Pith. sign in

REVIEW 3 major objections 5 minor 3 cited by

The paper claims that a no-reference image quality model can be trained to verify its own uncertainty by cropping into image regions before committing to a final score and rationale.

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

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

T0 review · deepseek-v4-flash

2026-08-03 12:25 UTC pith:N4QNQXHX

load-bearing objection A serious, well-engineered interactive IQA method that deserves referee time, but its headline 'reliable reasoning' claim rests on same-family VLM judges and the score gains over strong baselines are modest. the 3 major comments →

arxiv 2601.02918 v3 pith:N4QNQXHX submitted 2026-01-06 cs.CV

Zoom-IQA: Image Quality Assessment with Reliable Region-Aware Reasoning

Guoqiang Liang , Jianyi Wang , Zhonghua Wu , Shangchen Zhou , Chen Change Loy This is my paper
classification cs.CV
keywords image quality assessmentno-reference IQAvision language modelsreinforcement learningregion-aware reasoningchain-of-thoughtgrounded rationalesimage restoration
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Zoom-IQA sets out to show that image quality assessment should not be a one-shot numerical or textual prediction. The paper's claim is that a vision-language model can be taught to emulate a human verifier: first form a holistic impression, identify the region whose quality is uncertain, zoom into it, and only then issue a final score and rationale. To make this possible, the authors construct the Grounded-Rationale-IQA (GR-IQA) dataset of grounded reasoning trajectories and train in two stages — supervised fine-tuning to learn the zoom-and-verify format, then reinforcement learning to learn when zooming is worth doing. The reported experiments indicate that the resulting model matches or improves on score-regression baselines across common benchmarks and produces reasoning that automated judges rate as more accurate, complete, and confident, and that the same reasoning improves text-guided image restoration. A sympathetic reader would care because the work attacks a practical gap: existing IQA models either give numbers without explanation or give fluent text that can be visually unsupported.

Core claim

The central claim is that reliable IQA reasoning can be learned without exhaustively labeled evidence regions. Zoom-IQA is trained so that, on a hard case, it emits a structured rationale ending in a bounding box and a request to crop; the environment returns that crop as new visual input; and the model then writes a second rationale that confirms, upgrades, or downgrades its initial rating. This hypothesize-crop-verify loop replaces single-pass text-only reasoning. The paper argues that grounding rationales in actual image regions reduces hallucination, that the two-stage training pipeline (SFT on GR-IQA followed by GRPO with the KL-Coverage regularizer) teaches the model both how to zoom a

What carries the argument

The load-bearing mechanism is the iterative hypothesize-crop-verify loop. The model first produces a four-part textual rationale (quality summary, directions for improvement, issues to avoid, decision and rationale) together with a JSON answer containing a tool choice ('crop' or 'final'), a rating, and an optional bounding box; if 'crop' is chosen, the image is actually cropped and re-fed as context for a second turn ending in a final rating. Two supporting mechanisms carry the training: the GR-IQA dataset, filtered by Visual Reliance Filtering (VRF), which discards trajectories that do not depend on the image, and Hint-Augmented Consistency Filtering (HACF), which drops rationales inconsist

Load-bearing premise

The reasoning-reliability advantage depends on the assumption that the AI judges' ratings are a faithful proxy for human judgments, even though those judges come from the same model family as the teacher that wrote the training rationales and no human validation is provided.

What would settle it

Run Zoom-IQA on a held-out set with the crop tool disabled and with the crop tool forced to use a random crop; if final ratings and judged reasoning accuracy do not change, the learned zoom policy is not doing the claimed work. Then have human raters rank anonymized reasoning outputs from Zoom-IQA against single-pass baselines; if human preference does not reproduce the ordering given by the AI judges, the reliability claim is unsupported.

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

If this is right

  • IQA no longer needs to be one-shot: a single model can self-select magnified regions to inspect, which should reduce hallucinated descriptions and give users a concrete visual basis for each rating.
  • Because the rationale and score are produced together, the output can serve as a verifiable prompt for text-guided restoration, letting the restoration model target the exact flaws identified instead of applying generic enhancement.
  • The KL-Coverage regularizer directly addresses a known failure mode of RL-trained IQA models — score collapse — so reinforcement learning can explore diverse reasoning paths without narrowing the predicted rating distribution.
  • The GR-IQA curation pipeline offers a way to build grounded reasoning training data without expensive human annotations of evidence regions, using visual-reliance and consistency filtering to remove hallucinated trajectories.
  • If the interactive policy generalizes, the model should retain score-regression accuracy across in-the-wild and synthetic distortion datasets while adding explainable failure analysis for out-of-distribution images.

Where Pith is reading between the lines

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

  • Editorial inference: The hypothesize-crop-verify loop is a general template for any holistic perceptual judgment — video quality, audio quality, medical image assessment — where the final score aggregates over several interacting regions; the same two-stage SFT+RL recipe could be reused wherever grounded rationales are scarce.
  • Editorial inference: The KL-Coverage regularizer's focus on numerical tokens in the score answer suggests a reusable fix for reward hacking in any RL-trained regressor that outputs a scalar, not just IQA.
  • Editorial inference: The paper's reasoning-reliability claim rests on VLM-as-judge scores from the same model family as the teacher that generated the training rationales; without human validation, the improvement in judged reasoning quality could reflect stylistic preference for Gemini-like text rather than true reasoning fidelity. The paper itself notes in the Supplementary Material that closed
  • Editorial inference: A crisp testable extension is to ablate the zoom tool at inference — comparing model-selected crops against random crops or no crop — to quantify how much of the gain comes from the region-selection policy rather than from extra inference-time compute.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes Zoom-IQA, a VLM-based no-reference IQA method that performs iterative, region-aware reasoning in a "hypothesize-crop-verify" manner. The authors curate a Grounded-Rationale-IQA (GR-IQA) dataset from Gemini-2.5-pro outputs filtered by visual-reliance and consistency checks, then fine-tune Qwen2.5-VL-7B via SFT on this dataset, followed by GRPO reinforcement learning with a newly proposed KL-Coverage regularizer and a progressive re-sampling strategy. Experiments report score regression on seven IQA benchmarks, VLM-as-judge reasoning quality on KonIQ and SPAQ, and qualitative results for text-guided image restoration. The paper claims improved robustness, explainability, and generalization, and "consistently superior" performance over reasoning-IQA baselines.

