pith. machine review for the scientific record. sign in

arxiv: 2604.03061 · v2 · submitted 2026-04-03 · 💻 cs.CV

Recognition: 2 theorem links

· Lean Theorem

Can Nano Banana 2 Replace Traditional Image Restoration Models? An Evaluation of Its Performance on Image Restoration Tasks

Weixiong Sun , Xiang Yin , Chao Dong
Authors on Pith no claims yet

Pith reviewed 2026-05-13 20:07 UTC · model grok-4.3

classification 💻 cs.CV
keywords image restorationgenerative modelsNano Banana 2prompt designperceptual qualityfidelityuser studiesIQA metrics
0
0 comments X

The pith

Nano Banana 2 achieves competitive image restoration scores and user preference through prompt design yet produces over-enhanced details.

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

The paper evaluates whether the general-purpose generative model Nano Banana 2 can replace traditional specialized image restoration models across varied scenes and degradations. With concise prompts that include explicit fidelity constraints, the model reaches performance levels comparable to dedicated restorers on standard metrics while winning consistent preference in user studies and generalizing to difficult cases. This matters because a single versatile tool could simplify workflows that now require separate models for restoration versus editing. The evaluation also identifies a persistent gap where the outputs look richer but include inconsistencies and invented details that standard quality metrics fail to flag. A reader would care because the result points toward unified generative solutions for image tasks while underscoring the need for better fidelity controls.

Core claim

Nano Banana 2, when guided by concise prompts with explicit fidelity constraints, achieves competitive full-reference performance on diverse image restoration tasks and is consistently preferred in user studies while showing strong generalization in challenging scenarios. The model tends to produce visually rich results with over-enhanced details and inconsistencies, an issue not well captured by existing IQA metrics or standard user studies, indicating that general-purpose models show promise as unified IR solvers from a perceptual perspective but require improved controllability and fidelity-aware evaluation.

What carries the argument

Prompt engineering with concise instructions and fidelity constraints applied to Nano Banana 2 outputs, evaluated via full-reference metrics and user preference studies against traditional restorers.

If this is right

  • Concise prompts with fidelity constraints produce a better balance between accurate reconstruction and perceptual quality.
  • The model generalizes effectively to challenging degradation scenarios.
  • Standard IQA metrics and user studies overlook inconsistencies in generated details.
  • General-purpose generative models offer promise as unified image restoration solvers from a perceptual standpoint.
  • Improved controllability is needed to close the observed fidelity gap.

Where Pith is reading between the lines

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

  • New metrics focused on invented structures and consistency would give a clearer picture of generative restorers.
  • Hybrid systems that combine Nano Banana 2 outputs with traditional priors could reduce over-enhancement.
  • The same evaluation approach could test other general-purpose editing models on restoration benchmarks.
  • Future model development should embed explicit fidelity constraints rather than relying on post-hoc prompts.

Load-bearing premise

Existing IQA metrics and standard user studies are sufficient to detect gaps between perceptual quality and restoration fidelity when the model produces over-enhanced details and inconsistencies.

What would settle it

A targeted user study or new metric that specifically rates content accuracy and detail consistency would show whether the over-enhancements count as flaws that reverse the reported user preference.

Figures

Figures reproduced from arXiv: 2604.03061 by Chao Dong, Weixiong Sun, Xiang Yin.

