Pith. sign in

REVIEW 4 major objections 4 minor 70 references

Position: Agentic Systems Constitute a Key Component of Next-Generation Intelligent Image Processing

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This position paper argues that agentic systems that dynamically select, combine, and optimize existing image processing tools are a key component of next-generation intelligent image processing, complementing model-centric research.

desk verdict A well-framed position paper whose useful six-level agentic capability taxonomy is undercut by an asserted rather than demonstrated premise that model-centric limitations are intrinsic. read the letter →

arxiv 2505.16007 v1 pith:FVC7YO3D submitted 2025-05-21 cs.CV

Jinjin Gu This is my paper
classification cs.CV
keywords agenticimageprocessinglargelanguagemodelstoolorchestrationrestorationgeneralizationcognitivearchitecturepositionpaperqualityassessment
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This position paper argues that the image processing field's next leap will come not from bigger or more general deep models alone, but from agentic systems that act as intelligent coordinators, dynamically selecting, combining, and optimizing existing image processing tools. The paper claims current deep image processing models are limited by a fundamental trade-off between task coverage and performance, and by their tendency to overfit training degradations, making them brittle on real-world problems. A system that reasons about the input image, plans a sequence of tool calls, reflects on intermediate results, and adjusts its plan can emulate human experts and reach generality that monolithic models cannot. The author proposes a six-level scale of agentic capability, from fixed operations to fully autonomous creative systems, and outlines design principles and research problems for building such systems.

What carries the argument

The central object is the agentic image processing system: an AI agent that treats image processing models as tools and coordinates them through a cognitive architecture. The paper lays out a reference architecture with five stages—Perception, Scheduling, Execution, Reflection, and Rescheduling—where the agent examines the input, plans a sequence of operations, runs the tools, judges whether the intermediate result is acceptable, and modifies the plan if not. It also defines six levels of agentic capability, from Level 0 (fixed operations) through Level 1 (AI-augmented functions), Level 2 (adaptive model selection), Level 3 (reflection and flow control), Level 4 (self-evolution), to Level 5 (fully autonomous and creative systems). The large language model is the component that supplies reasoning, planning, and natural-language interaction, and multimodal LLMs act as the system's 'eyes' for content recognition and quality assessment.

What would settle it

Train or find a single image processing model that, with scaling alone, matches the performance of an LLM-orchestrated agent on a held-out benchmark of never-seen degradations and out-of-distribution images, with the agent limited to tools no stronger than that model; if the single model ties or beats the agent at equal or lower total compute, the claim that static architectures cannot reach generality is falsified.

Watch

Extended reading notes

Core claim

The paper's central claim is that AI-agent-based systems, powered by large language models, constitute a key component of next-generation intelligent image processing, alongside continued model development. The argument is that every image processing model, however advanced, is a tool whose effectiveness depends on preprocessing, postprocessing, iterative application, and combination with other tools, and that an agent with perception, planning, execution, and reflection modules can perform this orchestration the way a human expert would. The author asserts that the limitations of end-to-end deep models—limited generalization, the accuracy-generality trade-off, and lack of semantic understanding—are inherent to static architectures and cannot be fully resolved by scaling alone. Therefore, the field should broaden from model-centric research to agentic-oriented research, designing systems that integrate multiple models, adapt strategies to input content, and continuously improve.

Load-bearing premise

The load-bearing premise is that the generalization limits of static deep image processing models are inherent and cannot be resolved by scale or new architectures; the paper's own Alternative Views section acknowledges that scaling might subsume agentic capabilities but rejects this without empirical evidence, so if that rejection is wrong the central claim collapses.

Editorial extensions

If this is right

  • Research priorities shift from training ever-broader models toward building orchestration frameworks, evaluation benchmarks, and tool interfaces that make models composable.
  • A single all-in-one restoration model can be combined with other operations to extend coverage, so model performance and agent-level performance become separable axes of progress.
  • Image quality assessment and content analysis become first-class components of the processing system, since the agent's decisions depend on judging intermediate results accurately.
  • Knowledge infusion into agents, via fine-tuning or retrieval-augmented generation, becomes a core research topic because the planning model must know what operations are possible and when to use them.
  • Human-computer interaction changes from one-shot tool invocation to step-by-step visible, correctable collaboration between user and agent.

