Pith. sign in

REVIEW 4 major objections 5 minor 3 cited by

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

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read The paper claims that supplying diffusion-based image-restoration models with text captions whose length matches the degradation level—generated by its Res-Captioner module—reactivates their generative priors on out-of-distribution…

desk verdict Useful plug-and-play captioner for diffusion-based restoration, with honest experiments and a real benchmark; the main caveat is that the SUPIR-trained length policy's cross-model transfer is shown empirically but not yet explained. read the letter →

arxiv 2412.00878 v2 pith:WVDJ5KL6 submitted 2024-12-01 cs.CV

Haoze Sun , Wenbo Li , Jiayue Liu , Kaiwen Zhou , Yongqiang Chen , Yong Guo , Yanwei Li , Renjing Pei
show 2 more authors
Long Peng Yujiu Yang
This is my paper · ORCID
classification cs.CV
keywords imagerestorationtext-guideddiffusionmodelsgeneralizationout-of-distributioncaptioninggenerativecapabilitydeactivationplug-and-playreal-worldbenchmark
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

Diffusion-based image restoration models, when shown out-of-distribution real-world photos, often revert to blurred or featureless outputs because their text-conditioned generative priors stop engaging—the paper names this 'generative capability deactivation.' The authors argue that text can serve as an auxiliary invariant representation that compensates: descriptions that are rich enough, relevant enough, and free of blur-inducing terms re-trigger the generative prior even when the image itself is badly degraded. They support this with controlled observations on two text properties, richness and relevance, then build Res-Captioner, a captioner that predicts the right caption length for each image and writes a detailed description of it. Inserted between any low-quality image and a pretrained diffusion restorer, Res-Captioner improves fidelity and detail on a new RealIR benchmark and on existing real-world sets without retraining the restorer. The paper's bet is that a good text description is a stable, degradation-invariant channel that can rescue restoration when pixel evidence fails.

What carries the argument

The carrying object is Res-Captioner, a restoration-specific image captioner built by fine-tuning LLaVA-1.5 with LoRA. It has three distinctive parts: a Chain-of-Thought prompting scheme that first predicts the optimal caption length in tokens and then generates the caption; a degradation-aware visual encoder that compresses a pretrained degradation extractor's signal into tokens and injects them to keep length decisions sensitive to how badly the image is degraded; and a training set of 5,500 low-quality-image and caption pairs curated by having human annotators pick, among GPT-4 captions of varying length, the one that gave the best SUPIR restoration. The framework treats restoration as dual-conditioned generation: the low-quality image remains the primary condition, and the caption supplies an auxiliary invariant channel that becomes more important as degradation grows. The mechanism the whole system tests is that caption length controls texture richness, caption relevance controls fidelity, and the right length is proportional to how far the test degradation sits from the training distribution.

What would settle it

One decisive experiment would be to take a restoration model whose backbone was never used in training, run Res-Captioner on a held-out set of genuinely new camera degradations, and compare its outputs with the model's default caption under DISTS and LPIPS; if there is no consistent improvement, or if the predicted token lengths no longer track the manually annotated optimal lengths on those images, the transfer claim collapses.

Watch

Extended reading notes

Core claim

The paper's central claim is that the failure of diffusion-based restorers on real-world images is not primarily a limit of the generative backbone but a failure to supply the right text condition, and that this failure is fixable at the input level. It establishes that longer captions enrich restored textures up to a point, that the optimal caption length grows with the gap between training and test degradation, that captions irrelevant to the high-quality content lower fidelity, and that words about blur, focus, and bokeh actively blur outputs. From these observations it derives Res-Captioner, which predicts an image- and degradation-dependent token count and then writes a detailed, content-focused caption; the authors show that plugging these captions into StableSR, SUPIR, CoSeR, and SeeSR improves no-reference quality scores on RealIR and lowers DISTS and LPIPS on multi-degradation pairs, with larger gains under heavier degradation.

Load-bearing premise

The load-bearing premise is that the optimal caption richness and content, as judged by human annotators on SUPIR-produced restorations, transfer to other restoration backbones and to degradation types that never appeared in training; if the learned length rule is really tuned to SUPIR's behavior, the plug-and-play gains would not hold.

Editorial extensions

If this is right

  • Adding Res-Captioner before StableSR and SUPIR improves every no-reference quality metric reported on the RealIR benchmark, with SUPIR's LIQE gain reaching about 36% on camera-captured images.
  • The fidelity gain grows with degradation severity: heavy-degradation DISTS improves by about 8.6% for StableSR and 13.3% for SUPIR relative to the unmodified model.
  • Caption length should be adaptive: too-short captions leave the generative prior dormant, while too-long captions introduce artifacts in non-rigid structures such as faces, text, and crowds.
  • Captions that mention blur, focus, or bokeh impair restoration even when the description is accurate, so restoration prompts must be filtered for such terms.
  • The plug-and-play captioner transfers across restoration architectures such as StableSR, SUPIR, CoSeR, and SeeSR, and a small fine-tuning set of about 200 pairs adapts it to a new text-to-image backbone.