Significance. If validated, Zoom-IQA would be a meaningful advance: it introduces a trainable, interactive visual-grounding mechanism for IQA, goes beyond static descriptions by localizing quality-relevant regions, and provides a reusable pipeline (GR-IQA dataset, two-stage training, KL-Coverage objective) that could benefit explainable perceptual assessment. The dataset and training recipe are concrete, reproducible contributions. However, the current evidence for the paper's central claim — that the region-aware reasoning is more reliable — is weakened by evaluation design and reporting gaps, so the significance is conditional.

major comments (3)
  1. [Sec. 4.2, Table 1] The text states that Zoom-IQA achieves "consistently superior performance" against reasoning baselines. Table 1 does not support this. On SPAQ, Zoom-IQA trails Q-Insight (0.902/0.900 vs 0.903/0.903); on KADID it trails both Q-Insight (0.701/0.700 vs 0.702/0.702) and VisualQuality-R1 (0.703/0.712); on AGIQA the results are effectively tied with Q-Insight and slightly below VisualQuality-R1 on PLCC. No error bars, confidence intervals, or significance tests are reported. The differences are small and within typical noise. The claim should be tempered to "comparable or moderately better on most benchmarks" or supported with statistical analysis.
  2. [Sec. 3.1.1 and Sec. 4.2, Table 2] The reasoning-quality evaluation is not convincing as a measure of reliability. GR-IQA rationales were generated by Gemini-2.5-pro, and the primary evaluator is Gemini-2.5-Flash — a sibling model. The VLM judge may reward stylistic mimicry of the teacher format rather than factual grounding. The supplementary prompt for the judge (Table 7) defines Confidence as assertiveness "regardless of correctness," which is not a reliability property. There is no human validation of the reasoning chains, bounding-box localization, or the causal link between cropping and the final rating. The claim "superiority of our reasoning reliability" in Sec. 4.2 is therefore not established by Table 2. The authors should add a human study or, at minimum, an independent evaluation that checks whether the emitted bboxes match artifact regions and whether the described degradations are visually present.
  3. [Sec. 4.2 and Supplementary Sec. 8] The text-guided restoration experiment is used as evidence of downstream utility, but the experimental description is inconsistent. Section 4.2 says the reasoning is fed into SUPIR, whereas Supplementary Section 8 says the authors use DreamClear with SwinIR and LLaVA-1.6. This discrepancy makes the exact setup unclear. Moreover, the evaluation is qualitative only, with a handful of examples and no quantitative metric, human rating, or statistical test. The strong claim of "far superior restoration" is not supported by the presented evidence. The authors should clarify the framework used and provide quantitative or perceptual evaluation.
minor comments (5)
  1. [Sec. 3.2] The Progressive Re-sampling Strategy is described only briefly. The number of stages, the oversampling rates for low-frequency score intervals, and the stopping criterion are not specified. Please provide implementation details.
  2. [Sec. 3.1.1] The Visual Reliance Filtering thresholds (rating difference 0.05, bbox IoU 0.5, entropy 0.01) appear only in the supplementary. The main text should either state them or refer explicitly to the supplemental section.
  3. [Eq. (5)] The notation for preference probabilities is confusing: p_{ij} appears both as the ground-truth preference and as part of the predicted probability p_{k,ij}. Please distinguish the two to avoid ambiguity.
  4. [Table 2] The VLM-as-judge results do not report sample sizes, standard deviations, or inter-rater agreement. Since the evaluation is based on a small sample (implied by qualitative examples), the margins in Table 2 should be contextualized with these statistics.
  5. [Figure 2] The label "KL-Converage Loss" contains a typo; it should be "KL-Coverage Loss."

Circularity Check

0 steps flagged

No load-bearing circularity found; training and evaluation are anchored to external MOS or independent judge protocols, and the author self-citations are peripheral.

full rationale

Zoom-IQA's derivation is a training pipeline rather than a formal theorem. The score-regression evidence (Table 1) is anchored to external MOS datasets (KonIQ, SPAQ, KADID, PIPAL, LIVE-Wild, AGIQA, CSIQ), using the same KonIQ-trained protocol for all baselines; no fitted parameter is renamed as a prediction, and no equation defines a claimed output in terms of its training target. GR-IQA rationales are generated by Gemini-2.5-pro and filtered by VRF/HACF, but the reasoning-quality evaluation also uses a different-family judge (GPT-5-mini), and the RL score and rank rewards are grounded in ground-truth MOS rather than in the model's own text. The author self-citations [59,61,62] appear only as background references for CLIP-IQA and restoration backbones, not as load-bearing support for the central claim. The concern that Gemini-2.5-Flash may prefer Gemini-style text is a legitimate evaluation-validity limitation, but it is not a circular reduction: the Table 2 margins are empirical measurements, not identities, and the external MOS benchmarks provide an independent anchor for the method's score-prediction claims. Therefore, no specific circular step can be quoted or exhibited.

