Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

Segmenting and Understanding: Region-aware Semantic Attention for Fine-grained Image Quality Assessment with Large Language Models

T0 review · 4 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read The paper claims that no-reference image quality assessment improves when an image is segmented into semantic regions, each region is described by a multimodal language model for content and distortions, and a learned attention map weights

desk verdict Plausible region-aware IQA method, but the core MLLM distortion-perception step is unvalidated and the supplied full text is unreadable. read the letter →

arxiv 2508.07818 v1 pith:B3LAEL5W submitted 2025-08-11 cs.CV

Chenyue Song , Chen Hui , Haiqi Zhu , Feng Jiang , Yachun Mi , Wei Zhang , Shaohui Liu This is my paper
classification cs.CV
keywords no-referenceimagequalityassessmentregion-awareattentionfine-grainedIQASegmentAnythingModelmultimodallargelanguagesemanticsegmentationvisualperception
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

The paper is trying to establish that no-reference image quality assessment can be improved by treating quality as a property of semantic regions rather than of the whole image. Its model, RSFIQA, uses a segmentation model to split an image into non-overlapping regions, prompts a multimodal large language model to describe each region's content and its distortions, and then introduces a Region-Aware Semantic Attention mechanism that learns a global attention map over those region representations before regressing a quality score. The authors argue that existing global-representation methods miss local quality variation, while methods that average region features uniformly dilute salient local defects. If the paper is right, quality models can be made fine-grained and interpretable—region descriptions double as explanations—without sacrificing end-to-end deep-learning baselines, since the design is backbone-agnostic. The claimed payoff is competitive or better prediction accuracy across multiple benchmark datasets.

What carries the argument

The central mechanism is Region-Aware Semantic Attention: a module that builds a global attention map from fine-grained local region representations, replacing uniform region pooling with learned, content-dependent weights. The load-bearing inputs are the non-overlapping semantic region masks produced by SAM and the per-region descriptive text generated by a multimodal large language model; the text is encoded into region features, and RSA decides how much each region contributes to the global quality judgment. It also provides a way to inject region-level semantics into arbitrary backbones, which is the mechanism by which the paper claims fine-grained quality awareness without retraining th

What would settle it

Run RSFIQA on a benchmark, then replace the MLLM region descriptions with generic captions that contain no distortion information; if the accuracy does not drop measurably, the claimed distortion-perception step is not carrying the result. A second check is to shuffle region descriptions across regions; stable predictions would show the model is not actually using region-level content.

Watch

Extended reading notes

Core claim

RSFIQA's central claim is that fine-grained, region-level distortion information, generated by a multimodal large language model from SAM-segmented image regions, can be aggregated by a learned Region-Aware Semantic Attention map to produce a global quality representation that predicts human perceptual scores at least as well as current global or uniformly-pooled methods. The model pipeline has three stages: the Segment Anything Model partitions the input into non-overlapping semantic regions; a fine-tuned multimodal language model produces for each region a description covering both semantic content and multi-dimensional distortion cues; and the RSA module converts those region descriptions

Load-bearing premise

The method stands or falls on the multimodal language model correctly perceiving and reporting the distortions present in each cropped region; if those region descriptions are inaccurate, the attention mechanism has nothing reliable to weight.

Editorial extensions

If this is right

  • If RSFIQA is right, NR-IQA systems gain a modular way to inject fine-grained, semantic region information without retraining the backbone, since the RSA module is backbone-agnostic.
  • Quality prediction should become more sensitive to localized distortions in semantically important regions, because region weights are learned rather than uniform.
  • The framework suggests a reusable protocol for teaching multimodal language models to act as region-level distortion annotators, producing descriptive quality cues that downstream regressors can exploit.
  • Competitive results across multiple benchmark datasets imply that region-level semantic descriptions and global representations are complementary rather than redundant for perceptual quality.

