pith. sign in

arxiv: 2509.06904 · v3 · submitted 2025-09-08 · 💻 cs.CV

BIR-Adapter: A parameter-efficient diffusion adapter for blind image restoration

Cem Eteke , Alexander Griessel , Wolfgang Kellerer , Eckehard Steinbach This is my paper

Pith reviewed 2026-05-18 17:44 UTC · model grok-4.3

classification 💻 cs.CV
keywords blind image restorationdiffusion modelsparameter-efficient adapterattention mechanismimage degradationsplug-and-playsampling guidance
0
0 comments X

The pith

BIR-Adapter shows that a lightweight attention adapter can turn pretrained diffusion models into competitive blind image restorers while using up to 36 times fewer trained parameters.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents BIR-Adapter as a plug-and-play module that adds a simple attention mechanism to large pretrained diffusion models for blind image restoration. It builds on the idea that these models already hold useful representations of degraded images, so only a small adapter needs training rather than full fine-tuning or extra feature networks. The method also incorporates sampling guidance to limit hallucinations in restored outputs. Experiments cover both synthetic and real-world degradations and show the adapter matches or exceeds prior results while enabling existing models to tackle new degradation types without major changes.

Core claim

BIR-Adapter is a parameter-efficient diffusion adapter that introduces a plug-and-play attention mechanism into pretrained diffusion models. By leveraging the informative representations retained by these models under image degradations and adapting a sampling guidance mechanism, it achieves competitive or superior performance on blind image restoration tasks with significantly fewer trained parameters, up to 36 times less than state-of-the-art methods. The adapter design further allows seamless integration into existing models to handle additional unknown degradations.

What carries the argument

The BIR-Adapter, a lightweight plug-and-play attention module that extracts and applies retained representations from pretrained diffusion models to guide the restoration process.

Load-bearing premise

Large pretrained diffusion models keep sufficiently useful internal representations of degraded images that a small attention adapter can access and use effectively without full retraining or extra feature extractors.

What would settle it

A test on a novel degradation combination, such as heavy motion blur plus heavy compression artifacts absent from the training data, where the adapter's performance falls substantially below fully fine-tuned diffusion models would indicate the retained-representation premise does not hold broadly.

Figures

Figures reproduced from arXiv: 2509.06904 by Alexander Griessel, Cem Eteke, Eckehard Steinbach, Wolfgang Kellerer.

Figure 1
Figure 1. Figure 1: Example outputs of BIR-Adapter under various degradations. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Cosine similarity between the features of a clean image [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Usage of BIR-Adapter in a denoising diffusion model [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Effect of the guidance parameter ξ on the restoration of ↓4 in terms of hallucinations (PSNR) and quality (CLIP-IQA). For both metrics, higher is better. 4.2. Guided sampling As the resolution of an image increases and goes beyond the supported resolution of an LDM, a common practice is to tile the latent space with overlaps and execute the diffu￾sion model on the tiles. Finally, the tiles are merged with … view at source ↗
Figure 5
Figure 5. Figure 5: Example degraded and restored images using the baselines and our method. We used synthetic degradation on the DIV2K dataset [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 8
Figure 8. Figure 8: Example outputs of Variant 1 and Variant 2 of the ab [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
Figure 6
Figure 6. Figure 6: Effect of the guidance parameter ξ in terms of PSNR and CLIP-IQA. An increase in CLIP-IQA denotes higher quality images, while a sudden drop in PSNR hints at potential hallucina￾tions. ξ ∈ [0.75, 0.90] provides the best trade-off. (a) Variant 1 (b) Variant 2 [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Variant 1 utilizes the degraded features in the self [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: Visual effect of the guidance parameter ξ. Less guidance (higher ξ) results in more details, but no guidance leads to hallucinations. 13 [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Visual results of baseline methods and ours on a sample frame from the DIV2K validation set under synthetic degradations [ [PITH_FULL_IMAGE:figures/full_fig_p014_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Visual results of baseline methods and ours on a sample frame from the DIV2K validation set under synthetic degradations [ [PITH_FULL_IMAGE:figures/full_fig_p015_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Visual results of baseline methods and ours on a sample frame from the DIV2K validation set under synthetic degradations [ [PITH_FULL_IMAGE:figures/full_fig_p016_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Visual results from RealSR with real-world unknown degradations [ [PITH_FULL_IMAGE:figures/full_fig_p017_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Visual results from RealSR with real-world unknown degradations [ [PITH_FULL_IMAGE:figures/full_fig_p018_14.png] view at source ↗
read the original abstract