Reading between the lines

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

  • Inference: if caption length is compensating for a measurable drop in information from the degraded pixels, the same auxiliary-text-as-invariant-condition idea should transfer to other conditional generative tasks—video restoration, denoising, or text-to-speech under noisy input—where the conditioning signal degrades but a language description of the desired content does not.
  • Inference: the human-annotation step over SUPIR candidates means the training set encodes one backbone's generative preferences; a model-agnostic captioner would likely need per-backbone length calibration, and the paper's own 200-pair fine-tuning experiments suggest such calibration is cheap.
  • Inference: a sharper quantitative version of the paper's core observation would measure the optimal caption length against an information-theoretic proxy for degradation severity, for example the drop in CLIP similarity between low-quality and high-quality images, turning the learned length predictor into a testable regression.
  • Inference: because harmful descriptions such as blur, bokeh, and shallow depth of field degrade outputs even when accurate, restoration captioning may need a forbidden-terms filter that is checked against the output of any future captioner, not just Res-Captioner.
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 / 5 minor

Summary. This paper argues that text descriptions can act as an auxiliary invariant representation for diffusion-based image restoration models. It identifies two properties of text input, richness (roughly length) and relevance, shows empirically that longer and more relevant captions improve restoration quality up to a point, and that degradation- or photography-related descriptions are harmful. It then introduces Res-Captioner, an LLaVA-1.5-based captioner fine-tuned with LoRA, which predicts an adaptive caption length via chain-of-thought before generating a detailed caption, conditioned on a degradation-aware visual encoder. Training pairs are produced by generating multiple GPT-4 captions of varying length for HQ images, producing LQ versions with LDM degradation models, restoring them with SUPIR, and having human annotators select the caption whose restoration looks best. The method is evaluated on a newly collected RealIR benchmark (152 camera images and 53 internet images), a synthetic multi-degradation test set, RealSR, and DRealSR, across SUPIR, StableSR, and CoSeR. The main claim is that Res-Captioner is a fully plug-and-play module that improves generalization without retraining.

Significance. If the central claim holds, the paper would make a useful and practical contribution: the ability to improve multiple pretrained restoration models by simply replacing the captioner is valuable, and the observations about text richness, relevance, and harmful descriptions give concrete design guidance for text-guided restoration. The paper also introduces a new benchmark and reports a user study. The strengths are the clear experimental framing across several restoration backbones and the explicit ablation of richness, relevance, and harmful descriptions. However, the load-bearing transfer claim—that a caption-length policy optimized on SUPIR transfers to other models—is not directly established, and the statistical grounding of the quantitative results is thin. The contribution is therefore promising but not yet fully substantiated.

major comments (4)
  1. [Section 2.2 / Table B.5] The core plug-and-play claim is that a caption-length policy learned from human selections over SUPIR-generated restorations transfers without retraining to StableSR, SeeSR, and CoSeR. This transfer is asserted but not directly tested: the training data are generated exclusively with SUPIR candidates, and the only cross-model evidence is that the captioner improves numbers in Tables 2 and C.8. Those gains could arise from a generic "more detailed captions mitigate OOD degradation" effect rather than from a model-invariant adaptive policy. The supplementary fine-tuning experiment (Table B.5) shows that fine-tuning on 200 StableSR pairs improves StableSR results over the unfine-tuned captioner, which implies the SUPIR-trained policy is not already optimal for StableSR. The authors should provide a direct test: collect human annotations of optimal caption length for StableSR (and ideally SeeSR) on a small set of LQ images and compare the policy learned from those annotations with the SUPIR-trained policy, or otherwise demonstrate model invariance using a controlled analysis that removes the generic length effect.
  2. [Section 3.2 / Table 2] The multi-degradation test set is generated using pre-trained latent diffusion models (LDMs), and the training LQ images are also generated with five LDM-based degradation models. Although the test set uses held-out zoom ratios, it still shares the same degradation simulation family, so it is closer to the training distribution of the caption-length policy than the RealIR benchmark. This weakens the generalization conclusion drawn from Table 2. The authors should report results on a test set generated by a different degradation pipeline (for example Real-ESRGAN-style degradation) and should report per-condition variance or significance of the differences in Table 2; currently all entries are single runs, and several StableSR improvements (e.g., 2.4% DISTS under light degradation) are small relative to plausible run-to-run variability.
  3. [Section 3.3.1 / Table 1] All quantitative claims on the RealIR benchmark are based exclusively on non-reference metrics (MUSIQ, MANIQA, LIQE, NIQE, CLIP-IQA) with no confidence intervals or significance tests. The reported improvements, e.g., StableSR MUSIQ from 66.15 to 69.28 on cameras, are plausible but not statistically established; non-reference metrics are known to be noisy on out-of-distribution content. The user study (Figure 7) is reported only as mean scores without per-item variance or paired test results. Please report error bars over multiple evaluation runs or bootstrapped samples, and provide per-participant or per-image statistics for the user study to support the claim of "significant improvements".
  4. [Section 2.1.1 / Observation 2] Observation 2, which motivates adaptive length prediction, is supported by Figure 4(c,d) only qualitatively and on two models. The paper should quantify the claimed proportional relationship between domain gap (e.g., zoom ratio) and optimal text richness, and it should report the agreement between the human annotators used to construct the training pairs in Section 2.2. Without annotator-agreement data, the supervision signal for the length predictor is itself uncharacterized.
minor comments (5)
  1. [Throughout] The text is not typeset consistently: "LLaV A-1.5" appears with broken spacing in the main text, and "Chain-of-Thought" is hyphenated inconsistently.
  2. [Section 3.1] The implementation details report only batch size, steps, and learning rate; please specify the LoRA rank and alpha, the exact LLaVA-1.5 checkpoint, and the number of human annotators and their agreement for the training-data selection in Section 2.2.
  3. [Section 2.2] The statement "we leverage five pre-trained latent diffusion models (LDM) to generate LQ images" defers the key configuration to the appendix; consider moving the degradation types, zoom ratio ranges, and the held-out zoom ratios to the main text, since they are essential for assessing the multi-degradation evaluation.
  4. [Section 2.2 / Figure 5] The prompt in the supplementary material asks GPT-4 for captions of about XXX words, while the model is trained to predict token lengths; the relationship between word counts and token counts should be clarified.
  5. [Section 3.2 / RealIR] The RealIR benchmark is announced as "will be made publicly available," but no release link, license, or access procedure is provided; this should be completed before publication.