Reading between the lines

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

  • A testable extension the paper does not run: replacing the MLLM region descriptions with human-written captions would isolate how much of the gain comes from the perception step versus the attention-pooling mechanism.
  • Because the method relies on non-overlapping SAM regions, an implicit consequence is that coarse or misaligned segmentation should degrade scores; the paper does not directly measure this sensitivity.
  • The RSA mechanism is a general soft-attention over segmentation-derived tokens, so it could transfer to other fine-grained image regression tasks such as aesthetic scoring or artifact localization.
  • If the MLLM's distortion descriptions are noisy, the learned attention may implicitly learn to discount unreliable regions, which would make the method robust to weak region captions; the paper does not isolate that behavior.
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. The paper proposes RSFIQA, a no-reference image quality assessment (NR-IQA) model that partitions an image into non-overlapping semantic regions using SAM, then uses a multimodal large language model (MLLM) to generate region-level descriptions and multi-dimensional distortion perceptions, and finally aggregates these region features with a Region-Aware Semantic Attention (RSA) mechanism to predict a global quality score. The abstract claims the model is backbone-agnostic and achieves competitive performance on multiple benchmarks. However, the provided full text is largely garbled and contains an unrelated arXiv header; no experimental numbers, method equations, or baseline comparisons are legible. Consequently, the central empirical claim cannot currently be verified.

Significance. If the claimed results hold, the approach is a timely and plausible contribution to NR-IQA, combining recent progress in segmentation, MLLMs, and attention mechanisms to obtain region-level quality awareness. The idea of using SAM to obtain semantic regions and MLLMs to produce region-specific distortion descriptions is relevant and potentially more interpretable than global-representation methods. The manuscript does not provide machine-checked proofs, reproducible code, or parameter-free derivations; its support is intended to be empirical, but that evidence is not visible in the current artifact. The direction has merit, but the submission as presented is not reviewable.

major comments (4)
  1. [Full text / integrity] The provided full text is not a readable manuscript: it contains the header 'arXiv:2508.07821v1 [astro-ph.HE] 11 Aug 2025', which is an unrelated astrophysics identifier, and the body is heavily corrupted. I cannot verify the method equations, the experimental section, citations, or any limitation statements. This is a load-bearing issue because the central claim of competitive performance is an empirical claim, and no experiments are legible. A clean, complete manuscript with correct metadata must be supplied before substantive review.
  2. [Abstract / Experiments] The abstract states 'Extensive experiments demonstrate the robustness and effectiveness of the proposed method, which achieves competitive quality prediction performance across multiple benchmark datasets,' but no quantitative results are visible. There are no dataset names, no SROCC/LCC/PLCC values, no baseline comparisons, and no statistical significance tests. Without these, the central claim is an assertion. The experiments section should report results on standard NR-IQA benchmarks and include ablations that isolate the contributions of SAM segmentation, MLLM region descriptions, and RSA.
  3. [Method / MLLM regional perception] The central mechanism requires teaching an MLLM to 'perceive multi-dimensional distortions' from cropped SAM regions. The abstract and the readable portions do not specify the training data, prompt template, or validation of this perception. Because SAM regions are class-agnostic and can be partial-object fragments, it is non-trivial that an MLLM can reliably distinguish subtle distortions such as blur versus compression when given a cropped region. If this step fails, the RSA attention map would reward content salience rather than quality degradation, potentially collapsing the method to a global backbone plus noisy auxiliary features. The paper must validate this component independently, for example by comparing MLLM-generated distortion labels with ground-truth distortion types and by reporting final IQA performance as a function of MLLM perception accuracy.
  4. [RSA mechanism] The formal definition of Region-Aware Semantic Attention is not legible in the provided text. It is not clear how the 'global attention map' is computed from region-level features and whether the aggregation is genuinely different from uniform weighting or global average pooling. Please provide the exact equation and an ablation study comparing RSA against uniform regional aggregation to substantiate the claim that the mechanism improves sensitivity to local quality variations.
minor comments (4)
  1. [Method / SAM] SAM typically produces overlapping masks. The abstract says 'non-overlapping semantic regions'; the manuscript should specify how overlap is resolved (e.g., mask ordering, merging, or post-processing).
  2. [Experiments] The claim that RSFIQA is 'backbone-agnostic' needs to be substantiated by experiments with at least two different backbone architectures, not just one.
  3. [Experiments] When reporting results, please include both monotonic (SROCC) and linear (PLCC/LCC) correlation coefficients, as well as standard deviations or confidence intervals; the current abstract gives no metric definitions.
  4. [Full text] The final manuscript must not contain unrelated arXiv headers or corrupted encoding. The current text is unreadable and would be unusable in production.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the method is an empirically evaluated feature-extraction pipeline.