We introduce the BIR-Adapter, a parameter-efficient diffusion adapter for blind image restoration. Diffusion-based restoration methods have demonstrated promising performance in addressing this fundamental problem in computer vision, typically relying on auxiliary feature extractors or extensive fine-tuning of pre-trained models. Building on the observation that large-scale pretrained diffusion models can retain informative representations under image degradations, BIR-Adapter introduces a parameter-efficient, plug-and-play attention mechanism that substantially reduces the number of trained parameters. To further improve reliability, we adapt a sampling guidance mechanism that mitigates hallucinations during restoration. Experiments on synthetic and real-world degradations demonstrate that BIR-Adapter achieves competitive, and in several settings superior, performance compared to state-of-the-art methods while requiring up to 36x fewer trained parameters. Moreover, the adapter-based design enables integration into existing models. We validate this generality by extending a super-resolution-only diffusion model to handle additional unknown degradations, highlighting the adaptability of our approach for broader image restoration tasks.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces BIR-Adapter, a parameter-efficient, plug-and-play attention adapter for blind image restoration that operates on frozen large-scale pretrained diffusion models. Building on the premise that such models retain informative representations under degradations, the method adds a lightweight attention mechanism, incorporates sampling guidance to reduce hallucinations, and reports competitive or superior performance on synthetic and real-world degradations with up to 36x fewer trained parameters. It further demonstrates generality by extending a super-resolution-only diffusion model to additional unknown degradations.

Significance. If the empirical claims hold under rigorous controls, the work offers a practical route to efficient adaptation of diffusion models for restoration tasks, lowering the barrier to using large pretrained backbones without full fine-tuning or auxiliary extractors. The adapter design and generality experiment could influence parameter-efficient transfer in other vision domains.

major comments (2)
  1. Abstract and §4 (Experiments): the claim of 'up to 36x fewer trained parameters' and competitive/superior performance is central but lacks explicit reporting of baseline parameter counts, exact measurement protocol (e.g., trainable vs. total parameters), and statistical significance across runs. Without these, the efficiency advantage cannot be verified as load-bearing for the main contribution.
  2. §4.1 and Table 2 (synthetic degradations): the experimental controls for blind restoration (e.g., whether test degradations match training distributions, choice of baselines, and handling of unknown degradations) are not described in sufficient detail to support the 'competitive and in several settings superior' claim; this directly affects the reliability of the performance results.
minor comments (2)
  1. §3 (Method): clarify the exact architecture of the attention adapter (e.g., query/key/value dimensions, insertion points in the diffusion U-Net) and whether any components are frozen vs. trained.
  2. Figure 3 or §4.3 (generality experiment): provide quantitative metrics for the extended super-resolution model on the additional degradations rather than qualitative examples alone.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment below and commit to revisions that will strengthen the experimental reporting and clarity of the manuscript.

read point-by-point responses

  1. Referee: Abstract and §4 (Experiments): the claim of 'up to 36x fewer trained parameters' and competitive/superior performance is central but lacks explicit reporting of baseline parameter counts, exact measurement protocol (e.g., trainable vs. total parameters), and statistical significance across runs. Without these, the efficiency advantage cannot be verified as load-bearing for the main contribution.

    Authors: We agree that explicit documentation of parameter counts and measurement details is needed to make the efficiency claims fully verifiable. In the revised manuscript we will add a dedicated subsection and table in §4 that lists the exact number of trainable parameters for BIR-Adapter and every baseline, with a clear protocol stating that only parameters updated during adapter training are counted while the pretrained diffusion backbone remains frozen. We will also report mean performance and standard deviation over at least three independent runs for the key tables to address statistical significance. revision: yes

  2. Referee: §4.1 and Table 2 (synthetic degradations): the experimental controls for blind restoration (e.g., whether test degradations match training distributions, choice of baselines, and handling of unknown degradations) are not described in sufficient detail to support the 'competitive and in several settings superior' claim; this directly affects the reliability of the performance results.