Circularity Check

1 steps flagged · score 3.0 of 10

SUPIR evaluation is partly an in-domain fit because Res-Captioner's training labels are human selections over SUPIR restorations; the cross-model plug-and-play results supply independent evidence and keep the central claim non-circular.

  1. fitted input called prediction [Section 2.2, Training data generation (Figure 5a), evaluated in Tables 1-2 and Table C.8]
    "These descriptions are fed into the restoration model, producing multiple restored candidates for each LQ image. Human annotators select the optimal text input that provides the best visual result, balancing texture richness and fidelity. In total, we curate 5,500 LQ image-caption pairs for training our Res-Captioner. It is important to note that although we used a specific restoration model, SUPIR [68], in our training data collection pipeline, Res-Captioner performs effectively across other restoration models in a plug-and-play manner."

    Res-Captioner's supervision is defined by SUPIR restorations: human annotators pick, among GPT-4 captions of different lengths, the one whose SUPIR output looks best. Training the captioner to reproduce those choices and then reporting 'SUPIR w/ Ours' gains on unseen LQ images is an in-domain fit to SUPIR's prompt-length sensitivity, not an out-of-sample test of a model-invariant text policy. The fitted target (optimal caption for SUPIR) and the evaluation target (SUPIR restoration quality) coincide in model and in selection criterion, so the SUPIR-specific improvement is partly forced by construction. The circularity is only partial: the same captioner improves StableSR, SeeSR, and CoSeR without retraining, which provides independent evidence beyond the SUPIR fit.

full rationale

The only concrete circular step is the SUPIR-specific evaluation: training labels are human choices among SUPIR-generated candidates, so the Table 2 SUPIR gains partly reflect fitting to SUPIR's behavior rather than a purely external test. However, the paper's central claim is plug-and-play generalization across restoration models, and that claim is supported by genuinely independent results: the SUPIR-trained captioner is applied without retraining to StableSR (Table 2), SeeSR (Figure 6), and CoSeR (Table C.8), none of which generated the training labels. These results are not obtained through a self-citation chain, and no load-bearing uniqueness theorem or ansatz is imported solely from the authors' prior work. Table B.5 does show that 200 StableSR fine-tuning pairs further improve StableSR, which weakens the phrase 'fully plug-and-play' but is a transfer limitation rather than a circular derivation. Overall, the central contribution has independent empirical content; only the SUPIR benchmark result is partially self-consistent by construction.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central method relies on several domain assumptions: that caption length is a valid proxy for text richness, that synthetic LDM degradations represent real-world degradations, that perceptual metrics used for RealIR are reliable, and that human annotations define the optimal caption. The only hand-tuned parameters are the degradation token count N, the caption length set, the zoom ratio ranges, and the offset formula constants; no physical constants or invented entities are introduced.

free parameters (4)
  • Degradation encoder token count N = 36
    Chosen based on Table C.9 as a tradeoff between offset level E and computational cost; this is a hand-picked hyperparameter affecting caption length prediction accuracy.
  • Caption length levels = 80, 110, 140, 200, 260, 350, 440 words
    Discrete set of lengths used to generate training captions with GPT-4; the intervals increase progressively because longer captions cause smaller texture changes.
  • Zoom ratio ranges for degradation levels = 3-7 (light), 8-10 (moderate), 15-20 (heavy)
    Used to categorize the multi-degradation test set; chosen ad hoc to represent increasing domain shift.
  • Offset level formula constants = threshold 15, denominator 30
    Used in E = max(|Lo-L|-15,0)/30 to measure caption length prediction error; these constants are arbitrary choices.
assumptions (5)
  • domain assumption Text richness is equivalent to caption length in tokens
    The paper equates richness with text length (Section 2.1.1) and uses token count as the control variable in experiments.
  • domain assumption LDM-generated LQ images with varying zoom ratios simulate real-world device degradations
    Training data and the multi-degradation test set rely on LDMs conditioned on HQ images plus zoom ratios to emulate camera zoom and device degradation (Section B.2).
  • domain assumption The non-reference metrics (MUSIQ, MANIQA, LIQE, NIQE, CLIP-IQA) are valid proxies for perceptual quality
    The paper uses these for the RealIR benchmark without ground truth and states that pixel-level metrics are ignored because they correlate weakly with perception (Section 3.2).
  • domain assumption Human annotators' selection of the best restoration result is the ground truth for optimal caption content and length
    Training data generation uses human annotators to choose the text input that yields the best visual result (Section 2.2, Figure 5a).
  • domain assumption The richness and relevance properties observed on SUPIR and StableSR generalize to other T2I diffusion models
    Res-Captioner is trained with SUPIR but claimed to be plug-and-play for StableSR, SeeSR, CoSeR, and PromptFix (Section 2.2, Appendix C).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Beyond Pixels: Text Enhances Generalization in Real-World Image Restoration." pith.science (2026). https://pith.science/paper/WVDJ5KL6