Reading between the lines

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

  • If the position is right, the limiting resource for practical image processing becomes the availability of diverse, well-documented, composable tools and the quality of the planning model, rather than the pixel-level capacity of any single network.
  • A concrete testable consequence is that on held-out degradation types unseen during training, an agentic system that iterates and reflects should outperform its strongest component model by a margin that grows with task complexity; a benchmark that measures this margin would separate the paradigm's contribution from the models'.
  • The six-level agentic scale invites a comparison with autonomous-driving levels, but unlike driving, image processing is reversible and low-risk, so the field may reach Level 4-5 capabilities long before physical systems do.
  • One could also read the argument as a roadmap for making model scaling less central to generality: if orchestration of simpler models plus reflection closes the gap, research investment in massive multi-task restoration models may be better spent on tool libraries and agent evaluation.
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 / 4 minor

Summary. This position paper argues that image processing research should broaden its focus from training monolithic deep models toward building agentic systems that dynamically select, combine, and optimize existing image processing tools. It reviews limitations of current model-centric approaches (generalization failures, task-versus-performance trade-offs, limited interactivity), introduces AI agents and LLM-based planning, proposes a taxonomy of agentic capability levels from fixed operations to fully autonomous creative agents, and discusses open problems in cognitive architecture, quality assessment, knowledge acquisition, human-computer interaction, and self-evolution. The contribution is a research agenda and design vocabulary rather than a new system or an empirical study.

Significance. If the central claim is accepted, the paper articulates a plausible and timely reorientation of image processing research: system-level orchestration of existing tools may complement or supersede further model scaling. The paper has clear strengths: a useful six-level taxonomy of agentic capability, explicit identification of open problems (perception/reflection, knowledge injection, interaction, self-evolution), and an acknowledgment of alternative views. It is also honest that the higher capability levels are speculative. However, the significance depends on a premise that is not established: that the limitations of model-centric approaches are intrinsic and cannot be overcome by scale, data, or architecture. The empirical support for the diagnosis comes almost entirely from the author's own prior papers, and the proposed LLM-based solution may reintroduce the same distribution-mismatch problem one level up. The taxonomy itself is descriptive and should be useful to the community even if the stronger normative claim is weakened.

major comments (4)
  1. [§1 (Alternative Views), §2.2] The paper's central claim—that agentic systems are a key component of next-generation image processing—rests on the assertion that static deep models cannot achieve the required generality and intelligence, making orchestration necessary. This is a load-bearing premise, but the paper supports it only with observations that current models overfit to training degradations [20,38] and exhibit an accuracy–generalization trade-off [66], all of which are consistent with the alternative that larger models, richer degradation distributions, or new architectures will close the gap. The 'Alternative Views' paragraph dismisses scaling without a scaling comparison or a theoretical argument for why distribution mismatch is irreducible. The authors should either provide a concrete test (e.g., a systematic scaling study of generalist restoration models versus an agentic orchestrator on a held-out task distribution) or reframe the claim as 'current model-centric approaches are insufficient,' which the evidence supports. The evidential base would also be strengthened by independent reproductions, since the key empirical citations ([20], [38], [66]) are all from the author's own prior work.
  2. [§2.2] The statement 'essentially, current image processing networks are not intelligent' overreaches. The cited interpretability studies [19,37,22] show that super-resolution and restoration networks rely on local attributions or specific causal features rather than human-aligned semantic representations; they do not demonstrate the absence of task-relevant intelligence, and they certainly do not license a claim about all 'current image processing networks.' In particular, the same section later acknowledges that generative-prior methods [62] exhibit content-aware behaviors. The claim should be qualified to 'many end-to-end supervised models do not exhibit the kind of semantic understanding we associate with intelligent processing,' which is sufficient for the argument.
  3. [§4.4, §5.3] The proposed remedy reintroduces the diagnosed problem one level up. The planner in the agentic system is an LLM (or multimodal LLM), which is itself a static model trained on a finite corpus; the paper acknowledges in §5.3 that LLMs contain only basic knowledge and may fail on unseen problems, but it does not explain why this distribution mismatch is more tractable than the one facing image processing networks. No evidence is offered that LLM-based planning generalizes to novel image processing tasks beyond the demonstrated benchmarks (e.g., [9], [69]). Without such evidence, the paper has not shown that agentic systems overcome the fundamental limitation it attributes to monolithic models; it has only moved the limitation to a different component. At minimum, the authors should discuss why the LLM's large-scale pretraining and tool-use loop mitigate the same overfitting concern, or weaken the claim to a research hypothesis.
  4. [§4.3] The six-level taxonomy is presented as a reference framework, but the criteria for assigning a system to a level are not defined, and the transitions are not operationalized. In particular, Level 3 ('Reflection') and Level 4 ('Self-evolution') are distinguished by qualitative features (reflecting on results versus updating workflows), but no tests or metrics are offered to determine when a system crosses from one level to the next. For a position paper this is acceptable as a starting point, but the claim that 'Level 4 ... surpasses what fixed architectures can achieve' is an unsupported assertion that should be marked as a conjecture.
minor comments (4)
  1. [§4.3] The text says the six tiers are 'shown in Figure 2,' but Figure 2 depicts processing paradigms; the levels appear in Figure 3. Please correct the cross-reference.
  2. [§3] In the sentence 'frameworks like HuggingGPT [44] and Visual ChatGPT [44]', the second citation should be [52] (Wu et al., Visual ChatGPT); [44] is the HuggingGPT paper.
  3. [§4.3] In Level 5, the phrase 'don't know about.).' has stray punctuation; it should read 'don't know about).'
  4. [§4.1] The claim that 'this paper is the first to advocate for the construction of an agentic system' for image processing is difficult to reconcile with the immediately cited works [44], [52], [9], and [69], which already construct LLM-based systems that select and combine image processing tools. Please either clarify the sense in which the paper is first (e.g., first to propose a systematic research framework) or soften the claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the position paper argues a research direction and supports its empirical premises with independent prior studies, including the author's own, rather than reducing a prediction to its inputs.