    Authors: We accept that the current description of the blind-restoration protocol is insufficiently detailed. We will expand §4.1 with an explicit paragraph that (i) states how synthetic test degradations are generated to ensure they lie outside the training distribution, (ii) justifies the selection of baselines, and (iii) clarifies the evaluation procedure for truly unknown degradations. These additions will directly support the reliability of the reported performance comparisons. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper is an empirical engineering contribution that introduces BIR-Adapter as a plug-and-play attention mechanism on frozen pretrained diffusion models, building directly on the stated observation that such models retain informative representations under degradations. No derivation chain, first-principles equations, or predictions are presented that reduce by construction to fitted parameters, self-citations, or renamed inputs. Performance claims rest on experimental comparisons to SOTA methods across synthetic and real degradations, with the efficiency and generality results following from the adapter design and sampling guidance without internal reduction to the inputs. The approach is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The approach relies on the empirical observation that pretrained diffusion models preserve useful features under degradation; this is treated as a domain assumption rather than a derived result. No explicit free parameters or invented entities are named in the abstract.

axioms (1)
  • domain assumption Large-scale pretrained diffusion models retain informative representations under image degradations
    Stated in the abstract as the key observation enabling the lightweight adapter design.

pith-pipeline@v0.9.0 · 5704 in / 1330 out tokens · 30703 ms · 2026-05-18T17:44:55.303643+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 1 Pith paper

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

  1. A Causal Diffusion Model for Video Reconstruction from Ultra-Low-Bitrate Representations

    cs.CV 2026-02 unverdicted novelty 7.0

    A causal diffusion model reconstructs videos from ultra-low-bitrate semantics and compressed frames using temporal distillation from a bidirectional teacher, outperforming prior baselines.

Reference graph

Works this paper leans on

40 extracted references · 40 canonical work pages · cited by 1 Pith paper · 2 internal anchors

  1. [1]

    Ntire 2017 challenge on single image super-resolution: Dataset and study

    Eirikur Agustsson and Radu Timofte. Ntire 2017 challenge on single image super-resolution: Dataset and study. InPro- ceedings of the IEEE conference on computer vision and pat- tern recognition workshops, pages 126–135, 2017. 5, 11, 14, 15, 16

    work page 2017

  2. [2]

    Toward real-world single image super-resolution: A new benchmark and a new model

    Jianrui Cai, Hui Zeng, Hongwei Yong, Zisheng Cao, and Lei Zhang. Toward real-world single image super-resolution: A new benchmark and a new model. InProceedings of the IEEE/CVF international conference on computer vision, pages 3086–3095, 2019. 5, 17, 18

    work page 2019

  3. [3]

    Transfer clip for gen- eralizable image denoising

    Jun Cheng, Dong Liang, and Shan Tan. Transfer clip for gen- eralizable image denoising. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 25974–25984, 2024. 2

    work page 2024

  4. [4]

    Effective diffusion transformer architecture for image super- resolution

    Kun Cheng, Lei Yu, Zhijun Tu, Xiao He, Liyu Chen, Yong Guo, Mingrui Zhu, Nannan Wang, Xinbo Gao, and Jie Hu. Effective diffusion transformer architecture for image super- resolution. InProceedings of the AAAI Conference on Arti- ficial Intelligence, pages 2455–2463, 2025. 3, 6, 20

    work page 2025

  5. [5]

    Image denoising by sparse 3-d transform- domain collaborative filtering.IEEE Transactions on image processing, 16(8):2080–2095, 2007

    Kostadin Dabov, Alessandro Foi, Vladimir Katkovnik, and Karen Egiazarian. Image denoising by sparse 3-d transform- domain collaborative filtering.IEEE Transactions on image processing, 16(8):2080–2095, 2007. 2

    work page 2080

  6. [6]

    A re- view on generative adversarial networks for image genera- tion.Computers & Graphics, 114:13–25, 2023

    Vinicius Luis Trevisan De Souza, Bruno Augusto Dorta Mar- ques, Harlen Costa Batagelo, and Jo ˜ao Paulo Gois. A re- view on generative adversarial networks for image genera- tion.Computers & Graphics, 114:13–25, 2023. 2

    work page 2023

  7. [7]

    Diffusion models beat gans on image synthesis.Advances in neural informa- tion processing systems, 34:8780–8794, 2021

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis.Advances in neural informa- tion processing systems, 34:8780–8794, 2021. 2, 3

    work page 2021

  8. [8]

    Image super-resolution using deep convolutional net- works.IEEE transactions on pattern analysis and machine intelligence, 38(2):295–307, 2015

    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

    work page 2015

  9. [9]

    Generative adversarial networks.Commu- nications of the ACM, 63(11):139–144, 2020

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial networks.Commu- nications of the ACM, 63(11):139–144, 2020. 2

    work page 2020

  10. [10]