@misc{pith2026241200878,
  author       = {Pith},
  title        = {Pith review of: Beyond Pixels: Text Enhances Generalization in Real-World Image Restoration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WVDJ5KL6}},
  note         = {Machine review of arXiv:2412.00878}
}
read the original abstract

Generalization has long been a central challenge in real-world image restoration. While recent diffusion-based restoration methods, which leverage generative priors from text-to-image models, have made progress in recovering more realistic details, they still encounter "generative capability deactivation" when applied to out-of-distribution real-world data. To address this, we propose using text as an auxiliary invariant representation to reactivate the generative capabilities of these models. We begin by identifying two key properties of text input: richness and relevance, and examine their respective influence on model performance. Building on these insights, we introduce Res-Captioner, a module that generates enhanced textual descriptions tailored to image content and degradation levels, effectively mitigating response failures. Additionally, we present RealIR, a new benchmark designed to capture diverse real-world scenarios. Extensive experiments demonstrate that Res-Captioner significantly enhances the generalization abilities of diffusion-based restoration models, while remaining fully plug-and-play.

Figures

Figures reproduced from arXiv: 2412.00878 by the authors.

Figure 1
Figure 1. State-of-the-art methods like SUPIR [68] are limited in utilizing their full generative capacity, often yielding blurred or otherwise unsatisfactory results on out-of-distribution (OOD) data, a phenomenon we term as “generative capability deactivation”. Our Res-captioner can reactivate their generative capabilities by providing detailed and accurate descriptions. Abstract Generalization has long been a central chall… view at source ↗
Figure 2
Figure 2. Visualization of the text richness property. ( [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Visualization and demonstration of the text relevance property. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Demonstration of the richness property. (a, b): There is a positive correlation between text richness and the richness of textures in the restored results. (c, d): The optimal text richness (indicated by an asterisk) is proportional to the degree of deviation between t…
Figure 5
Figure 5. Figure 5: (a) The generation and annotation process of our training data. (b) Chain-of-Thought captioning of our Res-Captioner. (c) Network structure of our Res-Captioner. 2.2. Adaptive Caption Learning As discussed in Section 2.1, text is crucial in controlling both the richnes…
Figure 6
Figure 6. Figure 6: Qualitative comparisons on in-the-wild images. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: User study. Given the relatively simple and light degradation in Re￾alSR and DRealSR, we use SUPIR as the reference model for evaluation. Our Res-Captioner significantly improves the performance of SUPIR in [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

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

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

    cs.CV 2026-08 conditional novelty 6.0 of 10

    A new benchmark (MDTD-Art) shows image editing models generally beat dedicated restoration models on art images degraded by textured semi-transparent overlays.

  2. TurboVSR: Fantastic Video Upscalers and Where to Find Them

    cs.CV 2025-06 conditional novelty 6.0 of 10

    TurboVSR uses a high-compression video autoencoder with factorized conditioning and non-uniform shortcut sampling to achieve near-state-of-the-art perceptual video super-resolution at roughly 100x lower compute cost.

  3. Enhancing Zero-Shot Brain Tumor Subtype Classification via Fine-Grained Patch-Text Alignment

    cs.CV 2025-08 unverdicted novelty 4.0 of 10

    FG-PAN improves zero-shot brain tumor subtype classification by aligning refined visual patch features with LLM-generated fine-grained text prototypes.

Reference graph

Works this paper leans on

76 extracted references · 52 canonical work pages · cited by 3 Pith papers

  1. [1]

    Multimodal prompt perceiver: Empower adap- tiveness generalizability and fidelity for all-in-one image restoration

    Yuang Ai, Huaibo Huang, Xiaoqiang Zhou, Jiexiang Wang, and Ran He. Multimodal prompt perceiver: Empower adap- tiveness generalizability and fidelity for all-in-one image restoration. In CVPR, pages 25432–25444, 2024. 2, 9

    work page 2024

  2. [2]

    Invariant risk minimization

    Martin Arjovsky, L ´eon Bottou, Ishaan Gulrajani, and David Lopez-Paz. Invariant risk minimization. arXiv preprint arXiv:1907.02893, 2019. 2

    arXiv 1907

  3. [3]