full rationale

The paper proposes an empirical NR-IQA framework rather than a derivation that reduces to its own inputs. The pipeline takes an input image, segments it with SAM, obtains per-region content and distortion descriptions from an MLLM, and uses a Region-Aware Semantic Attention mechanism to aggregate region features into a global representation for quality prediction. In the abstract and readable portions of the full text, no equation defines the final predicted quality in terms of the ground-truth MOS or in terms of an attention map that was itself fitted to that MOS. The RSA attention weights are computed from the region features, and the final score is trained with a regression loss against human opinion scores on standard NR-IQA benchmarks; the reported comparisons are against external ground truth, so success is not guaranteed by construction. The weakest step—that the MLLM reliably perceives multi-dimensional distortions from cropped SAM regions—is unvalidated in the provided text, but this is an empirical robustness and correctness concern, not circularity: if the MLLM fails, the method's performance should degrade rather than tautologically confirm its assumptions. No uniqueness theorem, load-bearing self-citation, ansatz-smuggling citation, or by-construction identity (Eq. X = Eq. Y) is evident, so no grounded circular step can be reported.

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

Only the abstract was available for review. The paper relies on the effectiveness of pre-trained models (SAM, MLLM) and on the reliability of the benchmark datasets for human quality judgments. No free parameters or invented entities were specified in the abstract.

assumptions (2)
  • domain assumption Pre-trained SAM produces meaningful non-overlapping semantic regions for arbitrary images.
    The method depends on SAM's segmentation quality; if the segmentation is poor, the regional descriptions will be uninformative. This is stated in the abstract as the first stage of the pipeline.
  • domain assumption The MLLM can accurately perceive multi-dimensional distortions from a cropped image region.
    The abstract states that the MLLM is taught to extract descriptive content and perceive distortions; the system's local quality awareness hinges on this capability, which is an empirical claim about the MLLM not proven in the abstract.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Segmenting and Understanding: Region-aware Semantic Attention for Fine-grained Image Quality Assessment with Large Language Models." pith.science (2026). https://pith.science/paper/B3LAEL5W

@misc{pith2026250807818,
  author       = {Pith},
  title        = {Pith review of: Segmenting and Understanding: Region-aware Semantic Attention for Fine-grained Image Quality Assessment with Large Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/B3LAEL5W}},
  note         = {Machine review of arXiv:2508.07818}
}
read the original abstract

No-reference image quality assessment (NR-IQA) aims to simulate the process of perceiving image quality aligned with subjective human perception. However, existing NR-IQA methods either focus on global representations that leads to limited insights into the semantically salient regions or employ a uniform weighting for region features that weakens the sensitivity to local quality variations. In this paper, we propose a fine-grained image quality assessment model, named RSFIQA, which integrates region-level distortion information to perceive multi-dimensional quality discrepancies. To enhance regional quality awareness, we first utilize the Segment Anything Model (SAM) to dynamically partition the input image into non-overlapping semantic regions. For each region, we teach a powerful Multi-modal Large Language Model (MLLM) to extract descriptive content and perceive multi-dimensional distortions, enabling a comprehensive understanding of both local semantics and quality degradations. To effectively leverage this information, we introduce Region-Aware Semantic Attention (RSA) mechanism, which generates a global attention map by aggregating fine-grained representations from local regions. In addition, RSFIQA is backbone-agnostic and can be seamlessly integrated into various deep neural network architectures. Extensive experiments demonstrate the robustness and effectiveness of the proposed method, which achieves competitive quality prediction performance across multiple benchmark datasets.

