Pith. sign in

REVIEW 3 major objections 4 minor 62 references

This paper claims that on a new benchmark for art images damaged by semi-transparent texture overlays, general-purpose image editing models outperform specialized restoration models, and that explicit degradation-aware prompts widen the gap

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-05 00:23 UTC pith:TFHDA5BC

load-bearing objection A useful new art-restoration benchmark whose headline claim is undermined by the mask-confounded comparison; worth peer review with required revisions. the 3 major comments →

arxiv 2608.00736 v1 pith:TFHDA5BC submitted 2026-08-01 cs.CV

MDTD-ArtIR: Benchmarking Image Editing and Restoration Models for Art Image Restoration under Texture-Overlay Degradations

Mridula Vijendran , Shuang Chen , Hubert P. H. Shum This is my paper
classification cs.CV
keywords art restorationbenchmark datasetimage editing modelsuniversal image restorationprompt engineeringtexture degradationalpha mask overlayimage quality assessment
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper sets out to establish an empirical claim: when artwork is degraded by semi-transparent texture overlays — the synthetic analogue of cracks, stains, flaking, and material loss — general-purpose image editing models reconstruct the clean painting more faithfully than models purpose-built for image restoration. To make the test controlled, it builds MDTD-Art, pairing 1,000 clean art images with 47 texture classes blended toward white at controlled alpha opacities (Eq. 1), producing 5,640 degraded images across low, medium, and high severity. Across L1, PSNR, SSIM, and LPIPS, the benchmark finds that editing models such as Qwen Image Edit 2511 and Nano Banana Pro consistently beat restoration-specific models such as AutoDIR, HYPIR, BIRD, and LanPaint, with the gap widening as mask opacity rises. An explicit degradation-aware prompt — telling the model the masked regions are corrupted, not intentional content — amplifies the gains for some models, most clearly at high opacity, while a few models respond negatively or not at all. The paper attributes the results to recoverable semantic information under the mask plus prompt controllability, and argues these, not degradation-specific priors, are what art restoration models actually need.

Core claim

The central discovery is empirical, carried by a new construction. The paper defines a degradation family by alpha-compositing a grayscale texture mask toward white, I_deg = α⊙1 + (1−α)⊙I_clean (Eq. 1), using DTD texture classes over WikiArt paintings; with the mask hidden at inference and no closed-form inverse, the task is blind restoration of a hybrid global fade plus local texture inpainting. On this task, the measurements show large image editing models (Qwen Image Edit 2511 and Nano Banana Pro lead) outperform universal restoration models (AutoDIR, BIRD, LanPaint, HYPIR) on L1, PSNR, SSIM, and LPIPS across opacity levels, and an explicit degradation-aware prompt — framing masked region

What carries the argument

The load-bearing mechanism is the alpha texture-mask overlay of Eq. 1 — I_deg = α⊙1 + (1−α)⊙I_clean — which blends a grayscale DTD texture toward white over the clean painting, controlling the severity of information loss (low/medium/high opacity) independently of the degradation class (47 texture types). This converts restoration into a hybrid of blind restoration and inpainting in which partial semantic signal survives under the mask. Carrying the argument is the benchmark built on it: a class-stratified, combinatorially generated dataset (MDTD-Art, 5,640 images), a proxy mask predictor (an EfficientNet-B0 UNet) that feeds mask-requiring baselines, and a two-condition prompt design (generi

Load-bearing premise

The entire benchmark rests on the premise that overlaying texture images faded toward white, at controlled opacity, reproduces the look and spatial statistics of authentic art damage closely enough that the measured model rankings transfer to real restoration — a premise supported only by a small qualitative comparison (Fig. 5), which the paper itself calls an incomplete proxy.