full rationale

This is a position paper, not a derivation or prediction pipeline. The central claim is a normative research recommendation: image processing should incorporate agentic system design. The supporting premises are empirical observations about model generalization (e.g., [20], [38]), the accuracy-generalization trade-off (e.g., [66]), and the feasibility of agentic image processing (e.g., [69], [9]). These are cited to published studies with external benchmarks, not to quantities fitted within this paper. Self-citation is present and frequent, but the cited results are independent, falsifiable empirical findings, so they do not make the argument circular under the stated rules. The 'Alternative Views' paragraph dismisses scaling-based counterarguments without a comparative experiment, but that is an argumentative weakness or missing evidence, not a circular step: no equation, fitted parameter, or imported uniqueness theorem reduces the conclusion to its premises. The paper also explicitly acknowledges the preliminary nature of the area ('we currently have only very preliminary results', Sec. 5.3), which further indicates it is not claiming a derivation. No step qualifies as self-definitional, fitted-input-called-prediction, load-bearing self-citation in the prohibited sense, imported uniqueness, ansatz-smuggling via citation, or renaming a known result as a new derivation. Therefore the correct circularity score is 0.

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

The paper rests on domain assumptions about model limitations and LLM reliability, all of which are plausible but unproven in this text. There are no free parameters or invented entities.

assumptions (4)
  • domain assumption Deep image processing models overfit to training data and fail to generalize to out-of-distribution degradations.
    Stated in Sec. 2.2 and supported by citations to the author's own prior studies ([8], [20], [38]); this premise motivates the entire need for agents.
  • domain assumption A single model cannot simultaneously achieve broad task coverage and high per-task performance.
    Invoked in Sec. 4.1 with citation [66] (author's own work); used to argue that no point model can replace tool orchestration.
  • domain assumption Large language models can reliably plan image processing workflows from image content and user instructions.
    Assumed in Sec. 4.4 without evidence or benchmarks; the paper cites LLM capabilities generally but offers no data that planning accuracy is sufficient for low-level vision.
  • ad hoc to paper Scaling or architecture improvements cannot overcome the fundamental mismatch between static models and dynamic task requirements.
    Sec. 1 'Alternative Views' rejects the mainstream counterargument without proof; this is a position-dependent assumption that is necessary for the paper's conclusion.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Position: Agentic Systems Constitute a Key Component of Next-Generation Intelligent Image Processing." pith.science (2026). https://pith.science/paper/FVC7YO3D

@misc{pith2026250516007,
  author       = {Pith},
  title        = {Pith review of: Position: Agentic Systems Constitute a Key Component of Next-Generation Intelligent Image Processing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FVC7YO3D}},
  note         = {Machine review of arXiv:2505.16007}
}
read the original abstract

This position paper argues that the image processing community should broaden its focus from purely model-centric development to include agentic system design as an essential complementary paradigm. While deep learning has significantly advanced capabilities for specific image processing tasks, current approaches face critical limitations in generalization, adaptability, and real-world problem-solving flexibility. We propose that developing intelligent agentic systems, capable of dynamically selecting, combining, and optimizing existing image processing tools, represents the next evolutionary step for the field. Such systems would emulate human experts' ability to strategically orchestrate different tools to solve complex problems, overcoming the brittleness of monolithic models. The paper analyzes key limitations of model-centric paradigms, establishes design principles for agentic image processing systems, and outlines different capability levels for such agents.