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

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

  1. FinPercep-RM: A Fine-grained Reward Model and Co-evolutionary Curriculum for RL-based Real-world Super-Resolution

    cs.CV 2025-12 unverdicted novelty 6.0 of 10

    FinPercep-RM with co-evolutionary curriculum learning stabilizes RL training for real-world image super-resolution by supplying local degradation feedback and reducing reward hacking.

Reference graph

Works this paper leans on

55 extracted references · 37 canonical work pages · cited by 1 Pith paper

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    L.; Almeida, D.; Altenschmidt, J.; Altman, S.; Anadkat, S.; et al

    Achiam, J.; Adler, S.; Agarwal, S.; Ahmad, L.; Akkaya, I.; Aleman, F. L.; Almeida, D.; Altenschmidt, J.; Altman, S.; Anadkat, S.; et al. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774

    arXiv 2023

  4. [4]

    Agrawal, H.; Desai, K.; Wang, Y.; Chen, X.; Jain, R.; Johnson, M.; Batra, D.; Parikh, D.; Lee, S.; and Anderson, P. 2019. Nocaps: Novel object captioning at scale. In Proceedings of the IEEE/CVF international conference on computer vision, 8948--8957

    work page 2019

  5. [5]

    Bai, S.; Chen, K.; Liu, X.; Wang, J.; Ge, W.; Song, S.; Dang, K.; Wang, P.; Wang, S.; Tang, J.; et al. 2025. Qwen2. 5-vl technical report. arXiv preprint arXiv:2502.13923

    arXiv 2025

  6. [6]

    Cai, X.; You, Z.; Zhang, H.; Gu, J.; Liu, W.; and Xue, T. 2024. Phocolens: Photorealistic and consistent reconstruction in lensless imaging. Advances in Neural Information Processing Systems, 37: 12219--12242

    work page 2024

  7. [7]

    Chen, C.; Yang, S.; Wu, H.; Liao, L.; Zhang, Z.; Wang, A.; Sun, W.; Yan, Q.; and Lin, W. 2024 a . Q-ground: Image quality grounding with large multi-modality models. In Proceedings of the 32nd ACM International Conference on Multimedia, 486--495

    work page 2024

  8. [8]

    Chen, X.; Fang, H.; Lin, T.-Y.; Vedantam, R.; Gupta, S.; Doll \'a r, P.; and Zitnick, C. L. 2015. Microsoft coco captions: Data collection and evaluation server. arXiv preprint arXiv:1504.00325

    arXiv 2015

Show all 55 references
  1. [9]

    Chen, Z.; He, Z.; Lu, Z.; Sun, X.; and Lu, Z.-M. 2024 b . Prompt-based test-time real image dehazing: a novel pipeline. In European Conference on Computer Vision, 432--449. Springer

    2024

  2. [10]

    E.; et al

    Chiang, W.-L.; Li, Z.; Lin, Z.; Sheng, Y.; Wu, Z.; Zhang, H.; Zheng, L.; Zhuang, S.; Zhuang, Y.; Gonzalez, J. E.; et al. 2023. Vicuna: An open-source chatbot impressing gpt-4 with 90\ See https://vicuna. lmsys. org (accessed 14 April 2023), 2(3): 6

    2023

  3. [11]

    Fang, Y.; Zhu, H.; Zeng, Y.; Ma, K.; and Wang, Z. 2020. Perceptual quality assessment of smartphone photography. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 3677--3686

    2020

  4. [12]