Axiom & Free-Parameter Ledger

7 free parameters · 5 axioms · 0 invented entities

The central claim rests on 5 domain/algorithm assumptions and 7 fitted or hand-set hyperparameters. No invented physical entities. The heaviest assumptions are the reliability of Gemini-generated rationales and the VLM-as-judge evaluation.

free parameters (7)
  • Score-reward width σ = 0.35
    Controls sensitivity of the Gaussian score reward (Eq. 4); chosen without reported sensitivity analysis.
  • Reward weights α, β = α=1, β=2
    Balance score and rank rewards in Eq. 6; hand-set.
  • KL coefficient β_KL = 0.04
    GRPO/KL penalty coefficient in training; reported in implementation details.
  • KL-Coverage mask fraction p = 0.02
    Fraction of numerical tokens with highest covariance that receive the KL-Coverage penalty (Eq. 2).
  • GRPO group size N = 8
    Number of sampled responses per prompt for advantage estimation in GRPO.
  • VRF thresholds = rating diff 0.05, bbox IoU 0.5, entropy diff 0.01
    Discard thresholds in Visual Reliance Filtering; specified in Supp Sec. 6.2.
  • Progressive re-sampling schedule = unspecified
    Multi-stage sampling weights for underrepresented score intervals are described qualitatively, no exact schedule given.
axioms (5)
  • domain assumption MOS scores in KonIQ/SPAQ/etc. are valid ground truth for image quality
    All score and rank rewards are computed against dataset MOS; if MOS is noisy, rewards are noisy.
  • domain assumption Gemini-2.5-pro generated rationales, after VRF/HACF filtering, are reliable supervision for grounded IQA reasoning
    SFT on GR-IQA builds on this; only heuristic consistency checks are used, no human verification of rationales.
  • domain assumption VLM-as-judge scores (Gemini-2.5-Flash/GPT-5-mini) measure reasoning reliability
    Table 2 claims rely entirely on these closed models; no human study validates the judge.
  • domain assumption Cropping and zooming is necessary for accurate IQA
    Motivates the whole method; cited DiffIQA but not rigorously demonstrated.
  • standard math GRPO with the specified rewards improves the policy without degenerate collapse
    Standard RL algorithm; KL-Coverage regularizer added because raw GRPO collapses.

pith-pipeline@v1.3.0-alltime-deepseek · 28401 in / 13926 out tokens · 125562 ms · 2026-08-03T12:25:25.023470+00:00 · methodology

0 comments
read the original abstract

Image Quality Assessment (IQA) is a long-standing problem in computer vision. Previous methods typically focus on predicting numerical scores without explanation or providing low-level descriptions lacking precise scores. Recent reasoning-based vision language models (VLMs) have shown strong potential for IQA by jointly generating quality descriptions and scores. However, existing VLM-based IQA methods often suffer from unreliable reasoning due to their limited capability of integrating visual and textual cues. In this work, we introduce Zoom-IQA, a VLM-based IQA model to explicitly emulate key cognitive behaviors: uncertainty awareness, region reasoning, and iterative refinement. Specifically, we present a two-stage training pipeline: 1) supervised fine-tuning (SFT) on our Grounded-Rationale-IQA (GR-IQA) dataset to teach the model to ground its assessments in key regions, and 2) reinforcement learning (RL) for dynamic policy exploration, stabilized by our KL-Coverage regularizer to prevent reasoning and scoring diversity collapse, with a Progressive Re-sampling Strategy for mitigating annotation bias. Extensive experiments show that Zoom-IQA achieves improved robustness, explainability, and generalization. The application to downstream tasks, such as image restoration, further demonstrates the effectiveness of Zoom-IQA.

Figures

Figures reproduced from arXiv: 2601.02918 by Chen Change Loy, Guoqiang Liang, Jianyi Wang, Shangchen Zhou, Zhonghua Wu.

Figure 1
Figure 1. Figure 1: (Upper) Current IQA methods are non-interactive, leading to inferior assessments. They either spot only partial flaws (e.g., slightly overexposed or slightly blurred) or make factually incorrect claims (clear and well-lit), resulting in erroneous judgments. Our Zoom-IQA uses interactive, region-aware reasoning: it first hypothesizes flaws (green text), then grounds them by cropping (orange text), and final… view at source ↗
Figure 2
Figure 2. Figure 2: An overview of our two-stage framework. Stage (1), Grounded Quality Rationale Learning (Sec. 3.1), first uses SFT to teach the model how to correctly execute the crop action. Stage (2), Self-Guided Exploration (Sec. 3.2), then uses RL to let the model learn what to crop, allowing it to discover regions that lead to a deeper understanding of image quality. Data Generation. We prompt the closed-source VLM, G… view at source ↗
Figure 3
Figure 3. Figure 3: The GR-IQA dataset curation pipeline. It uses [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison of Zoom-IQA with competing methods (Q-insight [ [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative evaluation of reasoning quality on the image restoration task. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Example of filtered-out data in Visual Reliance Filtering. [PITH_FULL_IMAGE:figures/full_fig_p018_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative comparison of Zoom-IQA against competing methods (Q-Insight [ [PITH_FULL_IMAGE:figures/full_fig_p019_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative comparison of Zoom-IQA against competing methods (Q-Insight [ [PITH_FULL_IMAGE:figures/full_fig_p020_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative comparison of Zoom-IQA against competing methods (Q-Insight [ [PITH_FULL_IMAGE:figures/full_fig_p021_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative comparison of Zoom-IQA with competing methods (Q-Insight [ [PITH_FULL_IMAGE:figures/full_fig_p022_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Qualitative comparison of Zoom-IQA with competing methods (Q-Insight [ [PITH_FULL_IMAGE:figures/full_fig_p023_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Qualitative comparison of Zoom-IQA with competing methods (Q-Insight [ [PITH_FULL_IMAGE:figures/full_fig_p024_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Qualitative comparison of Zoom-IQA with competing methods (Q-Insight [ [PITH_FULL_IMAGE:figures/full_fig_p025_13.png] view at source ↗

discussion (0)

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

Forward citations

Cited by 3 Pith papers

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

  1. MR-IQA: A Unified Margin View of Regression and Ranking for Blind Image Quality Assessment

    cs.CV 2026-06 unverdicted novelty 6.0

    MR-IQA unifies regression and ranking in BIQA via a quality-margin optimization framework in RL, showing competitive performance on six benchmarks.

  2. MR-IQA: A Unified Margin View of Regression and Ranking for Blind Image Quality Assessment

    cs.CV 2026-06 unverdicted novelty 5.0

    Regression and ranking in BIQA both fit quality margins, enabling a unified margin optimization framework called MR-IQA for RL-based assessment.