Figure 1
Figure 1. Figure 1: Qualitative comparison of prompt length. Long prompts [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Failure cases with fidelity constraints. Infidelity still [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 2
Figure 2. Figure 2: Effect of fidelity constraints. Prompts without fidelity in [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Failure cases in output stability. Repeated runs on the [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: User study results. Nano Banana 2 achieves the highest [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Over-generation artifacts. The model produces excessive [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative comparison on challenging scenarios and degradations. Nano Banana 2 produces clearer structures and more consis [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
read the original abstract

Recent advances in generative AI raise the question of whether general-purpose image editing models can serve as unified solutions for image restoration. We conduct a systematic evaluation of Nano Banana 2 across diverse scenes and degradations. Our results show that prompt design is critical, with concise prompts and explicit fidelity constraints achieving a better balance between reconstruction and perceptual quality. Nano Banana 2 achieves competitive full-reference performance and is consistently preferred in user studies, while showing strong generalization in challenging scenarios. However, we observe a gap between perceptual quality and restoration fidelity, as the model tends to produce visually rich results with over-enhanced details and inconsistencies. This issue is not well captured by existing IQA metrics or user studies. Overall, general-purpose models show promise as unified IR solvers from a perceptual perspective, but require improved controllability and fidelity-aware evaluation. Further comparisons and detailed analyses are available in our project repository: https://github.com/yxyuanxiao/NanoBanana2TestOnIR.

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 / 1 minor

Summary. The manuscript evaluates Nano Banana 2, a general-purpose generative model, on image restoration tasks across diverse scenes and degradations. It claims that concise prompts with explicit fidelity constraints yield competitive full-reference performance and consistent user preference over traditional models, with strong generalization in challenging cases, while acknowledging a gap between perceptual quality and restoration fidelity manifested as over-enhanced details and inconsistencies that existing IQA metrics and user studies fail to capture. The authors conclude that such models show promise as unified IR solvers from a perceptual perspective but require improved controllability and fidelity-aware evaluation.

Significance. If the empirical results and user studies hold under scrutiny, the work would demonstrate that general-purpose generative models can function as unified solutions for image restoration, potentially simplifying pipelines that currently rely on specialized traditional models. The explicit identification of metric limitations and the call for better controllability add constructive value by highlighting open problems in evaluation.

major comments (2)
  1. Abstract: The claim of 'competitive full-reference performance' is load-bearing for the replacement thesis yet is presented without any numerical scores, baseline comparisons, tables, or error analysis in the abstract; this absence directly weakens the central assertion given the paper's own statement that the observed over-enhancement and inconsistencies are not captured by the metrics used to support competitiveness.
  2. Abstract: The explicit admission that 'this issue is not well captured by existing IQA metrics or user studies' creates an internal tension with the use of precisely those tools to assert competitiveness and user preference; the evaluation framework therefore cannot securely underwrite the conclusion that Nano Banana 2 can replace traditional models.
minor comments (1)
  1. Abstract: The GitHub repository link is useful but the manuscript should embed at least one summary table of quantitative results and one representative visual comparison to allow readers to assess the claims without external access.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful and detailed comments. We address each major comment below and have revised the abstract to improve clarity and support for our claims.

read point-by-point responses

  1. Referee: Abstract: The claim of 'competitive full-reference performance' is load-bearing for the replacement thesis yet is presented without any numerical scores, baseline comparisons, tables, or error analysis in the abstract; this absence directly weakens the central assertion given the paper's own statement that the observed over-enhancement and inconsistencies are not captured by the metrics used to support competitiveness.

    Authors: We agree that the abstract would be strengthened by including concrete numerical support. In the revised manuscript we have added a concise statement of key full-reference results (average PSNR and LPIPS across the evaluated datasets relative to the strongest traditional baselines) while preserving brevity. The complete tables, per-degradation breakdowns, and error analysis remain in Section 4 and the supplementary material. revision: yes

  2. Referee: Abstract: The explicit admission that 'this issue is not well captured by existing IQA metrics or user studies' creates an internal tension with the use of precisely those tools to assert competitiveness and user preference; the evaluation framework therefore cannot securely underwrite the conclusion that Nano Banana 2 can replace traditional models.