    Div8k: Diverse 8k resolution image dataset

    Shuhang Gu, Andreas Lugmayr, Martin Danelljan, Manuel Fritsche, Julien Lamour, and Radu Timofte. Div8k: Diverse 8k resolution image dataset. In2019 IEEE/CVF Interna- tional Conference on Computer Vision Workshop (ICCVW), pages 3512–3516. IEEE, 2019. 5

    work page 2019

  11. [11]

    Diffusion models in low-level vision: A survey.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

    Chunming He, Yuqi Shen, Chengyu Fang, Fengyang Xiao, Longxiang Tang, Yulun Zhang, Wangmeng Zuo, Zhenhua Guo, and Xiu Li. Diffusion models in low-level vision: A survey.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025. 2

    work page 2025

  12. [12]

    Classifier-Free Diffusion Guidance

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance.arXiv preprint arXiv:2207.12598, 2022. 6

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  13. [13]

    Denoising dif- fusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020. 2, 3

    work page 2020

  14. [14]

    A style-based generator architecture for generative adversarial networks

    Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial networks. InProceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 4401–4410, 2019. 5

    work page 2019

  15. [15]

    Musiq: Multi-scale image quality transformer

    Junjie Ke, Qifei Wang, Yilin Wang, Peyman Milanfar, and Feng Yang. Musiq: Multi-scale image quality transformer. InProceedings of the IEEE/CVF international conference on computer vision, pages 5148–5157, 2021. 7

    work page 2021

  16. [16]

    Diffusion restoration adapter for real-world image restoration.arXiv preprint arXiv:2502.20679, 2025

    Hanbang Liang, Zhen Wang, and Weihui Deng. Diffusion restoration adapter for real-world image restoration.arXiv preprint arXiv:2502.20679, 2025. 3

    work page arXiv 2025

  17. [17]

    Swinir: Image restoration us- ing swin transformer

    Jingyun Liang, Jiezhang Cao, Guolei Sun, Kai Zhang, Luc Van Gool, and Radu Timofte. Swinir: Image restoration us- ing swin transformer. InProceedings of the IEEE/CVF inter- national conference on computer vision, pages 1833–1844,

    work page

  18. [18]

    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. InThe IEEE Conference on Com- puter Vision and Pattern Recognition (CVPR) Workshops,

    work page

  19. [19]

    Diff- bir: Toward blind image restoration with generative diffusion prior

    Xinqi Lin, Jingwen He, Ziyan Chen, Zhaoyang Lyu, Bo Dai, Fanghua Yu, Yu Qiao, Wanli Ouyang, and Chao Dong. Diff- bir: Toward blind image restoration with generative diffusion prior. InEuropean Conference on Computer Vision, pages 430–448. Springer, 2024. 3, 5, 6, 19

    work page 2024

  20. [20]

    Decoupled Weight Decay Regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017. 5

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  21. [21]

    Efficient diffusion models: A com- prehensive survey from principles to practices.IEEE Trans- actions on Pattern Analysis and Machine Intelligence, 2025

    Zhiyuan Ma, Yuzhu Zhang, Guoli Jia, Liangliang Zhao, Yichao Ma, Mingjie Ma, Gaofeng Liu, Kaiyan Zhang, Ning Ding, Jianjun Li, et al. Efficient diffusion models: A com- prehensive survey from principles to practices.IEEE Trans- actions on Pattern Analysis and Machine Intelligence, 2025. 3

    work page 2025

  22. [22]

    Deep multi-scale convolutional neural network for dynamic scene deblurring

    Seungjun Nah, Tae Hyun Kim, and Kyoung Mu Lee. Deep multi-scale convolutional neural network for dynamic scene deblurring. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 3883–3891,

    work page

  23. [23]

    Semantic image synthesis with spatially-adaptive nor- malization

    Taesung Park, Ming-Yu Liu, Ting-Chun Wang, and Jun-Yan Zhu. Semantic image synthesis with spatially-adaptive nor- malization. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2337–2346,

    work page

  24. [24]

    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. InInternational conference on machine learning, pages 8748–8763. PmLR, 2021. 2

    work page 2021

  25. [25]

    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. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 2, 3

    work page 2022

  26. [26]

    A survey of deep learning approaches to image restoration.Neurocomputing, 487:46–65, 2022