    Spatext: Spatio-textual representation for con- trollable image generation

    Omri Avrahami, Thomas Hayes, Oran Gafni, Sonal Gupta, Yaniv Taigman, Devi Parikh, Dani Lischinski, Ohad Fried, and Xi Yin. Spatext: Spatio-textual representation for con- trollable image generation. In CVPR, pages 18370–18380,

  4. [4]

    Improving image generation with better captions

    James Betker, Gabriel Goh, Li Jing, Tim Brooks, Jianfeng Wang, Linjie Li, Long Ouyang, Juntang Zhuang, Joyce Lee, Yufei Guo, et al. Improving image generation with better captions. Computer Science. https://cdn. openai. com/papers/dall-e-3. pdf, 2(3):8, 2023. 3

    2023

  5. [5]

    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. In ICCV, pages 3086– 3095, 2019. 2, 5, 6

    work page 2019

  6. [6]

    Real-world blind super-resolution via feature matching with implicit high- resolution priors

    Chaofeng Chen, Xinyu Shi, Yipeng Qin, Xiaoming Li, Xi- aoguang Han, Tao Yang, and Shihui Guo. Real-world blind super-resolution via feature matching with implicit high- resolution priors. In ACMMM, pages 1329–1338, 2022. 2

    work page 2022

  7. [7]

    Low-res leads the way: Improving generalization for super-resolution by self-supervised learning

    Haoyu Chen, Wenbo Li, Jinjin Gu, Jingjing Ren, Haoze Sun, Xueyi Zou, Zhensong Zhang, Youliang Yan, and Lei Zhu. Low-res leads the way: Improving generalization for super-resolution by self-supervised learning. InCVPR, pages 25857–25867, 2024. 2, 5, 9

    work page 2024

  8. [8]

    Sharegpt4v: Improving large multi-modal models with better captions

    Lin Chen, Jisong Li, Xiaoyi Dong, Pan Zhang, Conghui He, Jiaqi Wang, Feng Zhao, and Dahua Lin. Sharegpt4v: Improving large multi-modal models with better captions. arXiv preprint arXiv:2311.12793, 2023. 2, 5, 9, 12

    arXiv 2023

Show all 76 references
  1. [9]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In CVPR, pages 248–255, 2009. 5, 9

    2009

  2. [10]

    Diffusion mod- els beat gans on image synthesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion mod- els beat gans on image synthesis. NeurIPS, 34:8780–8794,

  3. [11]

    Image quality assessment: Unifying structure and texture similarity

    Keyan Ding, Kede Ma, Shiqi Wang, and Eero P Simoncelli. Image quality assessment: Unifying structure and texture similarity. TPAMI, 44(5):2567–2581, 2020. 6

    2020

  4. [12]

    Learning in- variant representation for unsupervised image restoration

    Wenchao Du, Hu Chen, and Hongyu Yang. Learning in- variant representation for unsupervised image restoration. In CVPR, pages 14483–14492, 2020. 9

    2020

  5. [13]

    Generative dif- fusion prior for unified image restoration and enhancement

    Ben Fei, Zhaoyang Lyu, Liang Pan, Junzhe Zhang, Weidong Yang, Tianyue Luo, Bo Zhang, and Bo Dai. Generative dif- fusion prior for unified image restoration and enhancement. In CVPR, pages 9935–9946, 2023. 9

    2023

  6. [14]

    Text-guided explorable image super-resolution

    Kanchana Vaishnavi Gandikota and Paramanand Chan- dramouli. Text-guided explorable image super-resolution. In CVPR, pages 25900–25911, 2024. 2

    2024

  7. [15]

    Optimizing prompts for text-to-image generation

    Yaru Hao, Zewen Chi, Li Dong, and Furu Wei. Optimizing prompts for text-to-image generation. NeurIPS, 36, 2024. 5

    2024

  8. [16]

    Lora: Low-rank adaptation of large language models

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen- Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685, 2021. 5

    2021 arXiv

  9. [17]

    Ella: Equip diffusion models with llm for enhanced semantic alignment

    Xiwei Hu, Rui Wang, Yixiao Fang, Bin Fu, Pei Cheng, and Gang Yu. Ella: Equip diffusion models with llm for enhanced semantic alignment. arXiv preprint arXiv:2403.05135, 2024. 5

    2024 arXiv

  10. [18]

    Denoising diffusion restoration models

    Bahjat Kawar, Michael Elad, Stefano Ermon, and Jiaming Song. Denoising diffusion restoration models. NeurIPS, 35: 23593–23606, 2022. 9

    2022

  11. [19]

    Musiq: Multi-scale image quality transformer

    Junjie Ke, Qifei Wang, Yilin Wang, Peyman Milanfar, and Feng Yang. Musiq: Multi-scale image quality transformer. In ICCV, pages 5148–5157, 2021. 3, 6

    2021

  12. [20]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 ,

    arXiv

  13. [21]

    Segment any- thing

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer White- head, Alexander C Berg, Wan-Yen Lo, et al. Segment any- thing. In ICCV, pages 4015–4026, 2023. 5

    2023

  14. [22]