Figures

Figures reproduced from arXiv: 2505.16007 by the authors.

Figure 1
Figure 1. The existing research paradigm focuses on devel [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. How image processing systems can embody different levels of agentic to enhance their [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Levels of agentic capability in image processing systems, illustrating the progression from [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Cognitive architecture for image processing systems, illustrating the iterative process of perception, scheduling, execution, reflection, and rescheduling to achieve satis￾factory results. The foundation for building more complex agentic systems lies in designing their…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

70 extracted references · 31 canonical work pages

  1. [66]

    Crafting training degradation distribution for the accuracy-generalization trade-off in real-world super-resolution

    Ruofan Zhang, Jinjin Gu, Haoyu Chen, Chao Dong, Yulun Zhang, and Wenming Yang. Crafting training degradation distribution for the accuracy-generalization trade-off in real-world super-resolution. In International conference on machine learning, pages 41078–41091. PMLR, 2023

    work page 2023

  2. [20]

    Networks are slacking off: Under- standing generalization problem in image deraining

    Jinjin Gu, Xianzheng Ma, Xiangtao Kong, Yu Qiao, and Chao Dong. Networks are slacking off: Under- standing generalization problem in image deraining. Advances in Neural Information Processing Systems, 36, 2023

    work page 2023

  3. [38]

    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. IEEE Transactions on pattern analysis and machine intelligence, 2023

    work page 2023

  4. [62]

    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 Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 25669–25680, 2024

    2024

  5. [9]

    Restoreagent: Autonomous image restoration agent via multimodal large language models

    Haoyu Chen, Wenbo Li, Jinjin Gu, Jingjing Ren, Sixiang Chen, Tian Ye, Renjing Pei, Kaiwen Zhou, Fenglong Song, and Lei Zhu. Restoreagent: Autonomous image restoration agent via multimodal large language models. Advances in Neural Information Processing Systems, 2024

    work page 2024

  6. [69]

    An intelligent agentic system for complex image restoration problems

    Kaiwen Zhu, Jinjin Gu, Zhiyuan You, Yu Qiao, and Chao Dong. An intelligent agentic system for complex image restoration problems. arXiv preprint arXiv:2410.17809, 2024

    arXiv 2024

  7. [1]

    Gpt-4 technical report

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report. arXiv preprint arXiv:2303.08774, 2023

    arXiv 2023

  8. [2]

    Graph of thoughts: Solving elaborate problems with large language models

    Maciej Besta, Nils Blach, Ales Kubicek, Robert Gerstenberger, Michal Podstawski, Lukas Gianinazzi, Joanna Gajda, Tomasz Lehmann, Hubert Niewiadomski, Piotr Nyczyk, et al. Graph of thoughts: Solving elaborate problems with large language models. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 17682–17690, 2024

    2024

Show all 70 references
  1. [3]

    Unpro- cessing images for learned raw denoising

    Tim Brooks, Ben Mildenhall, Tianfan Xue, Jiawen Chen, Dillon Sharlet, and Jonathan T Barron. Unpro- cessing images for learned raw denoising. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11036–11045, 2019

    2019

  2. [4]

    Language models are few-shot learners

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. Advances in neural information processing systems, 33:1877–1901, 2020. 11

    1901

  3. [5]

    Reconfiguring the imaging pipeline for computer vision

    Mark Buckler, Suren Jayasuriya, and Adrian Sampson. Reconfiguring the imaging pipeline for computer vision. In Proceedings of the IEEE International Conference on Computer Vision, pages 975–984, 2017

    2017

  4. [6]

    A survey on mixture of experts

    Weilin Cai, Juyong Jiang, Fan Wang, Jing Tang, Sunghun Kim, and Jiayi Huang. A survey on mixture of experts. arXiv preprint arXiv:2407.06204, 2024