    Ghadiyaram, D.; and Bovik, A. C. 2015. Massive online crowdsourced study of subjective and objective picture quality. IEEE transactions on image processing, 25(1): 372--387

    2015

  5. [13]

    Goyal, Y.; Khot, T.; Summers-Stay, D.; Batra, D.; and Parikh, D. 2017. Making the v in vqa matter: Elevating the role of image understanding in visual question answering. In Proceedings of the IEEE conference on computer vision and pattern recognition, 6904--6913

    2017

  6. [14]

    He, K.; Zhang, X.; Ren, S.; and Sun, J. 2016. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, 770--778

    2016

  7. [15]

    Hosu, V.; Lin, H.; Sziranyi, T.; and Saupe, D. 2020. KonIQ-10k: An ecologically valid database for deep learning of blind image quality assessment. IEEE Transactions on Image Processing, 29: 4041--4056

    2020

  8. [16]

    P.; Perelman, A.; Ramesh, A.; Clark, A.; Ostrow, A.; Welihinda, A.; Hayes, A.; Radford, A.; et al

    Hurst, A.; Lerer, A.; Goucher, A. P.; Perelman, A.; Ramesh, A.; Clark, A.; Ostrow, A.; Welihinda, A.; Hayes, A.; Radford, A.; et al. 2024. Gpt-4o system card. arXiv preprint arXiv:2410.21276

    2024 arXiv

  9. [17]

    Ke, J.; Wang, Q.; Wang, Y.; Milanfar, P.; and Yang, F. 2021. Musiq: Multi-scale image quality transformer. In Proceedings of the IEEE/CVF international conference on computer vision, 5148--5157

    2021

  10. [18]

    C.; and Chandler, D

    Larson, E. C.; and Chandler, D. M. 2010. Most apparent distortion: full-reference image quality assessment and the role of strategy. Journal of electronic imaging, 19(1): 011006--011006

    2010

  11. [19]

    Li, A.; Wu, J.; Liu, Y.; Li, L.; Dong, W.; and Shi, G. 2024. Blind image quality assessment based on perceptual comparison. IEEE Transactions on Multimedia, 26: 9671--9682

    2024

  12. [20]

    Li, C.; Zhang, Z.; Wu, H.; Sun, W.; Min, X.; Liu, X.; Zhai, G.; and Lin, W. 2023. Agiqa-3k: An open database for ai-generated image quality assessment. IEEE Transactions on Circuits and Systems for Video Technology, 34(8): 6833--6846

    2023

  13. [21]

    Lin, H.; Hosu, V.; and Saupe, D. 2019. KADID-10k: A large-scale artificially distorted IQA database. In 2019 Eleventh International Conference on Quality of Multimedia Experience (QoMEX), 1--3. IEEE

    2019

  14. [22]

    Liu, H.; Li, C.; Wu, Q.; and Lee, Y. J. 2023. Visual instruction tuning. Advances in neural information processing systems, 36: 34892--34916

    2023

  15. [23]

    Liu, Y.; Duan, H.; Zhang, Y.; Li, B.; Zhang, S.; Zhao, W.; Yuan, Y.; Wang, J.; He, C.; Liu, Z.; et al. 2024. Mmbench: Is your multi-modal model an all-around player? In European conference on computer vision, 216--233. Springer

    2024

  16. [24]

    Liu, Z.; Lin, Y.; Cao, Y.; Hu, H.; Wei, Y.; Zhang, Z.; Lin, S.; and Guo, B. 2021. Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF international conference on computer vision, 10012--10022

    2021

  17. [25]

    Lu, P.; Mishra, S.; Xia, T.; Qiu, L.; Chang, K.-W.; Zhu, S.-C.; Tafjord, O.; Clark, P.; and Kalyan, A. 2022. Learn to explain: Multimodal reasoning via thought chains for science question answering. Advances in Neural Information Processing Systems, 35: 2507--2521

    2022

  18. [26]

