutiongenerationdiffusionmodelsstructureguidancecross-attentionmaskasynchronousmulti-GPUparallelpatternrepetitiontext-to-image
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 claims that a pretrained diffusion model can generate images beyond its native resolution—without any fine-tuning—if the denoising of each patch is steered by a global structure signal: the low-resolution noise of a reference patch, weighted by a cross-attention mask. That structure signal removes the repetitive-pattern artifacts that plague existing training-free upsamplers. The paper further claims that this guidance can be made asynchronous, using the previous timestep's guidance for the current step, so that patches no longer wait for a global broadcast; each GPU denoises its own patch and the communication cost hides under computation. With this, the method runs multi-GPU parallel generation, cutting 2048x2048 inference time to 14 seconds on four GPUs (13.4x faster than DemoFusion) while using less memory per GPU. If true, this makes interactive high-resolution text-to-image generation practical without retraining.

What carries the argument

The mechanism that carries the argument is the asynchronous structure guidance (ASG) identity in Eq. (5)–(6): guidance from the previous timestep $G_t = w_t M[\epsilon(x_{t-1}^{(0)}, t-1) - \epsilon(x_t^{(i)}, t)]$ is added to each patch's noise prediction. The cross-attention mask $M$ is built from the upsampling-layer attention heatmaps, which focus on objects; it modulates how strongly the reference patch's prediction pulls other patches toward the global structure, preserving background clarity. The two-stage pipeline uses this guidance only in the first (global-structuring) stage, then refines details in the second stage, and the LR reference patch noise is selected as the first patch so it acts as the shared global-structure carrier.

What would settle it

Run the same prompts and resolution with synchronous structure guidance (waiting for $G_t$) against asynchronous guidance ($G_{t-1}$) and compare FID/CLIP and per-image visual artifacts; if the async variant shows measurable degradation—especially at high guidance scales or fewer total steps—the central speed-quality tradeoff claim fails. Alternatively, test at 4096x4096 where patch counts grow and guidance drift should be largest.

Watch

Extended reading notes

Core claim

The central discovery is that replacing synchronous structure guidance with guidance from the previous timestep does not noticeably hurt image quality while enabling parallel patch denoising. Formally, the corrected noise for patch $i$ is $\tilde{\epsilon}(x_t^{(i)}, t) = \epsilon(x_t^{(i)}, t) + G_t$ with $G_t = w_t\, M\,[\epsilon(x_{t-1}^{(0)}, t-1) - \epsilon(x_t^{(i)}, t)]$, where $M$ is the normalized cross-attention mask from the upsampling layers, and $x^{(0)}$ is the reference patch whose LR noise encodes global structure. Because consecutive diffusion steps are similar, the stale guidance $G_t$ computed from step $t-1$ is a sufficient substitute for the true $t$-step guidance; the paper states experiments showed minimal quality differences between synchronous and asynchronous variants (Sec. 5.5). This lets every patch denoise immediately, overlapping communication with computation.

Load-bearing premise

The asynchronous method assumes that guidance computed at the previous timestep is a good enough substitute for the current timestep's guidance, so that denoising without waiting preserves output quality; the paper supports this only with a qualitative statement and no quantitative sync-versus-async comparison.

Editorial extensions

If this is right

  • On SDXL, ASGDiffusion produces 2048x2048 and 3072x3072 images in 14s and 59.4s on 4 GPUs, 13.4x and 11.2x faster than DemoFusion at the same resolutions.
  • Per-GPU memory drops because no GPU denoises the full HR latent; each GPU only denoises its LR-sized patch, with the structure guidance broadcast asynchronously.
  • The method transfers across Stable Diffusion versions (1.5, 2.1, XL, 3) without retraining, as demonstrated on 4x upsampling.
  • At 1024x2048, the method achieves the best FID (64.27) and IS (15.98) among compared training-free methods, and the human study ranks it highest (1.68 vs 1.97/2.11).
  • The paper acknowledges that at 3072x3072 DemoFusion achieves better FID/IS, and that 4096x4096 still shows small-object repetition and partial body blur.