    2024 arXiv

  5. [7]

    Autoagents: A framework for automatic agent generation

    Guangyao Chen, Siwei Dong, Yu Shu, Ge Zhang, Jaward Sesay, Börje F Karlsson, Jie Fu, and Yemin Shi. Autoagents: A framework for automatic agent generation. arXiv preprint arXiv:2309.17288, 2023

    2023 arXiv

  6. [8]

    Masked image training for generalizable deep image denoising

    Haoyu Chen, Jinjin Gu, Yihao Liu, Salma Abdel Magid, Chao Dong, Qiong Wang, Hanspeter Pfister, and Lei Zhu. Masked image training for generalizable deep image denoising. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1692–1703, 2023

    2023

  7. [10]

    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. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pa...

    2024

  8. [11]

    Learning a low-level vision generalist via visual task prompt

    Xiangyu Chen, Yihao Liu, Yuandong Pu, Wenlong Zhang, Jiantao Zhou, Yu Qiao, and Chao Dong. Learning a low-level vision generalist via visual task prompt. arXiv preprint arXiv:2408.08601, 2024

    2024 arXiv

  9. [12]

    Mod-squad: Designing mixtures of experts as modular multi-task learners

    Zitian Chen, Yikang Shen, Mingyu Ding, Zhenfang Chen, Hengshuang Zhao, Erik G Learned-Miller, and Chuang Gan. Mod-squad: Designing mixtures of experts as modular multi-task learners. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11...

    2023

  10. [13]

    Deepseekmoe: Towards ultimate expert specialization in mixture-of-experts language models

    Damai Dai, Chengqi Deng, Chenggang Zhao, RX Xu, Huazuo Gao, Deli Chen, Jiashi Li, Wangding Zeng, Xingkai Yu, Y Wu, et al. Deepseekmoe: Towards ultimate expert specialization in mixture-of-experts language models. arXiv preprint arXiv:2401.06066, 2024

    2024 arXiv

  11. [14]

    Image super-resolution using deep convolutional networks

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

    2015

  12. [15]

    The llama 3 herd of models

    Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, et al. The llama 3 herd of models. arXiv preprint arXiv:2407.21783, 2024

    2024 arXiv

  13. [16]

    Agent ai: Surveying the horizons of multimodal interaction

    Zane Durante, Qiuyuan Huang, Naoki Wake, Ran Gong, Jae Sung Park, Bidipta Sarkar, Rohan Taori, Yusuke Noda, Demetri Terzopoulos, Yejin Choi, et al. Agent ai: Surveying the horizons of multimodal interaction. arXiv preprint arXiv:2401.03568, 2024

    2024 arXiv

  14. [17]

    Is it an agent, or just a program?: A taxonomy for autonomous agents

    Stan Franklin and Art Graesser. Is it an agent, or just a program?: A taxonomy for autonomous agents. In International workshop on agent theories, architectures, and languages, pages 21–35. Springer, 1996

    1996

  15. [18]

    Image quality assessment for perceptual image restoration: A new dataset, benchmark and metric

    Jinjin Gu, Haoming Cai, Haoyu Chen, Xiaoxing Ye, Jimmy Ren, and Chao Dong. Image quality assessment for perceptual image restoration: A new dataset, benchmark and metric. arXiv preprint arXiv:2011.15002, 2020

    2011 arXiv

  16. [19]

    Interpreting super-resolution networks with local attribution maps

    Jinjin Gu and Chao Dong. Interpreting super-resolution networks with local attribution maps. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9199–9208, 2021

    2021

  17. [21]

    Saycanpay: Heuristic planning with large language models using learnable domain knowledge

    Rishi Hazra, Pedro Zuidberg Dos Martires, and Luc De Raedt. Saycanpay: Heuristic planning with large language models using learnable domain knowledge. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 20123–20133, 2024

    2024

  18. [22]

    Interpreting low-level vision models with causal effect maps

    Jinfan Hu, Jinjin Gu, Shiyao Yu, Fanghua Yu, Zheyuan Li, Zhiyuan You, Chaochao Lu, and Chao Dong. Interpreting low-level vision models with causal effect maps. arXiv preprint arXiv:2407.19789, 2024

    2024 arXiv

  19. [23]

    Exposure: A white-box photo post-processing framework

    Yuanming Hu, Hao He, Chenxi Xu, Baoyuan Wang, and Stephen Lin. Exposure: A white-box photo post-processing framework. ACM Transactions on Graphics (TOG), 37(2):1–17, 2018. 12