    X.; Tan, J

    Masry, A.; Long, D. X.; Tan, J. Q.; Joty, S.; and Hoque, E. 2022. Chartqa: A benchmark for question answering about charts with visual and logical reasoning. arXiv preprint arXiv:2203.10244

    2022 arXiv

  19. [27]

    Mathew, M.; Karatzas, D.; and Jawahar, C. 2021. Docvqa: A dataset for vqa on document images. In Proceedings of the IEEE/CVF winter conference on applications of computer vision, 2200--2209

    2021

  20. [28]

    K.; and Bovik, A

    Mittal, A.; Moorthy, A. K.; and Bovik, A. C. 2012. No-reference image quality assessment in the spatial domain. IEEE Transactions on image processing, 21(12): 4695--4708

    2012

  21. [29]

    completely blind

    Mittal, A.; Soundararajan, R.; and Bovik, A. C. 2012. Making a “completely blind” image quality analyzer. IEEE Signal processing letters, 20(3): 209--212

    2012

  22. [30]

    K.; and Bovik, A

    Moorthy, A. K.; and Bovik, A. C. 2011. Blind image quality assessment: From natural scene statistics to perceptual quality. IEEE transactions on Image Processing, 20(12): 3350--3364

    2011

  23. [31]

    Ponomarenko, N.; Ieremeiev, O.; Lukin, V.; Egiazarian, K.; Jin, L.; Astola, J.; Vozel, B.; Chehdi, K.; Carli, M.; Battisti, F.; et al. 2013. Color image database TID2013: Peculiarities and preliminary results. In European workshop on visual information processing (EUVIP), 106-...

    2013

  24. [32]

    W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; et al

    Radford, A.; Kim, J. W.; Hallacy, C.; Ramesh, A.; Goh, G.; Agarwal, S.; Sastry, G.; Askell, A.; Mishkin, P.; Clark, J.; et al. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learning, 8748--8763. PmLR

    2021

  25. [33]

    Radford, A.; Wu, J.; Child, R.; Luan, D.; Amodei, D.; Sutskever, I.; et al. 2019. Language models are unsupervised multitask learners. OpenAI blog, 1(8): 9

    2019

  26. [34]

    Saha, A.; Mishra, S.; and Bovik, A. C. 2023. Re-iqa: Unsupervised learning for image quality assessment in the wild. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 5846--5855

    2023

  27. [35]

    R.; Sabir, M

    Sheikh, H. R.; Sabir, M. F.; and Bovik, A. C. 2006. A statistical evaluation of recent full reference image quality assessment algorithms. IEEE Transactions on image processing, 15(11): 3440--3451

    2006

  28. [36]

    Shi, H.; Xie, W.; Qin, H.; Li, Y.; and Fang, L. 2025. Visual state space model with graph-based feature aggregation for blind image quality assessment. IEEE Transactions on Circuits and Systems for Video Technology

    2025

  29. [37]

    Shin, N.-H.; Lee, S.-H.; and Kim, C.-S. 2024. Blind image quality assessment based on geometric order learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 12799--12808

    2024

  30. [38]

    Singh, A.; Natarajan, V.; Shah, M.; Jiang, Y.; Chen, X.; Batra, D.; Parikh, D.; and Rohrbach, M. 2019. Towards vqa models that can read. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 8317--8326

    2019

  31. [39]

    Team, I. 2023. Internlm: A multilingual language model with progressively enhanced capabilities

    2023

  32. [40]

    C.; and Loy, C

    Wang, J.; Chan, K. C.; and Loy, C. C. 2023. Exploring clip for assessing the look and feel of images. In Proceedings of the AAAI conference on artificial intelligence, volume 37, 2555--2563

    2023

  33. [41]

    Wu, H.; Zhang, Z.; Zhang, E.; Chen, C.; Liao, L.; Wang, A.; Xu, K.; Li, C.; Hou, J.; Zhai, G.; et al. 2024 a . Q-instruct: Improving low-level visual abilities for multi-modality foundation models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recogn...