    Jingwen Su, Boyan Xu, and Hujun Yin. A survey of deep learning approaches to image restoration.Neurocomputing, 487:46–65, 2022. 2

    work page 2022

  27. [27]

    Attention is all you need.Advances in neural information processing systems, 30, 2017

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia 9 Polosukhin. Attention is all you need.Advances in neural information processing systems, 30, 2017. 2, 3

    work page 2017

  28. [28]

    Ex- ploring clip for assessing the look and feel of images

    Jianyi Wang, Kelvin CK Chan, and Chen Change Loy. Ex- ploring clip for assessing the look and feel of images. In AAAI, 2023. 7

    work page 2023

  29. [29]

    Exploiting diffusion prior for real-world image super-resolution.International Journal of Computer Vision, 132(12):5929–5949, 2024

    Jianyi Wang, Zongsheng Yue, Shangchen Zhou, Kelvin CK Chan, and Chen Change Loy. Exploiting diffusion prior for real-world image super-resolution.International Journal of Computer Vision, 132(12):5929–5949, 2024. 3, 5, 6, 20

    work page 2024

  30. [30]

    Recovering realistic texture in image super-resolution by deep spatial feature transform

    Xintao Wang, Ke Yu, Chao Dong, and Chen Change Loy. Recovering realistic texture in image super-resolution by deep spatial feature transform. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 606–615, 2018. 5

    work page 2018

  31. [31]

    Real-esrgan: Training real-world blind super-resolution with pure synthetic data

    Xintao Wang, Liangbin Xie, Chao Dong, and Ying Shan. Real-esrgan: Training real-world blind super-resolution with pure synthetic data. InProceedings of the IEEE/CVF inter- national conference on computer vision, pages 1905–1914,

    work page 1905

  32. [32]

    Maniqa: Multi-dimension attention network for no-reference image quality assessment

    Sidi Yang, Tianhe Wu, Shuwei Shi, Shanshan Lao, Yuan Gong, Mingdeng Cao, Jiahao Wang, and Yujiu Yang. Maniqa: Multi-dimension attention network for no-reference image quality assessment. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1191–1200, 2022. 7

    work page 2022

  33. [33]

    Pixel-aware stable diffusion for realistic image super-resolution and personalized stylization

    Tao Yang, Rongyuan Wu, Peiran Ren, Xuansong Xie, and Lei Zhang. Pixel-aware stable diffusion for realistic image super-resolution and personalized stylization. InEuropean Conference on Computer Vision, pages 74–91. Springer,

    work page

  34. [34]

    Scaling up to excellence: Practicing model scaling for photo- realistic image restoration in the wild

    Fanghua Yu, Jinjin Gu, Zheyuan Li, Jinfan Hu, Xiangtao Kong, Xintao Wang, Jingwen He, Yu Qiao, and Chao Dong. Scaling up to excellence: Practicing model scaling for photo- realistic image restoration in the wild. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 25669–25680, 2024. 3, 5, 6, 19, 20

    work page 2024

  35. [35]

    A com- prehensive review of deep learning-based real-world image restoration.IEEE Access, 11:21049–21067, 2023

    Lujun Zhai, Yonghui Wang, Suxia Cui, and Yu Zhou. A com- prehensive review of deep learning-based real-world image restoration.IEEE Access, 11:21049–21067, 2023. 2

    work page 2023

  36. [36]

    Learning deep cnn denoiser prior for image restoration

    Kai Zhang, Wangmeng Zuo, Shuhang Gu, and Lei Zhang. Learning deep cnn denoiser prior for image restoration. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 3929–3938, 2017. 2

    work page 2017

  37. [37]

    Designing a practical degradation model for deep blind im- age super-resolution

    Kai Zhang, Jingyun Liang, Luc Van Gool, and Radu Timofte. Designing a practical degradation model for deep blind im- age super-resolution. InProceedings of the IEEE/CVF inter- national conference on computer vision, pages 4791–4800,

    work page

  38. [38]

    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 Proceedings of the IEEE/CVF international conference on computer vision, pages 3836–3847, 2023. 3

    work page 2023

  39. [39]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. InCVPR, 2018. 7

    work page 2018

  40. [40]

    export NEGATIVE_PROMPT =

    Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-image translation using cycle- consistent adversarial networks. InProceedings of the IEEE international conference on computer vision, pages 2223– 2232, 2017. 2 10 BIR-Adapter: A Low-Complexity Diffusion Model Adapter for Blind Image Restoration Supplementary Material Cem Etek...

    work page 2017