What would settle it

Run the same model suite and the same two prompts on genuinely damaged artworks that have clean reference versions — for instance, mural or heritage-painting damage with known reconstructions — and check whether the rankings of Tables 5 and 6 survive on real damage. A reversal (restoration models matching or beating the editors) or a large measured distributional mismatch between MDTD-Art overlays and real crack/stain photographs would show the benchmark's conclusion is an artifact of the synthetic proxy rather than a fact about art restoration.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • If the finding holds, art restoration tooling should treat prompt-guided editing models as a primary option, with explicit degradation-aware prompts as a per-model, per-severity lever: Nano Banana Pro gains 2.24 dB PSNR overall (9.63 to 11.87 dB at high opacity) from the explicit prompt, showing the prompt itself carries restorative signal.
  • The advantage is severity-dependent: at low opacity the field converges and AutoDIR stays competitive, while at medium and high opacity the editing models are the only ones that extrapolate plausible content, so the right tool depends on how much of the image survives.
  • The paper itself defers a key explanation: broad pretraining may drive the editing-model advantage, and separating pretraining, scale, and architecture effects requires controlled experiments this benchmark does not run.
  • Hard cases reveal a defined failure mode — masks with strong semantic content (grids, faces, patterns) hijack editing-model outputs, causing identity drift, hallucinated scene elements, and even complete identity loss, so controllability and structural fidelity, not raw IQA score, are the next open problems.
  • MDTD-Art is reusable: because Eq. 1 decouples texture class from opacity level and composes with any clean image domain, it gives the community a controlled instrument for separating genuine semantic priors from shallow interpolation off lightly corrupted regions.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Because the overlay blends toward white, the benchmark rewards models that infer the painting's true local color through the mask; testing non-white blend targets (aged-varnish yellow, gray, darkening) would show whether the ranking is partly an artifact of white-bias, and the paper's own observations of yellow tonal shift in Qwen and supersaturation in SD3 hint the ranking could shift.
  • The mask-based baselines were fed a predicted proxy mask whose errors were not propagated (Sec. 4.2); giving them the oracle mask instead could narrow or reorder the gap, so the headline claim is really about full-pipeline performance, not restoration capacity alone.
  • The core task — separating a semi-transparent structured overlay from underlying content — is a general overlay-disambiguation capability; the same alpha-overlay construction could transfer to document cleaning, old-photo scratch or reflection removal, and material-loss detection, where partial signal under an overlay is the shared structure.
  • No metric in the study isolates identity preservation, yet for heritage use altering subject identity is the unacceptable failure; applying identity-distance or segmentation-consistency metrics to the hard samples would quantify hallucination risk and could reorder model choice for conservation even where IQA scores favor the editors.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper introduces MDTD-Art, a synthetic benchmark for restoring art images damaged by semi-transparent textured overlays. Textured masks from the DTD dataset are alpha-blended toward white over clean WikiArt paintings (Eq. 1) at three opacity levels, and the paper evaluates a range of closed/open image-editing models and universal image restoration (UIR) models under generic and explicit degradation-aware prompts. The authors report that image-editing models, particularly Qwen Image Edit 2511 and Nano Banana Pro, outperform restoration-specific models and that explicit prompts improve certain models at high opacity. They also train a lightweight UNet to predict the overlay mask for mask-conditioned baselines (LanPaint, HYPIR, BIRD).

Significance. If the comparison were unconfounded, this would be a useful controlled stress test for an under-studied restoration setting: recovering content under semi-transparent, semantically patterned overlays. The paper is transparent about model versions, API dates, and the exploratory nature of the real-world comparison, and the idea of separating opacity severity from texture class is sound. However, the core claim is currently supported only through a comparison in which mask-conditioned baselines receive noisy proxy masks, and the benchmark's scope is narrower than the phrase 'arbitrary degradations' suggests. These issues are addressable with additional controls, so the work has potential, but the present evidence does not establish the headline conclusion.