    2018

  20. [24]

    Learning agile and dynamic motor skills for legged robots.Science Robotics, 4(26):eaau5872, 2019

    Jemin Hwangbo, Joonho Lee, Alexey Dosovitskiy, Dario Bellicoso, Vassilios Tsounis, Vladlen Koltun, and Marco Hutter. Learning agile and dynamic motor skills for legged robots.Science Robotics, 4(26):eaau5872, 2019

    2019

  21. [25]

    Mistral 7b

    Albert Q Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, et al. Mistral 7b. arXiv preprint arXiv:2310.06825, 2023

    arXiv

  22. [26]

    Mixtral of experts

    Albert Q Jiang, Alexandre Sablayrolles, Antoine Roux, Arthur Mensch, Blanche Savary, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Emma Bou Hanna, Florian Bressand, et al. Mixtral of experts. arXiv preprint arXiv:2401.04088, 2024

    2024 arXiv

  23. [27]

    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. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XI 16...

    2020

  24. [28]

    Thinking, fast and slow

    Daniel Kahneman. Thinking, fast and slow. Farrar, Straus and Giroux, 2011

    2011

  25. [29]

    Level-5 autonomous driving—are we there yet? a review of research literature

    Manzoor Ahmed Khan, Hesham El Sayed, Sumbal Malik, Talha Zia, Jalal Khan, Najla Alkaabi, and Henry Ignatious. Level-5 autonomous driving—are we there yet? a review of research literature. ACM Computing Surveys (CSUR), 55(2):1–38, 2022

    2022

  26. [30]

    Large language models are zero-shot reasoners

    Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. Large language models are zero-shot reasoners. Advances in neural information processing systems, 35:22199–22213, 2022

    2022

  27. [31]

    A preliminary exploration towards general image restoration

    Xiangtao Kong, Jinjin Gu, Yihao Liu, Wenlong Zhang, Xiangyu Chen, Yu Qiao, and Chao Dong. A preliminary exploration towards general image restoration. arXiv preprint arXiv:2408.15143, 2024

    2024 arXiv

  28. [32]

    Deep learning

    Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning. nature, 521(7553):436–444, 2015

    2015

  29. [33]

    Retrieval-augmented generation for knowledge- intensive nlp tasks

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, et al. Retrieval-augmented generation for knowledge- intensive nlp tasks. Advances in Neural Information Processing System...

    2020

  30. [34]

    Multimodal foundation models: From specialists to general-purpose assistants

    Chunyuan Li, Zhe Gan, Zhengyuan Yang, Jianwei Yang, Linjie Li, Lijuan Wang, Jianfeng Gao, et al. Multimodal foundation models: From specialists to general-purpose assistants. Foundations and Trends® in Computer Graphics and Vision, 16(1-2):1–214, 2024

    2024

  31. [35]

    Swiftsage: A generative agent with fast and slow thinking for complex interactive tasks

    Bill Yuchen Lin, Yicheng Fu, Karina Yang, Faeze Brahman, Shiyu Huang, Chandra Bhagavatula, Prithviraj Ammanabrolu, Yejin Choi, and Xiang Ren. Swiftsage: A generative agent with fast and slow thinking for complex interactive tasks. Advances in Neural Information Processing Syst...

    2024

  32. [36]

    Jarvisir: Elevating autonomous driving perception with intelligent image restoration

    Yunlong Lin, Zixu Lin, Haoyu Chen, Panwang Pan, Chenxin Li, Sixiang Chen, Yeying Jin, Wenbo Li, and Xinghao Ding. Jarvisir: Elevating autonomous driving perception with intelligent image restoration. arXiv preprint arXiv:2504.04158, 2025

    2025 arXiv

  33. [37]

    semantics

    Yihao Liu, Anran Liu, Jinjin Gu, Zhipeng Zhang, Wenhao Wu, Yu Qiao, and Chao Dong. Discovering distinctive" semantics" in super-resolution networks. arXiv preprint arXiv:2108.00406, 2021

    2021 arXiv

  34. [39]

    Texture-based error analysis for image super-resolution

    Salma Abdel Magid, Zudi Lin, Donglai Wei, Yulun Zhang, Jinjin Gu, and Hanspeter Pfister. Texture-based error analysis for image super-resolution. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2118–2127, 2022

    2022

  35. [40]

    Octopack: Instruction tuning code large language models

    Niklas Muennighoff, Qian Liu, Armel Zebaze, Qinkai Zheng, Binyuan Hui, Terry Yue Zhuo, Swayam Singh, Xiangru Tang, Leandro V on Werra, and Shayne Longpre. Octopack: Instruction tuning code large language models. arXiv preprint arXiv:2308.07124, 2023

    2023 arXiv

  36. [41]

    Training language models to follow instructions with human feedback

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. Training language models to follow instructions with human feedback. Advances in neural information processing systems, 35:2773...