  3. Q-DeepSight: Incentivizing Thinking with Images for Image Quality Assessment and Refinement

    cs.CV 2026-04 unverdicted novelty 5.0

    Q-DeepSight proposes a think-with-image multimodal CoT framework trained via RL with perceptual curriculum rewards and evidence gradient filtering to achieve SOTA IQA performance and enable training-free perceptual re...

Reference graph

Works this paper leans on

102 extracted references · 27 linked inside Pith · cited by 2 Pith papers

  1. [1]

    Dream- clear: High-capacity real-world image restoration with privacy-safe dataset curation.Proceedings of Advances in Neural Information Processing Systems (NeurIPS), 2024

    Yuang Ai, Xiaoqiang Zhou, Huaibo Huang, Xiaotian Han, Zhengyu Chen, Quanzeng You, and Hongxia Yang. Dream- clear: High-capacity real-world image restoration with privacy-safe dataset curation.Proceedings of Advances in Neural Information Processing Systems (NeurIPS), 2024. 1

    2024

  2. [2]

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, et al. Qwen2. 5-VL technical report.arXiv preprint arXiv:2502.13923, 2025. 2, 3, 6

    Pith/arXiv arXiv 2025

  3. [3]

    Deep neural net- works for no-reference and full-reference image quality as- sessment.IEEE Transactions on Image Processing (TIP),

    Sebastian Bosse, Dominique Maniry, Klaus-Robert M ¨uller, Thomas Wiegand, and Wojciech Samek. Deep neural net- works for no-reference and full-reference image quality as- sessment.IEEE Transactions on Image Processing (TIP),

  4. [4]

    DSPO: Direct semantic pref- erence optimization for real-world image super-resolution

    Miaomiao Cai, Simiao Li, Wei Li, Xudong Huang, Hanting Chen, Jie Hu, and Yunhe Wang. DSPO: Direct semantic pref- erence optimization for real-world image super-resolution. arXiv preprint arXiv:2504.15176, 2025. 2

    Pith/arXiv arXiv 2025

  5. [5]

    Q-Ponder: A unified train- ing pipeline for reasoning-based visual quality assessment

    Zhuoxuan Cai, Jian Zhang, Xinbin Yuan, Peng-Tao Jiang, Wenxiang Chen, Bowen Tang, Lujian Yao, Qiyuan Wang, Jinwen Chen, and Bo Li. Q-Ponder: A unified train- ing pipeline for reasoning-based visual quality assessment. arXiv preprint arXiv:2506.05384, 2025. 3

    Pith/arXiv arXiv 2025

  6. [6]

    Q-ground: Image quality grounding with large multi-modality models

    Chaofeng Chen, Sensen Yang, Haoning Wu, Liang Liao, Zicheng Zhang, Annan Wang, Wenxiu Sun, Qiong Yan, and Weisi Lin. Q-ground: Image quality grounding with large multi-modality models. InProceedings of the ACM Interna- tional Conference on Multimedia (ACM MM), 2024. 2

    2024

  7. [7]

    Mllm-as-a-judge: Assessing multimodal llm-as-a-judge with vision-language benchmark

    Dongping Chen, Ruoxi Chen, Shilin Zhang, Yaochen Wang, Yinuo Liu, Huichi Zhou, Qihui Zhang, Yao Wan, Pan Zhou, and Lichao Sun. Mllm-as-a-judge: Assessing multimodal llm-as-a-judge with vision-language benchmark. InPro- ceedings of International Conference on Machine Learning (ICML), 2024. 7

    2024

  8. [8]

    Toward gen- eralized image quality assessment: Relaxing the perfect ref- erence quality assumption

    Du Chen, Tianhe Wu, Kede Ma, and Lei Zhang. Toward gen- eralized image quality assessment: Relaxing the perfect ref- erence quality assumption. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025. 2

    2025

  9. [9]

    Grounding-IQA: Multimodal lan- guage grounding model for image quality assessment.arXiv preprint arXiv:2411.17237, 2024

    Zheng Chen, Xun Zhang, Wenbo Li, Renjing Pei, Feng- long Song, Xiongkuo Min, Xiaohong Liu, Xin Yuan, Yong Guo, and Yulun Zhang. Grounding-IQA: Multimodal lan- guage grounding model for image quality assessment.arXiv preprint arXiv:2411.17237, 2024. 2, 3

    arXiv 2024

  10. [10]

    Reason- ing with exploration: An entropy perspective.arXiv preprint arXiv:2506.14758, 2025

    Daixuan Cheng, Shaohan Huang, Xuekai Zhu, Bo Dai, Wayne Xin Zhao, Zhenliang Zhang, and Furu Wei. Reason- ing with exploration: An entropy perspective.arXiv preprint arXiv:2506.14758, 2025. 5

    Pith/arXiv arXiv 2025

  11. [11]

    Perceptual image quality assessment with trans- formers

    Manri Cheon, Sung-Jun Yoon, Byungyeon Kang, and Jun- woo Lee. Perceptual image quality assessment with trans- formers. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021. 3

    2021

  12. [12]

    Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities.arXiv preprint arXiv:2507.06261, 2025

    Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, Inderjit Dhillon, Marcel Blis- tein, Ori Ram, Dan Zhang, Evan Rosen, et al. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities.arXiv preprint arXiv:2507.06261, 2025. 2, 4, 7, 1

    Pith/arXiv arXiv 2025

  13. [13]

    The entropy mechanism of reinforce- ment learning for reasoning language models.arXiv preprint arXiv:2505.22617, 2025

    Ganqu Cui, Yuchen Zhang, Jiacheng Chen, Lifan Yuan, Zhi Wang, Yuxin Zuo, Haozhan Li, Yuchen Fan, Huayu Chen, Weize Chen, et al. The entropy mechanism of reinforce- ment learning for reasoning language models.arXiv preprint arXiv:2505.22617, 2025. 5

    Pith/arXiv arXiv 2025

  14. [14]

    Diffusion models beat gans on image synthesis.Proceedings of Advances in Neural Information Processing Systems (NeurIPS), 2021