    Wilds: A benchmark of in-the- wild distribution shifts

    Pang Wei Koh, Shiori Sagawa, Henrik Marklund, Sang Michael Xie, Marvin Zhang, Akshay Balsubra- mani, Weihua Hu, Michihiro Yasunaga, Richard Lanas Phillips, Irena Gao, et al. Wilds: A benchmark of in-the- wild distribution shifts. In ICML, pages 5637–5664. PMLR,

  15. [23]

    Reflash dropout in image super-resolution

    Xiangtao Kong, Xina Liu, Jinjin Gu, Yu Qiao, and Chao Dong. Reflash dropout in image super-resolution. In CVPR, pages 6002–6012, 2022. 2

    2022

  16. [24]

    Benchmarking single- image dehazing and beyond

    Boyi Li, Wenqi Ren, Dengpan Fu, Dacheng Tao, Dan Feng, Wenjun Zeng, and Zhangyang Wang. Benchmarking single- image dehazing and beyond. TIP, 28(1):492–505, 2018. 12

    2018

  17. [25]

    Invariant information bot- tleneck for domain generalization

    Bo Li, Yifei Shen, Yezhen Wang, Wenzhen Zhu, Dongsheng Li, Kurt Keutzer, and Han Zhao. Invariant information bot- tleneck for domain generalization. In AAAI, pages 7399– 7407, 2022. 2

    2022

  18. [26]

    Learning degradation rep- resentations for image deblurring

    Dasong Li, Yi Zhang, Ka Chun Cheung, Xiaogang Wang, Hongwei Qin, and Hongsheng Li. Learning degradation rep- resentations for image deblurring. In ECCV, pages 736–753. Springer, 2022. 2, 9

    2022

  19. [27]

    Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models

    Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. In ICML, pages 19730–19742. PMLR, 2023. 9

    2023

  20. [28]

    Best-buddy gans for highly detailed image super-resolution

    Wenbo Li, Kun Zhou, Lu Qi, Liying Lu, and Jiangbo Lu. Best-buddy gans for highly detailed image super-resolution. In AAAI, pages 1412–1420, 2022. 2

    2022

  21. [29]

    Learning distortion invariant representation for im- age restoration from a causality perspective

    Xin Li, Bingchen Li, Xin Jin, Cuiling Lan, and Zhibo Chen. Learning distortion invariant representation for im- age restoration from a causality perspective. In CVPR, pages 1714–1724, 2023. 9

    2023

  22. [30]

    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. In ICCVW, pages 1833–1844, 2021. 2

    2021

  23. [31]

    Efficient and degradation-adaptive network for real-world image super- resolution

    Jie Liang, Hui Zeng, and Lei Zhang. Efficient and degradation-adaptive network for real-world image super- resolution. In ECCV, pages 574–591. Springer, 2022. 2, 6

    2022

  24. [32]

    Details or artifacts: A locally discriminative learning approach to realistic image super-resolution

    Jie Liang, Hui Zeng, and Lei Zhang. Details or artifacts: A locally discriminative learning approach to realistic image super-resolution. In CVPR, pages 5657–5666, 2022. 4

    2022

  25. [33]

    Improving image restoration through removing degradations in textual repre- sentations

    Jingbo Lin, Zhilu Zhang, Yuxiang Wei, Dongwei Ren, Dong- sheng Jiang, Qi Tian, and Wangmeng Zuo. Improving image restoration through removing degradations in textual repre- sentations. In CVPR, pages 2866–2878, 2024. 4

    2024

  26. [34]

    Diff- bir: Towards blind image restoration with generative diffu- sion prior

    Xinqi Lin, Jingwen He, Ziyan Chen, Zhaoyang Lyu, Bo Dai, Fanghua Yu, Wanli Ouyang, Yu Qiao, and Chao Dong. Diff- bir: Towards blind image restoration with generative diffu- sion prior. arXiv preprint arXiv:2308.15070, 2023. 2, 9

    2023 arXiv

  27. [35]

    Blind image super-resolution: A survey and beyond.TPAMI, 45(5):5461–5480, 2022

    Anran Liu, Yihao Liu, Jinjin Gu, Yu Qiao, and Chao Dong. Blind image super-resolution: A survey and beyond.TPAMI, 45(5):5461–5480, 2022. 2

    2022

  28. [36]

    Accelerating diffusion models for inverse problems through shortcut sampling

    Gongye Liu, Haoze Sun, Jiayi Li, Fei Yin, and Yujiu Yang. Accelerating diffusion models for inverse problems through shortcut sampling. arXiv preprint arXiv:2305.16965, 2023. 9

    2023 arXiv

  29. [37]

    Improved baselines with visual instruction tuning

    Haotian Liu, Chunyuan Li, Yuheng Li, and Yong Jae Lee. Improved baselines with visual instruction tuning. In CVPR, pages 26296–26306, 2024. 2, 5, 9, 12

    2024

  30. [38]

    Visual instruction tuning

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning. NeurIPS, 36, 2024. 2, 5, 9

    2024

  31. [39]

    Degae: A new pretraining paradigm for low-level vision

    Yihao Liu, Jingwen He, Jinjin Gu, Xiangtao Kong, Yu Qiao, and Chao Dong. Degae: A new pretraining paradigm for low-level vision. In CVPR, pages 23292–23303, 2023. 5