    2022

  37. [42]

    Generative agents: Interactive simulacra of human behavior

    Joon Sung Park, Joseph O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S Bernstein. Generative agents: Interactive simulacra of human behavior. In Proceedings of the 36th annual acm symposium on user interface software and technology, pages 1–22, 2023. 13

    2023

  38. [43]

    Code generation with alphacodium: From prompt engineering to flow engineering

    Tal Ridnik, Dedy Kredo, and Itamar Friedman. Code generation with alphacodium: From prompt engineering to flow engineering. arXiv preprint arXiv:2401.08500, 2024

    2024 arXiv

  39. [44]

    Hugginggpt: Solving ai tasks with chatgpt and its friends in hugging face

    Yongliang Shen, Kaitao Song, Xu Tan, Dongsheng Li, Weiming Lu, and Yueting Zhuang. Hugginggpt: Solving ai tasks with chatgpt and its friends in hugging face. Advances in Neural Information Processing Systems, 36, 2024

    2024

  40. [45]

    Reflexion: Language agents with verbal reinforcement learning

    Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. Reflexion: Language agents with verbal reinforcement learning. Advances in Neural Information Processing Systems, 36, 2024

    2024

  41. [46]

    A general reinforcement learning algorithm that masters chess, shogi, and go through self-play

    David Silver, Thomas Hubert, Julian Schrittwieser, Ioannis Antonoglou, Matthew Lai, Arthur Guez, Marc Lanctot, Laurent Sifre, Dharshan Kumaran, Thore Graepel, et al. A general reinforcement learning algorithm that masters chess, shogi, and go through self-play. Science, 362(64...

    2018

  42. [47]

    Llama: Open and efficient foundation language models

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023

    2023 arXiv

  43. [48]

    An overview of image caption generation methods

    Haoran Wang, Yue Zhang, and Xiaosheng Yu. An overview of image caption generation methods. Computational intelligence and neuroscience, 2020(1):3062706, 2020

    2020

  44. [49]

    A survey on large language model based autonomous agents

    Lei Wang, Chen Ma, Xueyang Feng, Zeyu Zhang, Hao Yang, Jingsen Zhang, Zhiyuan Chen, Jiakai Tang, Xu Chen, Yankai Lin, et al. A survey on large language model based autonomous agents. Frontiers of Computer Science, 18(6):186345, 2024

    2024

  45. [50]

    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 Proceedings of the IEEE/CVF international conference on computer vision, pages 1905–1914, 2021

    1905

  46. [51]

    Chain-of-thought prompting elicits reasoning in large language 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 language models. Advances in neural information processing systems, 35:24824–24837, 2022

    2022

  47. [52]

    Visual chatgpt: Talking, drawing and editing with visual foundation models

    Chenfei Wu, Shengming Yin, Weizhen Qi, Xiaodong Wang, Zecheng Tang, and Nan Duan. Visual chatgpt: Talking, drawing and editing with visual foundation models. arXiv preprint arXiv:2303.04671, 2023

    2023 arXiv

  48. [53]

    Q-bench: A benchmark for general-purpose foundation models on low-level vision

    Haoning Wu, Zicheng Zhang, Erli Zhang, Chaofeng Chen, Liang Liao, Annan Wang, Chunyi Li, Wenxiu Sun, Qiong Yan, Guangtao Zhai, et al. Q-bench: A benchmark for general-purpose foundation models on low-level vision. arXiv preprint arXiv:2309.14181, 2023

    2023 arXiv

  49. [54]

    Q-align: Teaching lmms for visual scoring via discrete text-defined levels

    Haoning Wu, Zicheng Zhang, Weixia Zhang, Chaofeng Chen, Liang Liao, Chunyi Li, Yixuan Gao, Annan Wang, Erli Zhang, Wenxiu Sun, et al. Q-align: Teaching lmms for visual scoring via discrete text-defined levels. arXiv preprint arXiv:2312.17090, 2023