major comments (3)
  1. [Section 4.2 / Table 4 / Table 6] The comparison supporting the headline claim is confounded by mask quality. Table 6 evaluates LanPaint, HYPIR, and BIRD using proxy masks predicted by the EfficientNet-B0 UNet, whose reconstruction quality is only SSIM 0.48 / PSNR 16.21 dB (Table 4). Section 4.2 explicitly states that mask-prediction errors are not taken into account in the quantitative results. Since the degradation is exactly the alpha overlay of Eq. (1), an inaccurate mask directly removes task-critical information from these baselines, and BIRD is run in inpainting mode with the proxy mask. To support the claim that image editing models outperform specialized restoration architectures, the paper must add an oracle-mask control or otherwise quantify how much of the gap is attributable to mask error. Without this, the reported gap may reflect mask-estimation quality rather than restoration capability.
  2. [Abstract; Section 3.1-3.2; Figure 5] The phrase 'arbitrary degradations' overstates the benchmark. The degradation model is a single parametric family, Eq. (1): a white-alpha DTD texture overlay at varying opacity. The 47 DTD classes are a fixed texture set, not arbitrary degradations. The external validity of the benchmark therefore rests on the premise that these overlays approximate real cracks, stains, flaking, and material loss. The only evidence for this premise is the exploratory qualitative comparison in Fig. 5, and Section 4.4 explicitly cautions that the real-world samples are limited and the synthetic dataset is an incomplete proxy. Please either provide a quantitative validation against real damaged images or restrict the conclusions to the proposed texture-overlay family.
  3. [Section 3.2 / Table 1 / Section 4.1] The dataset construction is internally inconsistent. With N=1000 clean images, C=47 classes, and S=50 samples per class, Eqs. (2)-(3) define a corpus of N*(C*S) = 2,350,000 degraded images. Table 1 reports 5,640 images for MDTD-Art, and Section 4.1 states only 1000 clean images. For benchmarking, C=3 and an unspecified S are used, which is a different dataset from the released one. Please clarify the exact number of released degraded images, the role of C and S in each experiment, and why Table 1 reports 5,640.
minor comments (4)
  1. [Section 4.4] The first paragraph contains a duplicated phrase: 'indicates that indicates that'. Please fix.
  2. [Table 5 caption] The caption uses 'opacitys' instead of 'opacities'. Tables 5 and 6 also have identical captions; please differentiate them.
  3. [Abstract and dataset availability] The paper states the dataset is publicly open, but no dataset URL or release link appears in the manuscript. Please provide a link or specify the intended release venue.
  4. [Section 4.3 / Table 5] The claim that 'performance gains are amplified by structured prompt engineering' should be qualified. The gains are strong for NB Pro and Flux 2 at high opacity, but Qwen Image Edit is largely insensitive and SD3 Medium degrades under the explicit prompt. A sentence summarizing this heterogeneity would better reflect the data.

Circularity Check

0 steps flagged

No circularity: benchmark conclusion comes from direct head-to-head experiments; mask-proxy confound is a validity limitation, not a circular derivation.

full rationale

MDTD-ArtIR's central claim—that image editing models outperform specialized restoration architectures on this degradation family—is supported by direct side-by-side quantitative evaluation (Tables 5 and 6) of independently obtained model outputs on the fixed alpha-overlay protocol (Eq. 1). The conclusion is not encoded in Eq. 1 or in any fitted parameter; Eq. 1 only constructs the benchmark input. The mask-conditioned baselines receive proxy masks from a separately trained EfficientNet-B0 UNet (Sec 4.2, Table 4), and the paper transparently states that 'The errors in mask prediction are not taken into account' — a real internal-validity confound, because the restoration-versus-editing comparison is entangled with mask-estimation quality, but it is a limitation of the experimental design, not a circular derivation of the result from its inputs. Likewise, the synthetic-to-real transferability is explicitly hedged in Sec 4.4 ('exploratory', 'incomplete proxy'), which affects external validity, not circularity. Self-citations ([14], [28], [48]) occur only in related-work/framing contexts and do not supply the load-bearing evidence for the headline result. No equation, fitted value, or self-cited premise is equivalent by construction to the reported ranking, so there is no circularity.

Axiom & Free-Parameter Ledger

2 free parameters · 3 axioms · 0 invented entities

The central claim depends on the assumptions that the synthetic white-alpha texture overlays represent real artistic damage, that the evaluation protocol (resize, crop, full-reference metrics) captures restoration quality, and that the hand-chosen opacity thresholds and class/subset sizes define a meaningful difficulty axis. No new physical or model entities are introduced.

free parameters (2)
  • alpha opacity group thresholds = low [0,0.33), medium (0.33,0.66], high (0.66,1]
    Hand-chosen cutoffs define the low/medium/high difficulty levels; all reported metrics are aggregated within these bins, so the central comparison depends on these thresholds.
  • texture class count C and subset size S for mask sampling = C=47, S=50 for training; C=3, S=selected samples for benchmarking
    Hand-picked design choices in the dataset construction (Eq. 2-3); they determine the degradation distribution and the mask predictor training.
axioms (3)
  • domain assumption DTD texture alpha maps approximate the spatial statistics of real artistic damage (cracks, stains, flaking, material loss).
    Invoked in Section 3.1 to justify using DTD as a proxy; load-bearing because the benchmark's relevance to real restoration depends on it.
  • domain assumption The white-blend degradation I_deg = alpha*1 + (1-alpha)*I_clean models partial information retention in damaged artworks.
    Eq. (1) defines the degradation; if real damage is not white-fade-like (e.g., colored stains, dark cracks), the benchmark may not cover those cases.
  • domain assumption Full-reference IQA metrics on center-cropped 512x512 outputs are a valid fidelity measure even when models alter aspect ratio or composition.
    The evaluation protocol in Section 3.5 assumes deterministic resize+crop yields comparable geometry; the paper itself notes 'a spatial misalignment occurs with the varying aspect ratios of the native input dataset'.

