pith. sign in

arxiv: 2605.22192 · v1 · pith:O4XFY4IPnew · submitted 2026-05-21 · 💻 cs.CV

Ultra-High-Definition Image Quality Assessment via Graph Representation Learning

Shaode Yu , Enqi Chen , Ming Huang , Xuemin Ren , Songnan Zhao , Zhicheng Zhang , Qiurui Sun This is my paper

Pith reviewed 2026-05-22 08:00 UTC · model grok-4.3

classification 💻 cs.CV
keywords ultra-high-definitionblind image quality assessmentgraph neural networkgraph convolutionpatch samplingUHD-IQAregion relation modeling
0
0 comments X

The pith

Graph connections between sampled patches improve blind quality prediction for ultra-high-definition images.

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

The paper establishes that treating sampled patches from a UHD image as nodes in a hybrid graph and using residual graph convolutions to exchange information across regions produces more accurate quality scores than methods that process patches in isolation. This modeling step addresses the core tension in UHD assessment: full-resolution inference is too costly while resizing or isolated crops erase scale-sensitive distortions and break the link between local flaws and overall scene context. A sympathetic reader would care because the approach keeps computational demands manageable yet preserves the structural relations that matter for human perception of high-resolution content. On the UHD-IQA benchmark the method records the lowest root-mean-square error among compared approaches while maintaining competitive correlation with subjective ratings.

Core claim

The paper claims that a graph representation learning framework samples aspect-ratio-aligned patches from each UHD image, encodes them as graph nodes, builds a hybrid k-nearest-neighbor graph using both spatial proximity and feature similarity, applies residual graph convolution to propagate contextual information, and employs gated attention pooling to produce an image-level quality score. An exponential moving average normalized multi-objective loss stabilizes joint regression, correlation, and ranking training. Experiments on the UHD-IQA benchmark yield PLCC of 0.7784, SRCC of 0.8019, and RMSE of 0.0519, indicating that graph-based region relation modeling is effective for UHD image质量评估,尤

What carries the argument

The hybrid k-nearest-neighbor graph on spatial proximity and feature similarity, which turns sampled patches into nodes so that residual graph convolutions can share contextual information across regions before gated attention pooling aggregates evidence.

If this is right

  • Graph-based region modeling improves absolute quality score estimation under high-resolution content.
  • Residual graph convolution propagates contextual information across sampled regions effectively.
  • The exponential moving average normalized multi-objective loss stabilizes joint optimization of regression, correlation, and ranking.
  • The resulting scores achieve the lowest RMSE among the methods tested on the UHD-IQA benchmark.

Where Pith is reading between the lines

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

  • The same patch-graph construction could be tested on video quality assessment by adding temporal edges between corresponding regions across frames.
  • If the graph successfully preserves scale-sensitive information, the framework might reduce reliance on aggressive downsampling in other high-resolution vision pipelines.
  • Varying the balance between spatial-proximity and feature-similarity edges could reveal which relation type matters most for particular distortion families.

Load-bearing premise

Aspect-ratio-aligned patch sampling plus a hybrid graph built on spatial proximity and feature similarity captures the structural dependencies needed for quality assessment without suppressing scale-sensitive distortions or breaking the link between local artifacts and global context.

What would settle it

Retraining the same backbone with all graph convolution layers removed so that patches remain independent and then observing equal or lower RMSE on the UHD-IQA benchmark would show the graph step is not necessary.

Figures

Figures reproduced from arXiv: 2605.22192 by Enqi Chen, Ming Huang, Qiurui Sun, Shaode Yu, Songnan Zhao, Xuemin Ren, Zhicheng Zhang.