    2023 arXiv

  50. [55]

    Towards open-ended visual quality comparison

    Haoning Wu, Hanwei Zhu, Zicheng Zhang, Erli Zhang, Chaofeng Chen, Liang Liao, Chunyi Li, Annan Wang, Wenxiu Sun, Qiong Yan, et al. Towards open-ended visual quality comparison. arXiv preprint arXiv:2402.16641, 2024

    2024 arXiv

  51. [56]

    The rise and potential of large language model based agents: A survey

    Zhiheng Xi, Wenxiang Chen, Xin Guo, Wei He, Yiwen Ding, Boyang Hong, Ming Zhang, Junzhe Wang, Senjie Jin, Enyu Zhou, et al. The rise and potential of large language model based agents: A survey. arXiv preprint arXiv:2309.07864, 2023

    2023 arXiv

  52. [57]

    Tree of thoughts: Deliberate problem solving with large language models

    Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Tom Griffiths, Yuan Cao, and Karthik Narasimhan. Tree of thoughts: Deliberate problem solving with large language models. Advances in Neural Information Processing Systems, 36, 2024

    2024

  53. [58]

    React: Synergizing reasoning and acting in language models

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. React: Synergizing reasoning and acting in language models. arXiv preprint arXiv:2210.03629, 2022

    2022 arXiv

  54. [59]

    Teaching large language models to regress accurate image quality scores using score distribution

    Zhiyuan You, Xin Cai, Jinjin Gu, Tianfan Xue, and Chao Dong. Teaching large language models to regress accurate image quality scores using score distribution. arXiv preprint arXiv:2501.11561, 2025

    2025

  55. [60]

    Descriptive image quality assessment in the wild

    Zhiyuan You, Jinjin Gu, Zheyuan Li, Xin Cai, Kaiwen Zhu, Tianfan Xue, and Chao Dong. Descriptive image quality assessment in the wild. arXiv preprint arXiv:2405.18842, 2024

    2024

  56. [61]

    Depicting beyond scores: Advancing image quality assessment through multi-modal language models

    Zhiyuan You, Zheyuan Li, Jinjin Gu, Zhenfei Yin, Tianfan Xue, and Chao Dong. Depicting beyond scores: Advancing image quality assessment through multi-modal language models. arXiv preprint arXiv:2312.08962, 2023. 14

    2023 arXiv

  57. [63]

    Crafting a toolchain for image restoration by deep reinforcement learning

    Ke Yu, Chao Dong, Liang Lin, and Chen Change Loy. Crafting a toolchain for image restoration by deep reinforcement learning. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2443–2452, 2018

    2018

  58. [64]

    Systematic image processing for diagnosing brain tumors: A type-ii fuzzy expert system approach

    MH Fazel Zarandi, Marzie Zarinbal, and Mina Izadi. Systematic image processing for diagnosing brain tumors: A type-ii fuzzy expert system approach. Applied soft computing, 11(1):285–294, 2011

    2011

  59. [65]

    Beyond a gaussian denoiser: Residual learning of deep cnn for image denoising

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

    2017

  60. [67]

    Seal: A framework for systematic evaluation of real-world super-resolution

    Wenlong Zhang, Xiaohui Li, Xiangyu Chen, Yu Qiao, Xiao-Ming Wu, and Chao Dong. Seal: A framework for systematic evaluation of real-world super-resolution. arXiv preprint arXiv:2309.03020, 2023

    2023 arXiv

  61. [68]

    A closer look at blind super-resolution: Degradation models, baselines, and performance upper bounds

    Wenlong Zhang, Guangyuan Shi, Yihao Liu, Chao Dong, and Xiao-Ming Wu. A closer look at blind super-resolution: Degradation models, baselines, and performance upper bounds. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 527–536, 2022

    2022

  62. [70]

    Ghost in the minecraft: Generally capable agents for open-world environments via large language models with text-based knowledge and memory

    Xizhou Zhu, Yuntao Chen, Hao Tian, Chenxin Tao, Weijie Su, Chenyu Yang, Gao Huang, Bin Li, Lewei Lu, Xiaogang Wang, et al. Ghost in the minecraft: Generally capable agents for open-world environments via large language models with text-based knowledge and memory. arXiv preprin...

    2023 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.