pith-pipeline@v1.3.0-alltime-deepseek · 18088 in / 14607 out tokens · 143667 ms · 2026-08-05T00:23:36.683684+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of MDTD-ArtIR: Benchmarking Image Editing and Restoration Models for Art Image Restoration under Texture-Overlay Degradations." pith.science (2026). https://pith.science/paper/TFHDA5BC

@misc{pith2026260800736,
  author       = {Pith},
  title        = {Pith review of: MDTD-ArtIR: Benchmarking Image Editing and Restoration Models for Art Image Restoration under Texture-Overlay Degradations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TFHDA5BC}},
  note         = {Machine review of arXiv:2608.00736}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Restoring severely degraded visual media still remains a formidable challenge, as existing methods often hallucinate unnatural textures and contents, struggle with preserving color and texture, or fail to leverage partially retained image information. Existing restoration benchmarks assume known degradation operators and fail to capture the complex characteristics of artistic damage such as cracks, stains, and color/texture deviation. We introduce a controlled benchmark for blind restoration of semantic, semi-transparent image media degradations, accompanied by a new, publicly open degradation alpha texture mask dataset MDTD-Art. We present a new dataset and benchmark evaluating state-of-the-art universal restoration models against image editing and vision-language models across varying mask opacity levels. Our experiments demonstrate that image editing models consistently outperform specialized restoration architectures for arbitrary degradations, with performance gains amplified by structured prompt engineering emphasizing detail preservation and structural consistency. These findings position recoverable semantic information and prompt controllability as critical factors in art image restoration.

Figures

Figures reproduced from arXiv: 2608.00736 by Hubert P. H. Shum, Mridula Vijendran, Shuang Chen.

Figure 1
Figure 1. Figure 1: Visual comparison of image restoration results on four sample images from different domains with cropped native model outputs for same view comparisons. The Masked image is shown in the first column, followed by predictions from models GPT Image 1.5, Nano Banana Pro, Flux 2 and AutoDIR, and finally shows the ground truth . datasets provides value to bridge this gap, as seen with other domains in the same r… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of Data Distribution as a heuristic visual grouping. The DTD dataset is selected to approximate different types of damages found in artworks and manuscripts and further stratified into levels of mask opacities. The mask images are class-wise categorized into 3 levels of alpha groups based on the average alpha channel values. or specialized for fixed degradation operators have no direct prior for t… view at source ↗
Figure 3
Figure 3. Figure 3: Visual comparison of image restoration results on three sample images across different alpha texture mask opacitys. The Masked image is shown in the first column, followed by predictions from the models, Qwen Image Edit, Nano Banana (NB) Pro, Step1x Edit, AutoDIR, and LanPaint with Qwen backbone. Masked Image Qwen Image Edit NB Pro Flux 2 SD3 Medium Ground Truth Simple Complex [PITH_FULL_IMAGE:figures/ful… view at source ↗
Figure 4
Figure 4. Figure 4: Visual comparison of image restoration results on two sample images across different prompt types for image editing models. The first column explores restorations under generic prompts and the second explores explicit degradation-aware prompts. The Masked image is shown in the first column, followed by predictions from the models, NB Pro, Flux 2, SD3 Medium, and Qwen Image Edit . broadly generalizable insi… view at source ↗
Figure 5
Figure 5. Figure 5: Visual comparison of image restoration results on sample images from multiple datasets [61, 62, 6] with real world degradations. The Masked image is shown in the first column, followed by predictions from the models, NB Pro, AutoDIR, Step1x Edit, LanPaint, and Qwen Image Edit . details present in the underlying image itself are blurred. Flux 2, in contrast, produces more photorealistic images with vivid co… view at source ↗
Figure 6
Figure 6. Figure 6: Hard degradation samples for different models when utilizing the learned model priors for the non-explicit prompts (generic and explicit degradation-aware) under different levels of mask opacitys. controllability and robustness remain open problems. Future work should explore soft masking strategies, degradation￾aware prompt conditioning, and hybrid architectures that combine the semantic flexibility of ed… view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

62 extracted references · 51 canonical work pages · 1 internal anchor

  1. [1]