    2023

  32. [40]

    Evaluating the generalization ability of super- resolution networks

    Yihao Liu, Hengyuan Zhao, Jinjin Gu, Yu Qiao, and Chao Dong. Evaluating the generalization ability of super- resolution networks. TPAMI, 2023. 2

    2023

  33. [41]

    Evaluating the generalization ability of super- resolution networks

    Yihao Liu, Hengyuan Zhao, Jinjin Gu, Yu Qiao, and Chao Dong. Evaluating the generalization ability of super- resolution networks. TPAMI, 2023. 9

    2023

  34. [42]

    Desnownet: Context-aware deep network for snow removal

    Yun-Fu Liu, Da-Wei Jaw, Shih-Chia Huang, and Jenq-Neng Hwang. Desnownet: Context-aware deep network for snow removal. TIP, 27(6):3064–3073, 2018. 12

    2018

  35. [43]

    Controlling vision-language models for universal image restoration

    Ziwei Luo, Fredrik K Gustafsson, Zheng Zhao, Jens Sj¨olund, and Thomas B Sch¨on. Controlling vision-language models for universal image restoration. arXiv preprint arXiv:2310.01018, 2023. 2

    2023 arXiv

  36. [44]

    Photo-realistic image restoration in the wild with controlled vision-language models

    Ziwei Luo, Fredrik K Gustafsson, Zheng Zhao, Jens Sj¨olund, and Thomas B Sch ¨on. Photo-realistic image restoration in the wild with controlled vision-language models. arXiv preprint arXiv:2404.09732, 2024. 2

    2024 arXiv

  37. [45]

    Domain invariant representation learning with do- main density transformations

    A Tuan Nguyen, Toan Tran, Yarin Gal, and Atilim Gunes Baydin. Domain invariant representation learning with do- main density transformations. NeurIPS, 34:5264–5275,

  38. [46]

    Sdxl: Improving latent diffusion models for high-resolution image synthesis

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas M ¨uller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion models for high-resolution image synthesis. In ICLR, 2024. 2, 7

    2024

  39. [47]

    Learn- ing transferable visual models from natural language super- vision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learn- ing transferable visual models from natural language super- vision. In ICML, pages 8748–8763. PMLR, 2021. 3

    2021

  40. [48]

    Moe-diffir: Task-customized diffusion priors for universal compressed image restoration

    Yulin Ren, Xin Li, Bingchen Li, Xingrui Wang, Mengxi Guo, Shijie Zhao, Li Zhang, and Zhibo Chen. Moe-diffir: Task-customized diffusion priors for universal compressed image restoration. In ECCV, pages 116–134. Springer, 2025. 2

    2025

  41. [49]

    High-resolution image syn- thesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj¨orn Ommer. High-resolution image syn- thesis with latent diffusion models. In CVPR, pages 10684– 10695, 2022. 2, 5, 9

    2022

  42. [50]

    Coser: Bridging image and language for cognitive super-resolution

    Haoze Sun, Wenbo Li, Jianzhuang Liu, Haoyu Chen, Ren- jing Pei, Xueyi Zou, Youliang Yan, and Yujiu Yang. Coser: Bridging image and language for cognitive super-resolution. In CVPR, pages 25868–25878, 2024. 2, 4, 6, 9, 12, 13

    2024

  43. [51]

    Explore image deblurring via encoded blur kernel space

    Phong Tran, Anh Tuan Tran, Quynh Phung, and Minh Hoai. Explore image deblurring via encoded blur kernel space. In CVPR, pages 11956–11965, 2021. 2, 9

    2021

  44. [52]

    Unsplash dataset

    Unsplash. Unsplash dataset. 5

  45. [53]

    Navigating beyond dropout: An intriguing solution to- wards generalizable image super resolution

    Hongjun Wang, Jiyuan Chen, Yinqiang Zheng, and Tieyong Zeng. Navigating beyond dropout: An intriguing solution to- wards generalizable image super resolution. In CVPR, pages 25532–25543, 2024. 2

    2024

  46. [54]

    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, pages 2555–2563, 2023. 6

    2023

  47. [55]

    Exploiting diffusion prior for real-world image super-resolution

    Jianyi Wang, Zongsheng Yue, Shangchen Zhou, Kelvin CK Chan, and Chen Change Loy. Exploiting diffusion prior for real-world image super-resolution. IJCV, pages 1–21, 2024. 2, 3, 6, 9

    2024

  48. [56]

    Self-supervised deep image restoration via adaptive stochastic gradient langevin dynam- ics

    Weixi Wang, Ji Li, and Hui Ji. Self-supervised deep image restoration via adaptive stochastic gradient langevin dynam- ics. In CVPR, pages 1989–1998, 2022. 9

    1989

  49. [57]

    Cogvlm: Visual expert for pretrained language models

    Weihan Wang, Qingsong Lv, Wenmeng Yu, Wenyi Hong, Ji Qi, Yan Wang, Junhui Ji, Zhuoyi Yang, Lei Zhao, Xixuan Song, et al. Cogvlm: Visual expert for pretrained language models. arXiv preprint arXiv:2311.03079, 2023. 9