Figure 1
Figure 1. Figure 1: The proposed UHD-GCN-BIQA framework. A UHD image is partitioned into an aspect￾ratio–aligned patch sampling from which local features are extracted. Then, a hybrid KNN graph is built to connect adjacent regions, and residual GCN layers are used to propagate contextual information. Finally, gated attention pooling aggregates the patch-level representations to produce an image-level quality prediction. 3.1. … view at source ↗
Figure 2
Figure 2. Figure 2: Hybrid KNN graph construction. Each patch is treated as a node. Edges are formed based on (1) spatial proximity and (2) deep-feature similarity. The two adjacency matrices are linearly combined to obtain the final hybrid graph used for message passing. A purely spatial graph preserves geometric continuity but may fail to relate seman￾tically similar regions that are far apart (e.g., repeated textures or sy… view at source ↗
Figure 3
Figure 3. Figure 3: presents a coarse-to-fine search procedure for the neighborhood size k. The coarse scan first evaluates a compact candidate set between 4 and 28 with an equal interval of 4, and k = 24 is identified as the most promising option. A subsequent fine-grained examination around this region between 21 and 28 with an equal interval of 1 further confirms that k = 24 remains the best-performing neighborhood size fo… view at source ↗
read the original abstract

Blind image quality assessment (BIQA) for ultrahighdefinition (UHD) images remains challenging because native-resolution inference is computationally expensive, whereas aggressive resizing or isolated cropping may suppress scale-sensitive distortions and weaken the relationship between local artifacts and global scene context. This paper aims to improve UHD-BIQA by explicitly modeling the structural dependencies among sampled image regions rather than treating them as independent views, and a graph representation learning framework UHD-GCN-BIQA is proposed. The framework samples aspect-ratio-aligned patches from each UHD image, encodes them as graph nodes, and constructs a hybrid k-nearest-neighbor graph using spatial proximity and feature similarity. Residual graph convolution is used to propagate contextual information across regions, and gated attention pooling aggregates patchlevel evidence into an imagelevel quality prediction. An exponential moving average normalized multiobjective loss function is adopted to stabilize the joint optimization of regression, correlation, and ranking objectives. Experiments on the UHD-IQA benchmark show that UHD-GCN-BIQA achieves PLCC = 0.7784, SRCC = 0.8019, and RMSE = 0.0519, obtaining competitive correlation performance and the lowest RMSE among the compared methods. These results indicate that graph-based region relation modeling is effective for UHD image quality assessment, particularly for improving absolute quality score estimation under high-resolution visual content.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 3 minor

Summary. The manuscript presents UHD-GCN-BIQA, a graph representation learning framework for blind image quality assessment (BIQA) of ultra-high-definition (UHD) images. It samples aspect-ratio-aligned patches from each UHD image as graph nodes, constructs a hybrid k-nearest-neighbor graph based on spatial proximity and feature similarity, applies residual graph convolution to propagate contextual information, uses gated attention pooling to aggregate patch-level evidence into an image-level quality score, and optimizes with an exponential moving average normalized multi-objective loss. On the UHD-IQA benchmark the method reports PLCC = 0.7784, SRCC = 0.8019 and RMSE = 0.0519, achieving competitive correlation coefficients and the lowest RMSE among the listed baselines.

Significance. If the reported numbers hold under full experimental scrutiny, the work shows that explicitly modeling structural dependencies among regions via a hybrid graph can improve absolute quality score estimation for high-resolution content without aggressive resizing. The explicit definitions of graph construction, residual GCN propagation, gated attention pooling and the EMA-normalized loss constitute a clear, reproducible modeling choice whose value is measured against an external benchmark; this provides a concrete, falsifiable contribution to the BIQA literature.

major comments (2)
  1. [§3.2] §3.2 (Graph Construction): the hybrid kNN graph is stated to combine spatial proximity and feature similarity, yet the manuscript does not specify the relative weighting, normalization, or distance metric used to fuse the two criteria; because this choice directly determines whether scale-sensitive distortions are preserved or suppressed, it is load-bearing for the central claim that the graph captures necessary structural dependencies.
  2. [Table 1] Table 1 (UHD-IQA results): the reported RMSE = 0.0519 is presented as the lowest among baselines, but the section provides no information on the number of independent runs, standard deviation, or exact reproduction protocol for the competing methods; without these controls the superiority claim for absolute score estimation cannot be fully assessed.
minor comments (3)
  1. [Abstract / §4.1] The abstract and §4.1 would benefit from stating the concrete values chosen for k and the number of sampled patches per image, as these hyperparameters directly affect both computational cost and the validity of the reported metrics.
  2. [§3.3] Notation for the gated attention pooling operation is introduced without an accompanying equation; adding a compact formula would improve clarity for readers unfamiliar with the mechanism.
  3. [§2] A few citations to prior graph-based IQA or region-relation works appear to be missing from the related-work section; including them would better situate the novelty of the hybrid kNN construction.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment below and indicate the revisions planned for the next version.