    Authors: The referee correctly notes a presentational tension. We have partially revised the abstract to explicitly distinguish the two layers of evidence: standard IQA metrics and user studies are reported because they are the established benchmarks for competitiveness, yet the text now states that these same tools do not fully capture the observed fidelity gap. This framing preserves the empirical findings while making the limitations and the call for improved controllability and fidelity-aware evaluation the central takeaway, rather than an unqualified replacement claim. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical evaluation without derivations or self-referential modeling

full rationale

The paper is a systematic empirical evaluation of Nano Banana 2 on image restoration tasks. It reports performance via full-reference metrics, user studies, and observations of over-enhancement without any equations, parameter fitting, derivations, or modeling steps. Claims rest on independent test results and studies; no load-bearing step reduces to its own inputs by construction. The noted gap between perceptual quality and fidelity is presented as an observation, not a fitted or self-defined result.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The paper is an empirical evaluation with no mathematical derivations, free parameters, axioms, or invented entities; all claims rest on experimental comparisons and user studies.

pith-pipeline@v0.9.0 · 5469 in / 983 out tokens · 45818 ms · 2026-05-13T20:07:53.503956+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.

Reference graph

Works this paper leans on

51 extracted references · 51 canonical work pages · 7 internal anchors

  1. [1]

    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. InCVPR, pages 10684– 10695, 2022. 1, 2

    work page 2022

  2. [2]

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

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

    work page 2022

  3. [3]

    In- structpix2pix: Learning to follow image editing instructions

    Tim Brooks, Aleksander Holynski, and Alexei A Efros. In- structpix2pix: Learning to follow image editing instructions. InCVPR, pages 18392–18402, 2023

    work page 2023

  4. [4]

    Grok imagine image.https : / / docs

    xAI. Grok imagine image.https : / / docs . x . ai/developers/models/grok-imagine-image,

    work page

  5. [5]

    Accessed: 2026-03-22

    work page 2026

  6. [6]

    Hunyuanimage 3.0 technical report.arXiv preprint arXiv:2509.23951, 2025

    Siyu Cao, Hangting Chen, Peng Chen, Yiji Cheng, Yutao Cui, Xinchi Deng, Ying Dong, Kipper Gong, Tianpeng Gu, Xiusen Gu, et al. Hunyuanimage 3.0 technical report.arXiv preprint arXiv:2509.23951, 2025

    work page arXiv 2025

  7. [7]

    Seedream 4.0: Toward Next-generation Multimodal Image Generation

    Team Seedream, Yunpeng Chen, Yu Gao, Lixue Gong, Meng Guo, Qiushan Guo, Zhiyao Guo, Xiaoxia Hou, Weilin Huang, Yixuan Huang, et al. Seedream 4.0: Toward next- generation multimodal image generation.arXiv preprint arXiv:2509.20427, 2025. 1

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  8. [8]

    Improving image generation with better captions.Computer Science, 2(3):8, 2023

    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, 2(3):8, 2023

    work page 2023

  9. [9]

    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 mod- els for high-resolution image synthesis.arXiv preprint arXiv:2307.01952, 2023. 1

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  10. [10]

    Beyond a gaussian denoiser: Residual learning of deep cnn for image denoising.TIP, 26(7):3142–3155, 2017

    Kai Zhang, Wangmeng Zuo, Yunjin Chen, Deyu Meng, and Lei Zhang. Beyond a gaussian denoiser: Residual learning of deep cnn for image denoising.TIP, 26(7):3142–3155, 2017. 1

    work page 2017

  11. [11]

    Deblurgan: Blind motion deblurring using conditional adversarial networks

    Orest Kupyn, V olodymyr Budzan, Mykola Mykhailych, Dmytro Mishkin, and Ji ˇr´ı Matas. Deblurgan: Blind motion deblurring using conditional adversarial networks. InCVPR, pages 8183–8192, 2018. 1

    work page 2018

  12. [12]

    Image super-resolution using deep convolutional net- works.PAMI, 38(2):295–307, 2015