    2023 arXiv

  50. [58]

    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. In ICCVW, pages 1905–1914, 2021. 2, 3, 5, 6

    1905

  51. [59]

    Zero-shot im- age restoration using denoising diffusion null-space model

    Yinhuai Wang, Jiwen Yu, and Jian Zhang. Zero-shot im- age restoration using denoising diffusion null-space model. arXiv preprint arXiv:2212.00490, 2022. 9

    2022 arXiv

  52. [60]

    Chain-of-thought prompting elicits reasoning in large lan- guage models

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. Chain-of-thought prompting elicits reasoning in large lan- guage models. NeurIPS, 35:24824–24837, 2022. 5

    2022

  53. [61]

    Component divide-and-conquer for real-world image super-resolution

    Pengxu Wei, Ziwei Xie, Hannan Lu, Zongyuan Zhan, Qix- iang Ye, Wangmeng Zuo, and Liang Lin. Component divide-and-conquer for real-world image super-resolution. In ECCV, pages 101–117. Springer, 2020. 2, 5, 6

    2020

  54. [62]

    Seesr: Towards semantics-aware real-world image super-resolution

    Rongyuan Wu, Tao Yang, Lingchen Sun, Zhengqiang Zhang, Shuai Li, and Lei Zhang. Seesr: Towards semantics-aware real-world image super-resolution. In CVPR, pages 25456– 25467, 2024. 2, 4, 6, 9, 13

    2024

  55. [63]

    Llmga: Multimodal large language model based gener- ation assistant

    Bin Xia, Shiyin Wang, Yingfan Tao, Yitong Wang, and Jiaya Jia. Llmga: Multimodal large language model based gener- ation assistant. 2024. 3

    2024

  56. [64]

    Unsupervised image restoration with quality-task- perception loss

    Wei Xu, Xinyuan Chen, Haoming Guo, Xiaolin Huang, and Wei Liu. Unsupervised image restoration with quality-task- perception loss. TCSVT, 32(9):5736–5747, 2022. 9

    2022

  57. [65]

    Mastering text-to-image dif- fusion: Recaptioning, planning, and generating with multi- modal llms

    Ling Yang, Zhaochen Yu, Chenlin Meng, Minkai Xu, Ste- fano Ermon, and CUI Bin. Mastering text-to-image dif- fusion: Recaptioning, planning, and generating with multi- modal llms. In ICML, 2024. 3

    2024

  58. [66]

    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. In CVPR, pages 1191–1200,

  59. [67]

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

    Tao Yang, Rongyuan Wu, Peiran Ren, Xuansong Xie, and Lei Zhang. Pixel-aware stable diffusion for realistic im- age super-resolution and personalized stylization. In ECCV,

  60. [68]

    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. In CVPR, pages 25669–25680, 2024. 1, 2, 3, 4, 5, 6, 9

    2024

  61. [69]

    Promptfix: You prompt and we fix the photo

    Yongsheng Yu, Ziyun Zeng, Hang Hua, Jianlong Fu, and Jiebo Luo. Promptfix: You prompt and we fix the photo. arXiv preprint arXiv:2405.16785, 2024. 2, 9, 12

    2024 arXiv

  62. [70]

    Difface: Blind face restoration with diffused error contraction

    Zongsheng Yue and Chen Change Loy. Difface: Blind face restoration with diffused error contraction. arXiv preprint arXiv:2212.06512, 2022. 9

    2022 arXiv

  63. [71]

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

    Kai Zhang, Jingyun Liang, Luc Van Gool, and Radu Timo- fte. Designing a practical degradation model for deep blind image super-resolution. In ICCV, pages 4791–4800, 2021. 2

    2021

  64. [72]

    A feature-enriched completely blind image quality evaluator.IEEE Transactions on Image Processing, 24(8):2579–2591, 2015

    Lin Zhang, Lei Zhang, and Alan C Bovik. A feature-enriched completely blind image quality evaluator.IEEE Transactions on Image Processing, 24(8):2579–2591, 2015. 6

    2015

  65. [73]

    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. In CVPR, pages 586–595,

  66. [74]

    Blind image quality assessment via vision- language correspondence: A multitask learning perspective

    Weixia Zhang, Guangtao Zhai, Ying Wei, Xiaokang Yang, and Kede Ma. Blind image quality assessment via vision- language correspondence: A multitask learning perspective. In CVPR, pages 14071–14081, 2023. 6

    2023

  67. [75]

    Diff- restorer: Unleashing visual prompts for diffusion-based uni- versal image restoration

    Yuhong Zhang, Hengsheng Zhang, Xinning Chai, Zhengxue Cheng, Rong Xie, Li Song, and Wenjun Zhang. Diff- restorer: Unleashing visual prompts for diffusion-based uni- versal image restoration. arXiv preprint arXiv:2407.03636,

    arXiv

  68. [76]

    Self-supervised image restoration with blurry and noisy pairs

    Zhilu Zhang, RongJian Xu, Ming Liu, Zifei Yan, and Wang- meng Zuo. Self-supervised image restoration with blurry and noisy pairs. NeurIPS, 35:29179–29191, 2022. 9

    2022

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.