    Z. Wan, B. Zhang, D. Chen, P. Zhang, D. Chen, J. Liao, F. Wen, Bringing old photos back to life, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 2747–2757

    work page 2020

  2. [2]

    X. Luo, X. Zhang, P. Yoo, R. Martin-Brualla, J. Lawrence, S. M. Seitz, Time-travel rephotography, ACM Transactions on Graphics (Proceedings of ACM SIGGRAPH Asia 2021) 40 (2021)

    work page 2021

  3. [3]

    F.M.Garcia-Moreno,L.R.Rodríguez-Simón,M.V.Hurtado-Torres, etal.,Artinsight:Adetaileddatasetfordetectingdeteriorationineasel paintings, Data in Brief 61 (2025) 111811

    work page 2025

  4. [4]

    W.Cai,X.Xu,J.Xu,H.Zhang,H.Yang,K.Zhang,S.He, Hierarchi- caldamagecorrelationsforoldphotorestoration, InformationFusion 107 (2024) 102340

    work page 2024

  5. [5]

    H. Shah, V. Manjula, A performance analysis of deep convolutional neural networks using kuzushiji character recognition, in: 2020 International Conference on Decision Aid Sciences and Application (DASA), IEEE, 2020, pp. 1068–1071

    work page 2020

  6. [6]

    Z. Xu, Y. Yang, Q. Fang, W. Chen, T. Xu, J. Liu, Z. Wang, A comprehensive dataset for digital restoration of dunhuang murals, Scientific Data 11 (2024) 955

    work page 2024

  7. [8]

    4909–4919

    X.Yin,J.Hu,Z.You,K.Yan,Y.Tang,C.Dong,J.Gu, Howfarhave we gone in generative image restoration? a study on its capability, limitationsandevaluationpractices,in:ProceedingsoftheIEEE/CVF Conference on Computer Vision and Pattern Recognition, 2026, pp. 4909–4919

    work page 2026

  8. [9]

    W. Sun, X. Yin, C. Dong, Can nano banana 2 replace traditional imagerestorationmodels?anevaluationofitsperformanceonimage restoration tasks, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2026, pp. 4581–4590

    work page 2026

  9. [10]

    O’Brien, J

    C. O’Brien, J. Hutson, T. Olsen, J. Ratican, Limitations and pos- sibilities of digital restoration techniques using generative ai tools: Reconstituting antoine françois callet’s achilles dragging hector’s body past the walls of troy, Arts & Communication (2023)

    work page 2023

  10. [11]

    Huang, B

    S. Huang, B. Gong, Y. Feng, X. Chen, Y. Fu, Y. Liu, D. Wang, Learningdisentangledidentifiersforaction-customizedtext-to-image generation, in: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition, 2024, pp. 7797–7806

    work page 2024

  11. [12]

    Vinker, A

    Y. Vinker, A. Voynov, D. Cohen-Or, A. Shamir, Concept decompo- sition for visual exploration and inspiration, ACM Transactions on Graphics (TOG) 42 (2023) 1–13

    work page 2023

  12. [13]

    Zheng, Y

    C. Zheng, Y. Lan, Y. Wang, Lanpaint: Training-free diffusion inpainting with asymptotically exact and fast conditional sampling, arXiv preprint arXiv:2502.03491 (2025)

    work page arXiv 2025

  13. [14]

    Z.Chang,G.A.Koulieris,H.J.Chang,H.P.H.Shum, Onthedesign fundamentals of diffusion models: A survey, Pattern Recognition (2026) 111934

    work page 2026

  14. [15]

    M. Sah, S. Suman, J. Mathew, Retrieval augmented generation for smartcalorieestimationincomplexfoodscenarios, JournalofVisual Communication and Image Representation (2025) 104632

    work page 2025

  15. [16]

    M. A. Yılmaz, A. Bilican, B. C. Biner, A. M. Tekalp, Edit2restore: Few-shot image restoration via parameter-efficient adaptation of pre- trained editing models, arXiv preprint arXiv:2601.03391 (2026). :Preprint submitted to Elsevier Page 13 of 15 MDTD-ArtIR

    arXiv 2026

  16. [17]

    Y. Bai, C. Wang, S. Xie, C. Dong, C. Yuan, Z. Wang, Textir: A simple framework for text-based editable image restoration, IEEE Transactions on Visualization and Computer Graphics (2025)

    work page 2025

  17. [18]