    2024

  34. [42]

    Wu, H.; Zhang, Z.; Zhang, W.; Chen, C.; Liao, L.; Li, C.; Gao, Y.; Wang, A.; Zhang, E.; Sun, W.; et al. 2023. Q-align: Teaching lmms for visual scoring via discrete text-defined levels. arXiv preprint arXiv:2312.17090

    2023 arXiv

  35. [43]

    Wu, H.; Zhu, H.; Zhang, Z.; Zhang, E.; Chen, C.; Liao, L.; Li, C.; Wang, A.; Sun, W.; Yan, Q.; et al. 2024 b . Towards open-ended visual quality comparison. In European Conference on Computer Vision, 360--377. Springer

    2024

  36. [44]

    Wu, T.; Ma, K.; Liang, J.; Yang, Y.; and Zhang, L. 2024 c . A comprehensive study of multimodal large language models for image quality assessment. In European Conference on Computer Vision, 143--160. Springer

    2024

  37. [45]

    Xu, K.; Liao, L.; Xiao, J.; Chen, C.; Wu, H.; Yan, Q.; and Lin, W. 2024. Boosting image quality assessment through efficient transformer adaptation with local feature enhancement. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2662--2672

    2024

  38. [46]

    Yang, S.; Wu, T.; Shi, S.; Lao, S.; Gong, Y.; Cao, M.; Wang, J.; and Yang, Y. 2022. Maniqa: Multi-dimension attention network for no-reference image quality assessment. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 1191--1200

    2022

  39. [47]

    You, Z.; Cai, X.; Gu, J.; Xue, T.; and Dong, C. 2025. Teaching large language models to regress accurate image quality scores using score distribution. In Proceedings of the Computer Vision and Pattern Recognition Conference, 14483--14494

    2025

  40. [48]

    You, Z.; Li, Z.; Gu, J.; Yin, Z.; Xue, T.; and Dong, C. 2024. Depicting beyond scores: Advancing image quality assessment through multi-modal language models. In European Conference on Computer Vision, 259--276. Springer

    2024

  41. [49]

    Young, P.; Lai, A.; Hodosh, M.; and Hockenmaier, J. 2014. From image descriptions to visual denotations: New similarity metrics for semantic inference over event descriptions. Transactions of the association for computational linguistics, 2: 67--78

    2014

  42. [50]

    Yu, J.; Lin, Z.; Yang, J.; Shen, X.; Lu, X.; and Huang, T. S. 2019. Free-form image inpainting with gated convolution. In Proceedings of the IEEE/CVF international conference on computer vision, 4471--4480

    2019

  43. [51]

    Zhang, W.; Ma, K.; Zhai, G.; and Yang, X. 2021. Uncertainty-aware blind image quality assessment in the laboratory and wild. IEEE Transactions on Image Processing, 30: 3474--3486

    2021

  44. [52]

    Zhang, W.; Zhai, G.; Wei, Y.; Yang, X.; and Ma, K. 2023. Blind image quality assessment via vision-language correspondence: A multitask learning perspective. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 14071--14081

    2023

  45. [53]

    Zhang, Z.; Zhou, Y.; Li, C.; Zhao, B.; Liu, X.; and Zhai, G. 2024. Quality assessment in the era of large models: A survey. ACM Transactions on Multimedia Computing, Communications and Applications

    2024

  46. [54]

    Zheng, H.; Yang, H.; Fu, J.; Zha, Z.-J.; and Luo, J. 2021. Learning conditional knowledge distillation for degraded-reference image quality assessment. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 10242--10251

    2021

  47. [55]

    Zhu, D.; Chen, J.; Shen, X.; Li, X.; and Elhoseiny, M. 2023. Minigpt-4: Enhancing vision-language understanding with advanced large language models. arXiv preprint arXiv:2304.10592

    2023 arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.