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis.Proceedings of Advances in Neural Information Processing Systems (NeurIPS), 2021. 2

    2021

  15. [15]

    Perceptual quality assessment of smartphone pho- tography

    Yuming Fang, Hanwei Zhu, Yan Zeng, Kede Ma, and Zhou Wang. Perceptual quality assessment of smartphone pho- tography. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020. 6, 1

    2020

  16. [16]

    Live in the wild image quality challenge database.Online: http://live

    Deepti Ghadiyaram and Alan C Bovik. Live in the wild image quality challenge database.Online: http://live. ece. utexas. edu/research/ChallengeDB/index. html [Mar, 2017],

    2017

  17. [17]

    Videoscore: Building automatic metrics to simulate fine-grained human feedback for video genera- tion.arXiv preprint arXiv:2406.15252, 2024

    Xuan He, Dongfu Jiang, Ge Zhang, Max Ku, Achint Soni, Sherman Siu, Haonan Chen, Abhranil Chandra, Ziyan Jiang, Aaran Arulraj, et al. Videoscore: Building automatic metrics to simulate fine-grained human feedback for video genera- tion.arXiv preprint arXiv:2406.15252, 2024. 2

    Pith/arXiv arXiv 2024

  18. [18]

    Koniq-10k: An ecologically valid database for deep learning of blind image quality assessment.IEEE Transactions on Image Processing (TIP), 2020

    Vlad Hosu, Hanhe Lin, Tamas Sziranyi, and Dietmar Saupe. Koniq-10k: An ecologically valid database for deep learning of blind image quality assessment.IEEE Transactions on Image Processing (TIP), 2020. 4, 5, 6, 1

    2020

  19. [19]

    Vision-r1: Incentivizing reasoning capability in multimodal large language models.arXiv preprint arXiv:2503.06749,

    Wenxuan Huang, Bohan Jia, Zijie Zhai, Shaosheng Cao, Zheyu Ye, Fei Zhao, Zhe Xu, Yao Hu, and Shaohui Lin. Vision-r1: Incentivizing reasoning capability in multimodal large language models.arXiv preprint arXiv:2503.06749,

    Pith/arXiv arXiv

  20. [20]

    Self-introspective de- coding: Alleviating hallucinations for large vision-language models.arXiv preprint arXiv:2408.02032, 2024

    Fushuo Huo, Wenchao Xu, Zhong Zhang, Haozhao Wang, Zhicheng Chen, and Peilin Zhao. Self-introspective de- coding: Alleviating hallucinations for large vision-language models.arXiv preprint arXiv:2408.02032, 2024. 1

    Pith/arXiv arXiv 2024

  21. [21]

    VLM-R3: Region recognition, reasoning, and refinement for enhanced multimodal chain-of-thought.arXiv preprint arXiv:2505.16192, 2025

    Chaoya Jiang, Yongrui Heng, Wei Ye, Han Yang, Haiyang Xu, Ming Yan, Ji Zhang, Fei Huang, and Shikun Zhang. VLM-R3: Region recognition, reasoning, and refinement for enhanced multimodal chain-of-thought.arXiv preprint arXiv:2505.16192, 2025. 3

    Pith/arXiv arXiv 2025

  22. [22]

    Mme-cot: Benchmarking chain-of-thought in large multimodal models for reasoning quality, robustness, and efficiency.arXiv preprint arXiv:2502.09621, 2025

    Dongzhi Jiang, Renrui Zhang, Ziyu Guo, Yanwei Li, Yu Qi, Xinyan Chen, Liuhui Wang, Jianhan Jin, Claire Guo, Shen Yan, et al. Mme-cot: Benchmarking chain-of-thought in large multimodal models for reasoning quality, robustness, and efficiency.arXiv preprint arXiv:2502.09621, 2025. 7

    Pith/arXiv arXiv 2025

  23. [23]

    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. InProceedings of the European Conference on Computer Vi- sion (ECCV), 2020. 6

    2020

  24. [24]

    Convo- lutional neural networks for no-reference image quality as- sessment

    Le Kang, Peng Ye, Yi Li, and David Doermann. Convo- lutional neural networks for no-reference image quality as- sessment. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2014. 3

    2014

  25. [25]

    Musiq: Multi-scale image quality transformer

    Junjie Ke, Qifei Wang, Yilin Wang, Peyman Milanfar, and Feng Yang. Musiq: Multi-scale image quality transformer. 9 InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021. 3, 7

    2021

  26. [26]

    Most apparent distortion: full-reference image quality assessment and the role of strategy.Journal of Electronic Imaging, 2010

    Eric Cooper Larson and Damon Michael Chandler. Most apparent distortion: full-reference image quality assessment and the role of strategy.Journal of Electronic Imaging, 2010. 6

    2010

  27. [27]

    Mitigating object hallucinations in large vision-language models through vi- sual contrastive decoding

    Sicong Leng, Hang Zhang, Guanzheng Chen, Xin Li, Shi- jian Lu, Chunyan Miao, and Lidong Bing. Mitigating object hallucinations in large vision-language models through vi- sual contrastive decoding. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024. 2, 4, 1

    2024

  28. [28]

    Agiqa-3k: An open database for ai-generated image quality assessment.IEEE Transactions on Circuits and Sys- tems for Video Technology (TCSVT), 2023

    Chunyi Li, Zicheng Zhang, Haoning Wu, Wei Sun, Xiongkuo Min, Xiaohong Liu, Guangtao Zhai, and Weisi Lin. Agiqa-3k: An open database for ai-generated image quality assessment.IEEE Transactions on Circuits and Sys- tems for Video Technology (TCSVT), 2023. 6

    2023

  29. [29]

    Q-insight: Understanding im- age quality via visual reinforcement learning.arXiv preprint arXiv:2503.22679, 2025

    Weiqi Li, Xuanyu Zhang, Shijie Zhao, Yabin Zhang, Junlin Li, Li Zhang, and Jian Zhang. Q-insight: Understanding im- age quality via visual reinforcement learning.arXiv preprint arXiv:2503.22679, 2025. 2, 3, 6, 7, 8, 1, 9, 10, 11, 12, 13