    Avrahami, K

    O. Avrahami, K. Aberman, O. Fried, D. Cohen-Or, D. Lischinski, Break-a-scene: Extracting multiple concepts from a single image, in: SIGGRAPH Asia 2023 Conference Papers, 2023, pp. 1–12

    work page 2023

  18. [19]

    Safaee, A

    M. Safaee, A. Mikaeili, O. Patashnik, D. Cohen-Or, A. Mahdavi- Amiri, Clic: Concept learning in context, in: Proceedings of the IEEE/CVFConferenceonComputerVisionandPatternRecognition, 2024, pp. 6924–6933

    work page 2024

  19. [20]

    Y. Wang, C. Cao, J. Yu, K. Fan, X. Xue, Y. Fu, Towards enhanced image inpainting: Mitigating unwanted object insertion and preserv- ing color consistency., in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2025

    work page 2025

  20. [21]

    T. Yang, R. Wu, P. Ren, X. Xie, , L. Zhang, Pixel-aware stable diffu- sion for realistic image super-resolution and personalized stylization, in: The European Conference on Computer Vision (ECCV) 2024, 2023

    work page 2024

  21. [22]

    H. Guo, J. Li, T. Dai, Z. Ouyang, X. Ren, S.-T. Xia, Mambair: A simple baseline for image restoration with state-space model, in: European Conference on Computer Vision, Springer, 2024, pp. 222– 241

    work page 2024

  22. [23]

    A. F. Shamsuddin, K. Ragunathan, D. R. S. PM, P. Sankaran, et al., From synthetic to natural—single natural image dehazing deep net- works using synthetic dataset domain randomization, Journal of VisualCommunicationandImageRepresentation89(2022)103636

    work page 2022

  23. [24]

    J. Zuo, H. Deng, H. Zhou, J. Zhu, Y. Zhang, Y. Zhang, Y. Yan, K. Huang, W. Chen, Y. Deng, et al., Is nano banana pro a low-level vision all-rounder? a comprehensive evaluation on 14 tasks and 40 datasets, arXiv preprint arXiv:2512.15110 (2025)

    arXiv 2025

  24. [25]

    Y.Yang,X.Zeng,Z.Jiang,F.Yin,J.Liu,W.Cheng,S.Liu,Y.Peng, G. YU, S. Chen, et al., Realrestorer: Towards generalizable real- worldimagerestorationwithlarge-scaleimageeditingmodels, arXiv preprint arXiv:2603.25502 (2026)

    arXiv 2026

  25. [26]

    J. You, Y. Lin, B. Hu, Enhancing aesthetic image generation with re- inforcement learning guided prompt optimization in stable diffusion, Journal of Visual Communication and Image Representation (2025) 104641

    work page 2025

  26. [27]

    Cimpoi, S

    M. Cimpoi, S. Maji, I. Kokkinos, S. Mohamed, , A. Vedaldi, De- scribing textures in the wild, in: Proceedings of the IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2014

    work page 2014

  27. [28]

    S. Chen, A. Atapour-Abarghouei, H. P. H. Shum, Hint: High-quality inpainting transformer with mask-aware encoding and enhanced at- tention, IEEE Transactions on Multimedia 26 (2024) 7649–7660

    work page 2024

  28. [29]

    Jiang, Z

    Y. Jiang, Z. Zhang, T. Xue, J. Gu, Autodir: Automatic all-in-one image restoration with latent diffusion, in: European Conference on Computer Vision, Springer, 2024, pp. 340–359

    work page 2024

  29. [30]

    OpenAI, Introducing gpt image 1.5 — available today in the api and chatgpt, OpenAI Developer Community announcement, 2025.URL:https://openai.com/index/new-chatgpt-images-is-here/, released December 16, 2025

    work page 2025

  30. [31]

    G. Team, R. Anil, S. Borgeaud, J.-B. Alayrac, J. Yu, R. Soricut, J. Schalkwyk, A. M. Dai, A. Hauth, K. Millican, et al., Gemini: a family of highly capable multimodal models, arXiv preprint arXiv:2312.11805 (2023)

    Pith/arXiv arXiv 2023

  31. [32]

    Dockhorn, J

    B.F.Labs,S.Batifol,A.Blattmann,F.Boesel,S.Consul,C.Diagne, T. Dockhorn, J. English, Z. English, P. Esser, et al., Flux. 1 kontext: Flow matching for in-context image generation and editing in latent space, arXiv preprint arXiv:2506.15742 (2025)