Reading between the lines

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

  • If the stale-guidance approximation holds across step counts and corruption schedules, the same async pattern could be applied to other guidance signals (e.g., classifier-free guidance text conditioning) to parallelize other diffusion workloads.
  • The cross-attention mask's role suggests that using segmentation or saliency maps as a mask, or expanding the mask to cover entire objects, could reduce the remaining body blur the paper reports.
  • A quantitative sync-vs-async ablation with FID/CLIP at multiple resolutions and guidance scales would be the natural stress test; the paper currently relies on a qualitative statement.
  • The method's dependence on the reference patch's LR noise as the global structure carrier implies that the first patch's quality bounds the whole image; generating that patch with higher fidelity or ensembling multiple references could improve ultra-high-resolution outputs.
Share X Bluesky LinkedIn Reddit HN

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. ASGDiffusion is a training-free method for generating high-resolution images from pretrained Stable Diffusion models. The method operates in two patch-based stages: a global-structure stage in which the noise prediction of a designated first patch is used as structure guidance for all other patches, weighted by a cross-attention mask, and a detail-refinement stage. To allow multi-GPU parallelism, the paper replaces synchronous structure guidance with asynchronous guidance computed at the previous timestep (Eq. 5). The paper reports inference-time comparisons (Table 1), quantitative metrics FID/IS/CLIP and cropped variants at three resolutions (Table 2), qualitative comparisons, hyperparameter analyses, and a small human evaluation. The central claims are that the proposed structure guidance alleviates pattern repetition and that asynchronous guidance enables speedups and per-GPU memory savings without meaningful quality loss.

Significance. If the asynchronous approximation is valid, ASGDiffusion is a practical contribution: it is training-free, supports multi-GPU parallelism, and addresses pattern repetition through a masked structure-guidance mechanism. The paper shows integration with SD1.5, SD2.1, SDXL, and SD3, and it includes an honest limitation section acknowledging failure cases at 4096x4096 and residual blur. However, the central speed/quality tradeoff is not yet established quantitatively: the only support for the asynchronous step is a qualitative sentence in Section 5.5, and the abstract's state-of-the-art claim is contradicted by the authors' own Table 2 at 3072x3072. The method is promising, but the evidence as presented does not fully support the headline claims.

major comments (4)
  1. [§4.3, Eq. (5); §5.5] The asynchronous structure guidance in Eq. (5) is the load-bearing step for the multi-GPU speedups in Table 1, but the only support for the claim that stale guidance is harmless is the sentence in §5.5: "experiments showed minimal quality differences between synchronous and asynchronous approaches." No quantitative synchronous-versus-asynchronous comparison, metric table, or statistical test is provided anywhere in the paper or supplement. Because G_t multiplies the stale prediction by w_t and is used throughout the first stage, a quantitative ablation at multiple resolutions is required to establish that the speedup does not come at a hidden quality cost. Please add FID/IS/CLIP and timing for synchronous versus asynchronous variants, with variance estimates.
  2. [§5.4, Table 2; Abstract] The abstract's claim of "state-of-the-art HR generation" is contradicted by the authors' own Table 2 at 3072x3072, where DemoFusion achieves better FID (64.85 vs 73.32), IS (17.11 vs 12.68), FIDc (53.42 vs 59.82), and ISc (21.82 vs 16.99). At 2048x2048, DemoFusion also has better FID (66.85 vs 68.49), although ASGDiffusion has a slightly better CLIP score (30.94 vs 30.48). No quality metrics are reported at 4096x4096. The paper should either restrict the state-of-the-art claim to settings where it is supported or present additional evidence at high resolutions.
  3. [§5.4, Table 2] No error bars, confidence intervals, or significance tests are reported for any of the FID, IS, FIDc, ISc, or CLIP values. Several of the reported advantages are small relative to the scale of metric noise (e.g., 64.27 vs 64.39 FID at 1024x2048). Without repeated seeds or equivalent uncertainty estimates, the "best" and "second best" markings in Table 2 are not supported. Please report standard deviations or a significance test.
  4. [§5.2, Table 1; Fig. 2] Table 1 compares 4-GPU ASGDiffusion against single-GPU baselines, and the speedup claims in §5.2 need a hardware-normalized basis. Please report total GPU-seconds, per-GPU memory, and communication overhead so that the claimed speed and memory advantages can be evaluated fairly. The caption of Fig. 2 states that ASGDiffusion is "the fastest," but Table 1 shows that on 1 GPU ASGDiffusion (40s at 2048x2048) is slower than CutDiffusion (32s); the caption should clarify that "fastest" refers to the 4-GPU configuration.
minor comments (6)
  1. [References [3] and [4]] The same DemoFusion paper appears as both reference [3] and reference [4] with different author lists and slightly different titles; please merge into a single reference.
  2. [§6, Table 3] The text refers to "Tab. 4" but the table in the main text is labeled Table 3; please renumber or correct the cross-reference.
  3. [§5.5, Fig. 7] The ablation of Structure Guidance, Cross-Attention Mask, and Asynchronous Structure Guidance is presented only with qualitative images. A quantitative ablation table (FID/IS/CLIP) would strengthen the claim that each component contributes to quality, not just to computational efficiency.
  4. [§6, Human Evaluation] The statement that ASGDiffusion "significantly outperform[s]" ScaleCrafter and DemoFusion is not supported by a significance test; with 20 participants and mean ranks of 1.68, 1.97, and 2.11, please report p-values or confidence intervals for the pairwise comparisons.
  5. [§4.2, Eq. (4)] The cross-attention mask M is introduced without a precise definition of which layer or timestep it is taken from; please specify how M is computed, normalized, and applied, since this is central to the method.
  6. [§5.6] The hyperparameter analysis is shown only at 2048x2048; since the optimal T1/(T1+T2) ratio and guidance scale may depend on the target resolution, a brief statement of how these were set for 1024x2048, 3072x3072, and 4096x4096 would improve reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the method is an empirically evaluated heuristic whose claims are tested against external baselines, not derived from fitted inputs or self-citations.