    Pith/arXiv arXiv 2025

  30. [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. InProceedings of the IEEE/CVF Inter- national Conference on Computer Vision Workshops (ICCV- W), 2021. 1

    2021

  31. [31]

    Kadid-10k: A large-scale artificially distorted iqa database

    Hanhe Lin, Vlad Hosu, and Dietmar Saupe. Kadid-10k: A large-scale artificially distorted iqa database. In2019 Eleventh International Conference on Quality of Multimedia Experience (QoMEX), 2019. 6

    2019

  32. [32]

    Visual instruction tuning.Proceedings of Advances in Neural Information Processing Systems (NeurIPS), 2023

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning.Proceedings of Advances in Neural Information Processing Systems (NeurIPS), 2023. 1, 2, 3, 8, 7, 9

    2023

  33. [33]

    Dog- iqa: Standard-guided zero-shot mllm for mix-grained image quality assessment.arXiv preprint arXiv:2410.02505, 2024

    Kai Liu, Ziqing Zhang, Wenbo Li, Renjing Pei, Fenglong Song, Xiaohong Liu, Linghe Kong, and Yulun Zhang. Dog- iqa: Standard-guided zero-shot mllm for mix-grained image quality assessment.arXiv preprint arXiv:2410.02505, 2024. 3

    Pith/arXiv arXiv 2024

  34. [34]

    Paying more atten- tion to image: A training-free method for alleviating halluci- nation in lvlms

    Shi Liu, Kecheng Zheng, and Wei Chen. Paying more atten- tion to image: A training-free method for alleviating halluci- nation in lvlms. InProceedings of the European Conference on Computer Vision (ECCV), 2024. 2, 4

    2024

  35. [35]

    Rankiqa: Learning from rankings for no-reference image quality assessment

    Xialei Liu, Joost Van De Weijer, and Andrew D Bagdanov. Rankiqa: Learning from rankings for no-reference image quality assessment. InProceedings of the IEEE/CVF Inter- national Conference on Computer Vision (ICCV), 2017. 3

    2017

  36. [36]

    Visual- rft: Visual reinforcement fine-tuning.arXiv preprint arXiv:2503.01785, 2025

    Ziyu Liu, Zeyi Sun, Yuhang Zang, Xiaoyi Dong, Yuhang Cao, Haodong Duan, Dahua Lin, and Jiaqi Wang. Visual- rft: Visual reinforcement fine-tuning.arXiv preprint arXiv:2503.01785, 2025. 2, 3

    Pith/arXiv arXiv 2025

  37. [37]

    Learning a no-reference quality metric for single-image super-resolution.Computer Vision and Image Understanding (CVIU), 2017

    Chao Ma, Chih-Yuan Yang, Xiaokang Yang, and Ming- Hsuan Yang. Learning a no-reference quality metric for single-image super-resolution.Computer Vision and Image Understanding (CVIU), 2017. 3

    2017

  38. [38]

    Mm-eureka: Exploring the frontiers of multimodal reasoning with rule-based reinforce- ment learning.arXiv preprint arXiv:2503.07365, 2025

    Fanqing Meng, Lingxiao Du, Zongkai Liu, Zhixiang Zhou, Quanfeng Lu, Daocheng Fu, Tiancheng Han, Botian Shi, Wenhai Wang, Junjun He, et al. Mm-eureka: Exploring the frontiers of multimodal reasoning with rule-based reinforce- ment learning.arXiv preprint arXiv:2503.07365, 2025. 2, 3

    Pith/arXiv arXiv 2025

  39. [39]

    No-reference image quality assessment in the spa- tial domain.IEEE Transactions on Image Processing (TIP),

    Anish Mittal, Anush Krishna Moorthy, and Alan Conrad Bovik. No-reference image quality assessment in the spa- tial domain.IEEE Transactions on Image Processing (TIP),

  40. [40]

    completely blind

    Anish Mittal, Rajiv Soundararajan, and Alan C Bovik. Mak- ing a “completely blind” image quality analyzer.IEEE Sig- nal processing letters, 2012. 7

    2012

  41. [41]

    A two- step framework for constructing blind image quality indices

    Anush Krishna Moorthy and Alan Conrad Bovik. A two- step framework for constructing blind image quality indices. IEEE Signal Processing Letters, 2010

    2010

  42. [42]

    Blind image quality assessment: From natural scene statistics to perceptual quality.IEEE Transactions on Image Processing (TIP), 2011

    Anush Krishna Moorthy and Alan Conrad Bovik. Blind image quality assessment: From natural scene statistics to perceptual quality.IEEE Transactions on Image Processing (TIP), 2011. 3

    2011

  43. [43]

    Blind predicting similar quality map for image quality assessment

    Da Pan, Ping Shi, Ming Hou, Zefeng Ying, Sizhe Fu, and Yuan Zhang. Blind predicting similar quality map for image quality assessment. InProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR),

  44. [44]

    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. InProceedings of International Conference on Ma- chine Learning (ICML), 2021. 2, 3, 1

    2021

  45. [45]

    Exploring the limits of transfer learning with a unified text-to-text transformer.Journal of Machine Learn- ing Research (JMLR), 2020

    Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer.Journal of Machine Learn- ing Research (JMLR), 2020. 1

    2020

  46. [46]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022. 2

    2022

  47. [47]

    Proximal policy optimization algo- rithms.arXiv preprint arXiv:1707.06347, 2017

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Rad- ford, and Oleg Klimov. Proximal policy optimization algo- rithms.arXiv preprint arXiv:1707.06347, 2017. 5

    Pith/arXiv arXiv 2017

  48. [48]

    Visual cot: Advancing multi-modal language models with a comprehen- sive dataset and benchmark for chain-of-thought reasoning

    Hao Shao, Shengju Qian, Han Xiao, Guanglu Song, Zhuofan Zong, Letian Wang, Yu Liu, and Hongsheng Li. Visual cot: Advancing multi-modal language models with a comprehen- sive dataset and benchmark for chain-of-thought reasoning. Proceedings of Advances in Neural Information Processing Systems (NeurIPS), 2024. 2