    Pith/arXiv arXiv 2025

  32. [33]

    C. Wu, J. Li, J. Zhou, J. Lin, K. Gao, K. Yan, S.-m. Yin, S. Bai, X. Xu, Y. Chen, et al., Qwen-image technical report, arXiv preprint arXiv:2508.02324 (2025)

    Pith/arXiv arXiv 2025

  33. [34]

    S. I. Team, C. Qiao, C. Hui, C. Li, C. Wang, D. Song, J. Zhang, J. Li, Q. Xiang, R. Wang, et al., Firered-image-edit-1.0 technical report, arXiv preprint arXiv:2602.13344 (2026)

    arXiv 2026

  34. [35]

    S.Liu,Y.Han,P.Xing,F.Yin,R.Wang,W.Cheng,J.Liao,Y.Wang, H. Fu, C. Han, et al., Step1x-edit: A practical framework for general image editing, arXiv preprint arXiv:2504.17761 (2025)

    Pith/arXiv arXiv 2025

  35. [36]

    X. Lin, J. He, Z. Chen, Z. Lyu, B. Dai, F. Yu, Y. Qiao, W. Ouyang, C. Dong, Diffbir: Toward blind image restoration with generative diffusionprior,in:Europeanconferenceoncomputervision,Springer, 2024, pp. 430–448

    work page 2024

  36. [37]

    Chihaoui, A

    H. Chihaoui, A. Lemkhenter, P. Favaro, Blind image restoration via fast diffusion inversion, Advances in Neural Information Processing Systems 37 (2024) 34513–34532

    work page 2024

  37. [38]

    25669–25680

    F.Yu,J.Gu,Z.Li,J.Hu,X.Kong,X.Wang,J.He,Y.Qiao,C.Dong, Scaling up to excellence: Practicing model scaling for photo-realistic image restoration in the wild, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2024, pp. 25669–25680

    work page 2024

  38. [39]

    Y.Liu,J.He,Y.Liu,X.Lin,F.Yu,J.Hu,Y.Qiao,C.Dong, Adaptbir: Adaptiveblindimagerestorationwithlatentdiffusionpriorforhigher fidelity, Pattern Recognition 155 (2024) 110659

    work page 2024

  39. [40]

    X. Tang, X. Gu, X. He, X. Hu, J. Sun, Degradation-aware residual- conditioned optimal transport for unified image restoration, IEEE Transactions on Pattern Analysis and Machine Intelligence (2025)

    work page 2025

  40. [41]

    X. Chen, Z. Li, Y. Pu, Y. Liu, J. Zhou, Y. Qiao, C. Dong, A comparativestudyofimagerestorationnetworksforgeneralbackbone network design, in: European Conference on Computer Vision, Springer, 2024, pp. 74–91

    work page 2024

  41. [42]

    Y. Guo, Y. Gao, Y. Lu, H. Zhu, R. W. Liu, S. He, Onerestore: A universal restoration framework for composite degradation, in: European conference on computer vision, Springer, 2024, pp. 255– 272

    work page 2024

  42. [43]

    Eteke, A

    C. Eteke, A. Griessel, W. Kellerer, E. Steinbach, Bir-adapter: A parameter-efficientdiffusionadapterforblindimagerestoration, Pat- tern Recognition (2026) 113824

    work page 2026

  43. [44]

    Monga, H

    A. Monga, H. Nehete, T. K. R. Bollu, B. Raman, Dairnet: Degradation-aware all-in-one image restoration network with cross- channel feature interaction, Journal of Visual Communication and Image Representation (2025) 104659

    work page 2025

  44. [45]

    J. Jin, X. Qi, H. Dong, Q. Guan, G. Jin, L. Fan, Degradation removal anddetailrestorationdecompositionnetworkforsingleimagederain- ing, Journal of Visual Communication and Image Representation (2025) 104520

    work page 2025

  45. [46]

    2866–2878

    J.Lin,Z.Zhang,Y.Wei,D.Ren,D.Jiang,Q.Tian,W.Zuo,Improving imagerestorationthroughremovingdegradationsintextualrepresen- tations, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 2866–2878

    work page 2024

  46. [47]

    Beyond Pixels: Text Enhances Generalization in Real-World Image Restoration

    H.Sun,W.Li,J.Liu,K.Zhou,Y.Chen,Y.Guo,Y.Li,R.Pei,L.Peng, Y. Yang, Beyond pixels: Text enhances generalization in real-world image restoration, arXiv preprint arXiv:2412.00878 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  47. [48]