full rationale

The paper's derivation is not circular. The core proposal is an algorithmic construction: structure guidance is defined as the first patch's noise prediction, weighted by a cross-attention mask (Eq. 4), and made asynchronous by reusing guidance from timestep t-1 (Eq. 5-6). These equations define the method rather than deriving a predicted quantity from a fitted parameter; none of the quantities in them are fitted to the evaluation data. The central claims—reduced pattern repetition, faster generation, and lower per-GPU memory—are supported by comparisons to external baselines (MultiDiffusion, ScaleCrafter, DemoFusion, CutDiffusion) using FID, IS, CLIP, inference-time tables, and a user study. No fitted parameter is renamed as a prediction, and the hyperparameters are analyzed qualitatively rather than presented as derived predictions. The asynchronous approximation in Eq. 5 is the main load-bearing assumption for multi-GPU speedup, and the paper supports it only with the qualitative statement in Sec. 5.5 that 'experiments showed minimal quality differences between synchronous and asynchronous approaches'; this is a weakness in evidence strength, not circularity, because the method's validity is judged empirically against external benchmarks rather than being equivalent to its inputs. There are no load-bearing self-citations: the cited prior works, including CutDiffusion and DemoFusion, are external to the authors and their claims. Hence, the analysis finds no circular step that reduces the derivation to its own inputs.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The method rests on pretrained Stable Diffusion as a generative prior, the two-stage structure/detail assumption borrowed from CutDiffusion, the semantic validity of cross-attention masks, and the stability of guidance across adjacent denoising steps. Hyperparameters w_t and the stage ratio are manually tuned; no invented entities are introduced.