    2024

  49. [49]

    Deepseekmath: Pushing the limits of math- ematical reasoning in open language models.arXiv preprint arXiv:2402.03300, 2024

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Yang Wu, et al. Deepseekmath: Pushing the limits of math- ematical reasoning in open language models.arXiv preprint arXiv:2402.03300, 2024. 3, 2

    Pith/arXiv arXiv 2024

  50. [50]

    Vlm-r1: A stable and generaliz- able r1-style large vision-language model.arXiv preprint arXiv:2504.07615, 2025

    Haozhan Shen, Peng Liu, Jingcheng Li, Chunxin Fang, Yibo Ma, Jiajia Liao, Qiaoli Shen, Zilun Zhang, Kangjia Zhao, Qianqian Zhang, et al. Vlm-r1: A stable and generaliz- able r1-style large vision-language model.arXiv preprint arXiv:2504.07615, 2025. 2, 3 10

    Pith/arXiv arXiv 2025

  51. [51]

    Pixel reasoner: Incentivizing pixel-space rea- soning with curiosity-driven reinforcement learning.arXiv preprint arXiv:2505.15966, 2025

    Alex Su, Haozhe Wang, Weiming Ren, Fangzhen Lin, and Wenhu Chen. Pixel reasoner: Incentivizing pixel-space rea- soning with curiosity-driven reinforcement learning.arXiv preprint arXiv:2505.15966, 2025. 3

    Pith/arXiv arXiv 2025

  52. [52]

    Blindly assess image qual- ity in the wild guided by a self-adaptive hyper network

    Shaolin Su, Qingsen Yan, Yu Zhu, Cheng Zhang, Xin Ge, Jinqiu Sun, and Yanning Zhang. Blindly assess image qual- ity in the wild guided by a self-adaptive hyper network. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition (CVPR), 2020. 3, 7

    2020

  53. [53]

    Koniq++: Boosting no-reference image quality assessment in the wild by jointly predicting image quality and defects

    Shaolin Su, Vlad Hosu, Hanhe Lin, Yanning Zhang, and Di- etmar Saupe. Koniq++: Boosting no-reference image quality assessment in the wild by jointly predicting image quality and defects. InProceedings of the British Machine Vision Conference (BMVC), 2021. 1

    2021

  54. [54]

    Graphiqa: Learning distortion graph representations for blind image quality assessment.IEEE Transactions on Multimedia (TMM), 2022

    Simeng Sun, Tao Yu, Jiahua Xu, Wei Zhou, and Zhibo Chen. Graphiqa: Learning distortion graph representations for blind image quality assessment.IEEE Transactions on Multimedia (TMM), 2022. 3

    2022

  55. [55]

    RFSR: Improving isr diffusion models via reward feedback learning

    Xiaopeng Sun, Qinwei Lin, Yu Gao, Yujie Zhong, Chengjian Feng, Dengjie Li, Zheng Zhao, Jie Hu, and Lin Ma. RFSR: Improving isr diffusion models via reward feedback learning. arXiv preprint arXiv:2412.03268, 2024. 2

    Pith/arXiv arXiv 2024

  56. [56]

    Nima: Neural image assessment.IEEE Transactions on Image Processing (TIP),

    Hossein Talebi and Peyman Milanfar. Nima: Neural image assessment.IEEE Transactions on Image Processing (TIP),

  57. [57]

    A law of comparative judgment

    Louis L Thurstone. A law of comparative judgment. InScal- ing, pages 81–92. 2017. 6

    2017

  58. [58]

    Llama: Open and efficient foundation language models

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timoth´ee Lacroix, Baptiste Rozi`ere, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023. 2, 3

    Pith/arXiv arXiv 2023

  59. [60]

    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. InPro- ceedings of the AAAI Conference on Artificial Intelligence,

  60. [61]

    Chan, and Chen Change Loy

    Jianyi Wang, Zongsheng Yue, Shangchen Zhou, Kelvin C.K. Chan, and Chen Change Loy. Exploiting diffusion prior for real-world image super-resolution.International Journal of Computer Vision (IJCV), 2024. 8

    2024

  61. [62]

    Seedvr: Seeding in- finity in diffusion transformer towards generic video restora- tion

    Jianyi Wang, Zhijie Lin, Meng Wei, Yang Zhao, Ceyuan Yang, Chen Change Loy, and Lu Jiang. Seedvr: Seeding in- finity in diffusion transformer towards generic video restora- tion. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), 2025. 8

    2025

  62. [63]

    Unified multimodal chain-of-thought reward model through reinforcement fine- tuning.arXiv preprint arXiv:2505.03318, 2025

    Yibin Wang, Zhimin Li, Yuhang Zang, Chunyu Wang, Qinglin Lu, Cheng Jin, and Jiaqi Wang. Unified multimodal chain-of-thought reward model through reinforcement fine- tuning.arXiv preprint arXiv:2505.03318, 2025. 2

    arXiv 2025

  63. [64]

    Q-instruct: Improving low-level visual abilities for multi-modality foundation models

    Haoning Wu, Zicheng Zhang, Erli Zhang, Chaofeng Chen, Liang Liao, Annan Wang, Kaixin Xu, Chunyi Li, Jingwen Hou, Guangtao Zhai, et al. Q-instruct: Improving low-level visual abilities for multi-modality foundation models. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition (CVPR), 2024. 3

    2024

  64. [65]

    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. InProceedings of International Conference on Machine Learning (ICML),

  65. [66]

    DP2O-SR: Direct perceptual preference optimization for real-world image super-resolution.NeurIPS, 2025

    Rongyuan Wu, Lingchen Sun, Zhengqiang Zhang, Shihao Wang, Tianhe Wu, Qiaosi Yi, Shuai Li, and Lei Zhang. DP2O-SR: Direct perceptual preference optimization for real-world image super-resolution.NeurIPS, 2025. 2