    M.Vijendran,J.Deng,S.Chen,E.S.L.Ho,H.P.H.Shum, Artificial intelligence for geometry-based feature extraction, analysis and syn- thesis in artistic images: A survey, Artificial Intelligence Review 58 (2024) 64

    work page 2024

  48. [49]

    Y. Fang, H. Zhang, H. Ji, X. Dai, C. Yao, Region-guided representa- tion fusion and background consistency in mask-free image editing, Journal of Visual Communication and Image Representation 119 (2026) 104852

    work page 2026

  49. [50]

    Q. Yang, D. Chen, Z. Tan, Q. Liu, Q. Chu, J. Bao, L. Yuan, G. Hua, N. Yu, Hq-50k: A large-scale, high-quality dataset for image restora- tion, 2023.arXiv:2306.05390

    Pith/arXiv arXiv 2023

  50. [51]

    Chheda-Kothary, R

    A. Chheda-Kothary, R. Kanchi, C. Sanders, K. Xiao, A. Sengupta, M. Kneitmix, J. O. Wobbrock, J. E. Froehlich, Artinsight: Enabling ai-poweredartworkengagementformixedvisual-abilityfamilies, in: Proceedings of the 30th International Conference on Intelligent User Interfaces, 2025, pp. 190–210

    work page 2025

  51. [52]

    OpenAI, Introducing 4o image generation (2025)

    work page 2025

  52. [53]

    Rombach, A

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, B. Ommer, High- resolution image synthesis with latent diffusion models, 2021. arXiv:2112.10752

    Pith/arXiv arXiv 2021

  53. [54]

    X. Lin, F. Yu, J. Hu, Z. You, W. Shi, J. S. Ren, J. Gu, C. Dong, Harnessingdiffusion-yieldedscorepriorsforimagerestoration, ACM Transactions on Graphics (TOG) 44 (2025) 1–21. :Preprint submitted to Elsevier Page 14 of 15 MDTD-ArtIR

    work page 2025

  54. [55]

    Z. Wang, A. C. Bovik, H. R. Sheikh, E. P. Simoncelli, Image qualityassessment:fromerrorvisibilitytostructuralsimilarity, IEEE transactions on image processing 13 (2004) 600–612

    work page 2004

  55. [56]

    Zhang, P

    R. Zhang, P. Isola, A. A. Efros, E. Shechtman, O. Wang, The unreasonableeffectivenessofdeepfeaturesasaperceptualmetric, in: Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 586–595

    work page 2018

  56. [57]

    URL:https:// huggingface.co/datasets/huggan/wikiart

    huggan, WikiArt Dataset, Hugging Face, 2022. URL:https:// huggingface.co/datasets/huggan/wikiart

    work page 2022

  57. [58]

    Saleh, A

    B. Saleh, A. Elgammal, Large-scale classification of fine-art paint- ings: Learning the right metric on the right feature, arXiv preprint arXiv:1505.00855 (2015)

    Pith/arXiv arXiv 2015

  58. [59]

    URL:https: //huggingface.co/diffusers/FLUX.2-dev-bnb-4bit

    diffusers, FLUX.2-dev-bnb-4bit, Hugging Face, 2025. URL:https: //huggingface.co/diffusers/FLUX.2-dev-bnb-4bit

    work page 2025

  59. [60]

    T.Dettmers,A.Pagnoni,A.Holtzman,L.Zettlemoyer, QLoRA:Effi- cientfinetuningofquantizedLLMs, AdvancesinNeuralInformation Processing Systems 36 (2023)

    work page 2023

  60. [61]

    URL:https://www.kaggle.com/datasets/ pes1ug22am047/damaged-and-undamaged-artworks, dataset

    pes1ug22am047,DAMAGED_AND_UNDAMAGED_ARTWORKS, Kaggle, 2024. URL:https://www.kaggle.com/datasets/ pes1ug22am047/damaged-and-undamaged-artworks, dataset

    work page 2024

  61. [62]

    colabsss, Digital Restoration of Heritage Paintings, Kaggle,

    work page

  62. [2024]

    :Preprint submitted to Elsevier Page 15 of 15

    URL:https://www.kaggle.com/datasets/colabsss/ digital-restoration-of-heritage-paintings, dataset. :Preprint submitted to Elsevier Page 15 of 15

    work page