free parameters (2)
  • Structure guidance scale w_t = Tested values 0, 0.5, 1, 2, 4, 8, 16; moderate values around 1.0 or 2.0 recommended
    Controls the strength of the structure guidance in Eq. 4. It is chosen by qualitative inspection of generated images and reported as part of the hyperparameter analysis, not derived from any independent criterion.
  • Stage ratio T1/(T1+T2) = Tested 0.0 to 1.0; ratio near 0.5 recommended
    Splits denoising between global structure construction and detail refinement. The paper reports a qualitative grid search and recommends a balanced ratio, so the chosen value is fitted to perceived output quality.
assumptions (4)
  • domain assumption Pretrained Stable Diffusion models provide a sufficiently strong generative prior for both low-resolution structure and high-resolution details when used with the proposed patch-based inference.
    The entire method relies on SDXL (and other SD versions) as the frozen backbone; there is no fine-tuning and no correction for artifacts introduced by the base model.
  • domain assumption Diffusion models construct global semantic structure in early denoising steps and refine local details in later steps.
    This two-stage assumption is borrowed from CutDiffusion and is used to separate the pipeline into structure generation and detail denoising phases.
  • domain assumption Cross-attention heatmaps from the upsampling layers of the U-Net reliably indicate which regions are semantically important and can serve as the mask M in Eq. 4.
    The mask is central to the method, but the paper only provides qualitative attention-map visualizations (Fig. 6), not a quantitative justification that the mask improves or preserves fidelity across diverse prompts.
  • domain assumption Structure guidance computed at timestep t-1 is similar enough to guidance at timestep t that using stale guidance does not degrade output quality.
    This is the core of the asynchronous parallelism in Eq. 5. The paper asserts it is true based on a qualitative statement in Sec. 5.5, but no quantitative sync-versus-async comparison or statistical analysis is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ASGDiffusion: Parallel High-Resolution Generation with Asynchronous Structure Guidance." pith.science (2026). https://pith.science/paper/KQPEOCTX

@misc{pith2026241206163,
  author       = {Pith},
  title        = {Pith review of: ASGDiffusion: Parallel High-Resolution Generation with Asynchronous Structure Guidance},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KQPEOCTX}},
  note         = {Machine review of arXiv:2412.06163}
}
read the original abstract

Training-free high-resolution (HR) image generation has garnered significant attention due to the high costs of training large diffusion models. Most existing methods begin by reconstructing the overall structure and then proceed to refine the local details. Despite their advancements, they still face issues with repetitive patterns in HR image generation. Besides, HR generation with diffusion models incurs significant computational costs. Thus, parallel generation is essential for interactive applications. To solve the above limitations, we introduce a novel method named ASGDiffusion for parallel HR generation with Asynchronous Structure Guidance (ASG) using pre-trained diffusion models. To solve the pattern repetition problem of HR image generation, ASGDiffusion leverages the low-resolution (LR) noise weighted by the attention mask as the structure guidance for the denoising step to ensure semantic consistency. The proposed structure guidance can significantly alleviate the pattern repetition problem. To enable parallel generation, we further propose a parallelism strategy, which calculates the patch noises and structure guidance asynchronously. By leveraging multi-GPU parallel acceleration, we significantly accelerate generation speed and reduce memory usage per GPU. Extensive experiments demonstrate that our method effectively and efficiently addresses common issues like pattern repetition and achieves state-of-the-art HR generation.

Figures

Figures reproduced from arXiv: 2412.06163 by the authors.

Figure 1
Figure 1. The generated samples of ASGDiffusion based on Stable Diffusion 3 (SD3). While SD3 can synthesize images up to 1024x1024, [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The comparison of generated images, inference time, [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The pipeline of ASGDiffusion. Following recent works, our method also consists of two stages. In the first stage, we refine [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (12 more)
Figure 4
Figure 4. Figure 4: Timeline visualization of asynchronous structure guid [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Comparison of different methods. (a) SDXL+BSRGAN, (b) MultiDiffusion, (c) ScaleCrafter, (d) DemoFusion, (e) CutDiffusion, [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Cross-attention heatmap visualization. 5.3. Qualitative evaluation [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: The ablation study of three components introduced in [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: The Effect of the Ratio of Global Semantic Structuring to Details Denoising on Image Quality. [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Analysis of the Impact of the Structure Guidance Scale on Image Generation. [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: Left: Experimental results of our method on other ver [PITH_FULL_IMAGE:figures/full_fig_p008_10.png]
Figure 11
Figure 11. Figure 11: The figure illustrates the effects of varying guidance scale (g) and ratio (r) settings on the generated images. The top grid shows [PITH_FULL_IMAGE:figures/full_fig_p012_11.png]
Figure 12
Figure 12. Figure 12: The heatmap shows the win rates of each method in [PITH_FULL_IMAGE:figures/full_fig_p013_12.png]
Figure 13
Figure 13. Figure 13: Head-to-Head comparison between ASGDiffusion and [PITH_FULL_IMAGE:figures/full_fig_p013_13.png]
Figure 14
Figure 14. Figure 14: Additional generated results using ASGDiffusion. 5 [PITH_FULL_IMAGE:figures/full_fig_p015_14.png]
Figure 15
Figure 15. Figure 15: Additional generated results using ASGDiffusion. [PITH_FULL_IMAGE:figures/full_fig_p016_15.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. UPLiFT: Efficient Pixel-Dense Feature Upsampling with Local Attenders