    Chao Dong, Chen Change Loy, Kaiming He, and Xiaoou Tang. Image super-resolution using deep convolutional net- works.PAMI, 38(2):295–307, 2015. 1, 2

    work page 2015

  13. [13]

    Esrgan: En- hanced super-resolution generative adversarial networks

    Xintao Wang, Ke Yu, Shixiang Wu, Jinjin Gu, Yihao Liu, Chao Dong, Yu Qiao, and Chen Change Loy. Esrgan: En- hanced super-resolution generative adversarial networks. In ECCVW, pages 63–79, 2018. 1, 2

    work page 2018

  14. [14]

    Compression artifacts reduction by a deep convolu- tional network

    Chao Dong, Yubin Deng, Chen Change Loy, and Xiaoou Tang. Compression artifacts reduction by a deep convolu- tional network. InICCV, pages 576–584, 2015. 1

    work page 2015

  15. [15]

    Sinsr: diffusion-based image super- resolution in a single step

    Yufei Wang, Wenhan Yang, Xinyuan Chen, Yaohui Wang, Lanqing Guo, Lap-Pui Chau, Ziwei Liu, Yu Qiao, Alex C Kot, and Bihan Wen. Sinsr: diffusion-based image super- resolution in a single step. InCVPR, pages 25796–25805,

    work page

  16. [16]

    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. InCVPR, pages 9935–9946, 2023

    work page 2023

  17. [17]

    Exploiting diffusion prior for real-world image super-resolution.IJCV, 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.IJCV, 132(12):5929– 5949, 2024. 1, 2

    work page 2024

  18. [18]

    Unires: Universal im- age restoration for complex degradations

    Mo Zhou, Keren Ye, Mauricio Delbracio, Peyman Milanfar, Vishal M Patel, and Hossein Talebi. Unires: Universal im- age restoration for complex degradations. InICCV, pages 13237–13247, 2025. 1, 2

    work page 2025

  19. [19]

    Diffusion models in low-level vision: A survey.PAMI, 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.PAMI, 2025

    work page 2025

  20. [20]

    A survey on all-in-one image restoration: Tax- onomy, evaluation and future trends.PAMI, 2025

    Junjun Jiang, Zengyuan Zuo, Gang Wu, Kui Jiang, and Xi- anming Liu. A survey on all-in-one image restoration: Tax- onomy, evaluation and future trends.PAMI, 2025. 1, 2

    work page 2025

  21. [21]

    Generative modeling by esti- mating gradients of the data distribution.NeurIPS, 32, 2019

    Yang Song and Stefano Ermon. Generative modeling by esti- mating gradients of the data distribution.NeurIPS, 32, 2019. 1

    work page 2019

  22. [22]

    Denoising Diffusion Implicit Models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models.arXiv preprint arXiv:2010.02502, 2020. 1

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  23. [23]

    Gemini: A Family of Highly Capable Multimodal Models

    Gemini Team, Rohan Anil, Sebastian Borgeaud, Jean- Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M Dai, Anja Hauth, Katie Millican, et al. Gemini: a family of highly capable multimodal models.arXiv preprint arXiv:2312.11805, 2023. 1, 2

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  24. [24]

    Black Forest Labs, Stephen Batifol, Andreas Blattmann, Frederic Boesel, Saksham Consul, Cyril Diagne, Tim Dock- horn, Jack English, Zion English, Patrick Esser, et al. Flux. 1 kontext: Flow matching for in-context image generation and editing in latent space.arXiv preprint arXiv:2506.15742, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  25. [25]

    Z-Image: An Efficient Image Generation Foundation Model with Single-Stream Diffusion Transformer

    Huanqia Cai, Sihan Cao, Ruoyi Du, Peng Gao, Steven Hoi, Zhaohui Hou, Shijie Huang, Dengyang Jiang, Xin Jin, Liangchen Li, et al. Z-image: An efficient image generation foundation model with single-stream diffusion transformer. arXiv preprint arXiv:2511.22699, 2025. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  26. [26]