    2025

  66. [67]

    Visualquality-r1: Reasoning-induced image quality assess- ment via reinforcement learning to rank.arXiv preprint arXiv:2505.14460, 2025

    Tianhe Wu, Jian Zou, Jie Liang, Lei Zhang, and Kede Ma. Visualquality-r1: Reasoning-induced image quality assess- ment via reinforcement learning to rank.arXiv preprint arXiv:2505.14460, 2025. 2, 3, 5, 6, 7, 8, 1, 9, 10, 11, 12, 13

    arXiv 2025

  67. [68]

    Scalecap: Inference- time scalable image captioning via dual-modality debiasing

    Long Xing, Qidong Huang, Xiaoyi Dong, Pan Zhang, Yuhang Zang, Yuhang Cao, Jinsong Li, Shuangrui Ding, Weiming Zhang, Nenghai Yu, et al. Scalecap: Inference- time scalable image captioning via dual-modality debiasing. arXiv preprint arXiv:2506.19848, 2025. 1

    Pith/arXiv arXiv 2025

  68. [69]

    Qwen2.5 technical report.arXiv preprint arXiv:2412.15115, 2024

    An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jingren Zhou, Junyang Lin, Kai Dang, Keming Lu, Keqin Bao, Kexin Yang, Le Yu, Mei Li, Mingfeng Xue, Pei Zhang, Qin Zhu, Rui Men, Runji Lin, Tianhao Li, Tingyu X...

    Pith/arXiv arXiv 2024

  69. [70]

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

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

    2022

  70. [71]

    Descriptive image quality assessment in the wild.arXiv preprint arXiv:2405.18842,

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

    arXiv

  71. [72]

    Depicting beyond scores: Ad- vancing image quality assessment through multi-modal lan- guage models

    Zhiyuan You, Zheyuan Li, Jinjin Gu, Zhenfei Yin, Tian- fan Xue, and Chao Dong. Depicting beyond scores: Ad- vancing image quality assessment through multi-modal lan- guage models. InProceedings of the European Conference on Computer Vision (ECCV), 2024. 2, 3, 7

    2024

  72. [73]

    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. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025. 2, 3, 7

    2025

  73. [74]

    Perception-r1: Pioneering perception policy with reinforcement learning.arXiv preprint arXiv:2504.07954,

    En Yu, Kangheng Lin, Liang Zhao, Jisheng Yin, Yana Wei, Yuang Peng, Haoran Wei, Jianjian Sun, Chunrui Han, Zheng Ge, et al. Perception-r1: Pioneering perception policy with reinforcement learning.arXiv preprint arXiv:2504.07954,

    Pith/arXiv arXiv

  74. [75]

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

    Fanghua Yu, Jinjin Gu, Zheyuan Li, Jinfan Hu, Xiangtao Kong, Xintao Wang, Jingwen He, Yu Qiao, and Chao Dong. Scaling up to excellence: Practicing model scaling for photo- realistic image restoration in the wild. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024. 1, 8

    2024

  75. [76]

    R1-vl: Learn- ing to reason with multimodal large language models via step-wise group relative policy optimization.arXiv preprint arXiv:2503.12937, 2025

    Jingyi Zhang, Jiaxing Huang, Huanjin Yao, Shunyu Liu, Xikun Zhang, Shijian Lu, and Dacheng Tao. R1-vl: Learn- ing to reason with multimodal large language models via step-wise group relative policy optimization.arXiv preprint arXiv:2503.12937, 2025. 3

    Pith/arXiv arXiv 2025

  76. [77]

    Blind image quality assessment using a deep bilinear convolutional neural network.IEEE Transactions on Cir- cuits and Systems for Video Technology (TCSVT), 2018

    Weixia Zhang, Kede Ma, Jia Yan, Dexiang Deng, and Zhou Wang. Blind image quality assessment using a deep bilinear convolutional neural network.IEEE Transactions on Cir- cuits and Systems for Video Technology (TCSVT), 2018. 7

    2018

  77. [78]

    Chain-of-focus: Adaptive visual search and zooming for multimodal reasoning via rl.arXiv preprint arXiv:2505.15436, 2025

    Xintong Zhang, Zhi Gao, Bofei Zhang, Pengxiang Li, Xi- aowen Zhang, Yang Liu, Tao Yuan, Yuwei Wu, Yunde Jia, Song-Chun Zhu, et al. Chain-of-focus: Adaptive visual search and zooming for multimodal reasoning via rl.arXiv preprint arXiv:2505.15436, 2025. 3

    Pith/arXiv arXiv 2025

  78. [79]

    Deep- eyes: Incentivizing” thinking with images” via reinforce- ment learning.arXiv preprint arXiv:2505.14362, 2025

    Ziwei Zheng, Michael Yang, Jack Hong, Chenxiao Zhao, Guohai Xu, Le Yang, Chao Shen, and Xing Yu. Deep- eyes: Incentivizing” thinking with images” via reinforce- ment learning.arXiv preprint arXiv:2505.14362, 2025. 2, 3

    Pith/arXiv arXiv 2025

  79. [80]

    Metaiqa: Deep meta-learning for no- reference image quality assessment

    Hancheng Zhu, Leida Li, Jinjian Wu, Weisheng Dong, and Guangming Shi. Metaiqa: Deep meta-learning for no- reference image quality assessment. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020. 3

    2020

  80. [81]

    Adaptive image quality assess- ment via teaching large multimodal model to compare.Pro- ceedings of Advances in Neural Information Processing Sys- tems (NeurIPS), 2024

    Hanwei Zhu, Haoning Wu, Yixuan Li, Zicheng Zhang, Bao- liang Chen, Lingyu Zhu, Yuming Fang, Guangtao Zhai, Weisi Lin, and Shiqi Wang. Adaptive image quality assess- ment via teaching large multimodal model to compare.Pro- ceedings of Advances in Neural Information Processing Sys- tems (NeurIPS), 2024. 3, 7 12 Zoom-IQA: Image Quality Assessment with Reliab...

    2024

Showing first 80 references.