    cs.CV 2026-01 conditional novelty 6.0 of 10

    UPLiFT shows that iterative 2× feature upsampling with a locally-defined attention operator beats cross-attention-based upsamplers on dense prediction while scaling linearly with token count.

Reference graph

Works this paper leans on

29 extracted references · 13 canonical work pages · cited by 1 Pith paper

  1. [1]

    Multidiffusion: Fusing diffusion paths for controlled image generation

    Omer Bar-Tal, Lior Yariv, Yaron Lipman, and Tali Dekel. Multidiffusion: Fusing diffusion paths for controlled image generation. 2023. 2, 3, 5

    work page 2023

  2. [2]

    Image super-resolution using deep convolutional net- works

    Chao Dong, Chen Change Loy, Kaiming He, and Xiaoou Tang. Image super-resolution using deep convolutional net- works. IEEE transactions on pattern analysis and machine intelligence, 38(2):295–307, 2015. 2

    2015

  3. [3]

    Demofusion: Democratising high- resolution image generation with no $$$

    Ruoyi Du, Dongliang Chang, Timothy Hospedales, Yi-Zhe Song, and Zhanyu Ma. Demofusion: Democratising high- resolution image generation with no $$$. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6159–6168, 2024. 2, 3

    work page 2024

  4. [4]

    Generalized boundary conditions in closed cosmologies

    Yilun Du et al. Demofusion: Democratising high-resolution image generation. arXiv preprint arXiv:2303.04007, 2023. 5

    work page Pith review arXiv 2023

  5. [5]

    Scalecrafter: Tuning-free higher- resolution visual generation with diffusion models

    Yingqing He et al. Scalecrafter: Tuning-free higher- resolution visual generation with diffusion models. arXiv preprint arXiv:2310.07702, 2023. 2, 3, 5

    arXiv 2023

  6. [6]

    Gans trained by a two time-scale update rule converge to a local nash equilib- rium

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilib- rium. In Advances in neural information processing systems, pages 6626–6637, 2017. 7

    work page 2017

  7. [7]

    Classifier-free diffusion guidance

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. arXiv preprint arXiv:2207.12598, 2022. 3

    arXiv 2022

  8. [8]

    Denoising dif- fusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. In Advances in Neural Informa- tion Processing Systems (NeurIPS), 2020. 3

    work page 2020

Show all 29 references
  1. [9]

    Denoising dif- fusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. Advances in neural information processing systems, 33:6840–6851, 2020. 3

    2020

  2. [10]

    Cascaded diffu- sion models for high fidelity image generation

    Jonathan Ho, Chitwan Saharia, William Chan, David J Fleet, Mohammad Norouzi, and Tim Salimans. Cascaded diffu- sion models for high fidelity image generation. Journal of Machine Learning Research, 23(47):1–33, 2022. 3

    2022

  3. [11]

    Upsample guidance: Scale up diffusion models without training

    Juno Hwang, Yong-Hyun Park, and Junghyo Jo. Upsample guidance: Scale up diffusion models without training. arXiv preprint arXiv:2404.01709, 2024. 3

    2024 arXiv

  4. [12]

    Accurate image super-resolution using very deep convolutional net- works

    Jiwon Kim, Jung Kwon Lee, and Kyoung Mu Lee. Accurate image super-resolution using very deep convolutional net- works. Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1646–1654, 2016. 2

    2016

  5. [13]

    Diffusehigh: Training-free progressive high-resolution im- age synthesis through structure guidance

    Younghyun Kim, Geunmin Hwang, and Eunbyung Park. Diffusehigh: Training-free progressive high-resolution im- age synthesis through structure guidance. arXiv preprint arXiv:2406.18459, 2024. 3

    2024 arXiv

  6. [14]

    Photo- realistic single image super-resolution using a generative ad- versarial network

    Christian Ledig, Lucas Theis, Ferenc Husz´ar, Jose Caballero, Andrew Cunningham, Alejandro Acosta, Andrew Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, et al. Photo- realistic single image super-resolution using a generative ad- versarial network. In Proceedings of the IE...