    Qwen-Image Technical Report

    Chenfei Wu, Jiahao Li, Jingren Zhou, Junyang Lin, Kaiyuan Gao, Kun Yan, Sheng-ming Yin, Shuai Bai, Xiao Xu, Yilei Chen, et al. Qwen-image technical report.arXiv preprint arXiv:2508.02324, 2025. 1

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  27. [27]

    The perception-distortion tradeoff

    Yochai Blau and Tomer Michaeli. The perception-distortion tradeoff. InCVPR, pages 6228–6237, 2018. 1

    work page 2018

  28. [28]

    How far have we gone in gener- ative image restoration? a study on its capability, limitations and evaluation practices.arXiv preprint arXiv:2603.05010,

    Xiang Yin, Jinfan Hu, Zhiyuan You, Kainan Yan, Yu Tang, Chao Dong, and Jinjin Gu. How far have we gone in gener- ative image restoration? a study on its capability, limitations and evaluation practices.arXiv preprint arXiv:2603.05010,

    work page arXiv

  29. [29]

    Is nano banana pro a low-level vision all-rounder? a comprehensive evaluation on 14 tasks and 40 datasets.arXiv preprint arXiv:2512.15110,

    Jialong Zuo, Haoyou Deng, Hanyu Zhou, Jiaxin Zhu, Yicheng Zhang, Yiwei Zhang, Yongxin Yan, Kaixing Huang, Weisen Chen, Yongtai 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,

    work page arXiv

  30. [30]

    Pipal: a large-scale image quality assessment dataset for perceptual image restoration

    Gu Jinjin, Cai Haoming, Chen Haoyu, Ye Xiaoxing, Jimmy S Ren, and Dong Chao. Pipal: a large-scale image quality assessment dataset for perceptual image restoration. InECCV, pages 633–651, 2020. 2, 7

    work page 2020

  31. [31]

    Accurate image restora- tion with attention retractable transformer.arXiv preprint arXiv:2210.01427, 2022

    Jiale Zhang, Yulun Zhang, Jinjin Gu, Yongbing Zhang, Linghe Kong, and Xin Yuan. Accurate image restora- tion with attention retractable transformer.arXiv preprint arXiv:2210.01427, 2022. 2

    work page arXiv 2022

  32. [32]

    Image denoising by sparse 3-d transform- domain collaborative filtering.TIP, 16(8):2080–2095, 2007

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

    work page 2080

  33. [33]

    Bilateral filtering for gray and color images

    Carlo Tomasi and Roberto Manduchi. Bilateral filtering for gray and color images. InICCV, pages 839–846. IEEE,

    work page

  34. [34]

    Acceler- ating the super-resolution convolutional neural network

    Chao Dong, Chen Change Loy, and Xiaoou Tang. Acceler- ating the super-resolution convolutional neural network. In ECCV, pages 391–407, 2016. 2

    work page 2016

  35. [35]

    Restormer: Efficient transformer for high-resolution image restoration

    Syed Waqas Zamir, Aditya Arora, Salman Khan, Mu- nawar Hayat, Fahad Shahbaz Khan, and Ming-Hsuan Yang. Restormer: Efficient transformer for high-resolution image restoration. InCVPR, pages 5728–5739, 2022. 2

    work page 2022

  36. [36]

    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. InICCV, pages 1833–1844, 2021. 2

    work page 2021

  37. [37]

    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. InCVPR, pages 25669–25680, 2024. 2, 7

    work page 2024

  38. [38]

    Unicon: Unidirectional information flow for effec- tive control of large-scale diffusion models.arXiv preprint arXiv:2503.17221, 2025

    Fanghua Yu, Jinjin Gu, Jinfan Hu, Zheyuan Li, and Chao Dong. Unicon: Unidirectional information flow for effec- tive control of large-scale diffusion models.arXiv preprint arXiv:2503.17221, 2025. 2

    work page arXiv 2025

  39. [39]