  7. [15]

    Enhanced deep residual networks for single image super-resolution

    Bee Lim, Sanghyun Son, Heewon Kim, Seungjun Nah, and Kyoung Mu Lee. Enhanced deep residual networks for single image super-resolution. In Proceedings of the IEEE confer- ence on computer vision and pattern recognition workshops, pages 136–144, 2017. 2

    2017

  8. [16]

    Cutdiffusion: A simple, fast, cheap, and strong diffusion model for image generation

    Ming Lin et al. Cutdiffusion: A simple, fast, cheap, and strong diffusion model for image generation. arXiv preprint arXiv:2401.03003, 2024. 2, 3, 5

    2024 arXiv

  9. [17]

    Mostaque

    E. Mostaque. Post on x about ai development. https : / / x . com / emostaque / status / 1563870674111832066, 2022. [Accessed: Aug. 12, 2024]. 2

    2022

  10. [18]

    Benchmark for image synthesis in medical imag- ing

    Taewoo Park, Jun-Yan Kim, Alexei A Efros, and Richard Zhang. Benchmark for image synthesis in medical imag- ing. In International Conference on Medical Image Com- puting and Computer-Assisted Intervention, pages 429–438. Springer, 2021. 7

    2021

  11. [19]

    Sd xl 1.0: Next generation high-resolution image synthesis

    Ethan Podell et al. Sd xl 1.0: Next generation high-resolution image synthesis. arXiv preprint arXiv:2303.08934, 2023. 5

    2023 arXiv

  12. [20]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. In International conference on machine learning, ...

    2021

  13. [21]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 3

    2022

  14. [22]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI) , pages 234–

  15. [23]

    Improved techniques for training gans

    Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, Xi Chen, and Xi Chen. Improved techniques for training gans. In Advances in neural informa- tion processing systems, pages 2234–2242, 2016. 7

    2016

  16. [24]

    Laion-5b: An open large-scale dataset for train- ing next generation image-text models

    Christoph Schuhmann, Ross W Beaumont, Radu Vencu, Cade W Gordon, Ross Wightman, Mehdi Cherti, Theo Coombes, Anish Katta, Adam M ¨uller, Norah Yala, et al. Laion-5b: An open large-scale dataset for train- ing next generation image-text models. arXiv preprint arXiv:2111.02114, 2021. 6

    2021 arXiv

  17. [25]

    Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network

    Wenzhe Shi, Jose Caballero, Ferenc Husz ´ar, Johannes Totz, Andrew P Aitken, Rob Bishop, Daniel Rueckert, and Zehan Wang. Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network. In Proceedings of the IEEE conference on compu...

    2016

  18. [26]

    Denoising diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502, 2020. 3

    2010 arXiv

  19. [27]

    Relay diffusion: Unifying diffusion process across resolutions for image syn- thesis

    Jiayan Teng, Wendi Zheng, Ming Ding, Wenyi Hong, Jian- qiao Wangni, Zhuoyi Yang, and Jie Tang. Relay diffusion: Unifying diffusion process across resolutions for image syn- thesis. arXiv preprint arXiv:2309.03350, 2023. 3

    2023 arXiv

  20. [28]

    Designing a practical degradation model for deep blind 9 image super-resolution

    Kai Zhang, Jingyun Liang, Luc Van Gool, and Radu Timo- fte. Designing a practical degradation model for deep blind 9 image super-resolution. arXiv preprint arXiv:2103.14006 ,

    arXiv

  21. [2021]

    • Additional visualization results – High-resolution outputs generated by ASGDiffusion

    2 10 ASGDiffusion: Parallel High-Resolution Generation with Asynchronous Structure Guidance Supplementary Material Overview The following aspects are included in this supplementary material: • Supplementary experimental analysis – Hyperparameter experiments: – User study resul...

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.