    Ultraedit: Instruction-based fine-grained im- age editing at scale.NeurIPS, 37:3058–3093, 2024

    Haozhe Zhao, Xiaojian Ma, Liang Chen, Shuzheng Si, Ru- jie Wu, Kaikai An, Peiyu Yu, Minjia Zhang, Qing Li, and Baobao Chang. Ultraedit: Instruction-based fine-grained im- age editing at scale.NeurIPS, 37:3058–3093, 2024. 3

    work page 2024

  40. [40]

    Looks too good to be true: An information- theoretic analysis of hallucinations in generative restoration models.NeurIPS, 37:22596–22623, 2024

    Regev Cohen, Idan Kligvasser, Ehud Rivlin, and Daniel Freedman. Looks too good to be true: An information- theoretic analysis of hallucinations in generative restoration models.NeurIPS, 37:22596–22623, 2024. 3

    work page 2024

  41. [41]

    Harnessing diffusion-yielded score priors for image restoration.TOG, 44(6):1–21, 2025

    Xinqi Lin, Fanghua Yu, Jinfan Hu, Zhiyuan You, Wu Shi, Jimmy S Ren, Jinjin Gu, and Chao Dong. Harnessing diffusion-yielded score priors for image restoration.TOG, 44(6):1–21, 2025. 4

    work page 2025

  42. [42]

    Tsdsr: temporal–spatial domain de- noise super-resolution photon-efficient 3d reconstruction by deep learning

    Ziyi Tong, Xinding Jiang, Jiemin Hu, Lu Xu, Long Wu, Xu Yang, and Bo Zou. Tsdsr: temporal–spatial domain de- noise super-resolution photon-efficient 3d reconstruction by deep learning. InPhotonics, volume 10, page 744. MDPI,

    work page

  43. [43]

    Pixel-level and semantic-level adjustable super-resolution: A dual-lora approach

    Lingchen Sun, Rongyuan Wu, Zhiyuan Ma, Shuaizheng Liu, Qiaosi Yi, and Lei Zhang. Pixel-level and semantic-level adjustable super-resolution: A dual-lora approach. InCVPR, pages 2333–2343, 2025. 4

    work page 2025

  44. [44]

    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. InECCV, pages 430–448, 2024. 4

    work page 2024

  45. [45]

    Image quality assessment: from error visibility to structural similarity.TIP, 13(4):600–612, 2004

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. Image quality assessment: from error visibility to structural similarity.TIP, 13(4):600–612, 2004. 4

    work page 2004

  46. [46]

    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, pages 586–595,

    work page

  47. [47]

    Musiq: Multi-scale image quality transformer

    Junjie Ke, Qifei Wang, Yilin Wang, Peyman Milanfar, and Feng Yang. Musiq: Multi-scale image quality transformer. InCVPR, pages 5148–5157, 2021. 4

    work page 2021

  48. [48]

    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. InCVPR, pages 1191–1200,

    work page

  49. [49]

    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, volume 37, pages 2555–2563, 2023. 4

    work page 2023

  50. [50]

    The use of ranks to avoid the assumption of normality implicit in the analysis of variance.Journal of the american statistical association, 32(200):675–701, 1937

    Milton Friedman. The use of ranks to avoid the assumption of normality implicit in the analysis of variance.Journal of the american statistical association, 32(200):675–701, 1937. 5

    work page 1937

  51. [51]

    Position: Eval- uation of visual processing should be human-centered, not metric-centered.arXiv preprint arXiv:2603.00643, 2026

    Jinfan Hu, Fanghua Yu, Zhiyuan You, Xiang Yin, Hongyu An, Xinqi Lin, Chao Dong, and Jinjin Gu. Position: Eval- uation of visual processing should be human-centered, not metric-centered.arXiv preprint arXiv:2603.00643, 2026. 7 10

    work page arXiv 2026