read point-by-point responses

  1. Referee: [§3.2] §3.2 (Graph Construction): the hybrid kNN graph is stated to combine spatial proximity and feature similarity, yet the manuscript does not specify the relative weighting, normalization, or distance metric used to fuse the two criteria; because this choice directly determines whether scale-sensitive distortions are preserved or suppressed, it is load-bearing for the central claim that the graph captures necessary structural dependencies.

    Authors: We agree that §3.2 lacks the precise specification of how spatial proximity and feature similarity are fused in the hybrid kNN graph. The manuscript currently states only that both criteria are used without detailing the weighting, normalization, or distance metrics. We will revise this section to provide the exact formulation employed in our experiments, including the distance metric, normalization procedure, and relative weighting. This addition will directly support the claim that the graph construction preserves scale-sensitive structural dependencies. revision: yes

  2. Referee: [Table 1] Table 1 (UHD-IQA results): the reported RMSE = 0.0519 is presented as the lowest among baselines, but the section provides no information on the number of independent runs, standard deviation, or exact reproduction protocol for the competing methods; without these controls the superiority claim for absolute score estimation cannot be fully assessed.

    Authors: We acknowledge that Table 1 and the associated experimental description do not report the number of independent runs, standard deviations, or detailed reproduction protocols for the baselines. We will revise the experimental section to include results from multiple independent runs with reported means and standard deviations for all metrics. We will also add information on the reproduction protocol, including any re-implementations of competing methods. These updates will allow a more complete evaluation of the absolute score estimation performance. revision: yes

Circularity Check

0 steps flagged

No significant circularity; framework is independently defined and benchmarked

full rationale

The paper defines its UHD-GCN-BIQA pipeline through explicit, non-referential choices: aspect-ratio-aligned patch sampling, hybrid kNN graph on spatial proximity plus feature similarity, residual GCN propagation, gated attention pooling, and EMA-normalized multi-objective loss. These components are motivated directly from the problem of preserving scale-sensitive distortions and local-global context in UHD images, without reducing any prediction to a fitted parameter or prior self-citation. Reported PLCC/SRCC/RMSE values are obtained on the external UHD-IQA benchmark and compared against listed baselines, keeping the central claim (graph-based region modeling improves absolute score estimation) empirically grounded rather than tautological.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

Only the abstract is available, so the ledger is necessarily incomplete. The framework introduces several modeling decisions whose justification and sensitivity are not detailed.

free parameters (2)
  • k in k-nearest-neighbor graph
    The number of neighbors used to connect patches is a hyperparameter whose value is not stated and must be chosen or tuned.
  • weights in multi-objective loss
    The exponential moving average normalized combination of regression, correlation, and ranking losses implies tunable balancing coefficients.
axioms (2)
  • domain assumption Aspect-ratio-aligned patch sampling preserves scale-sensitive distortions and global context sufficiently for quality prediction.
    The abstract relies on this premise to justify avoiding native-resolution inference.
  • domain assumption Spatial proximity and feature similarity together define meaningful structural dependencies among image regions.
    This underpins the hybrid kNN graph construction.

pith-pipeline@v0.9.0 · 5785 in / 1584 out tokens · 46427 ms · 2026-05-22T08:00:06.836234+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

51 extracted references · 51 canonical work pages · 1 internal anchor

  1. [1]

    A survey on objective evaluation of image sharpness

    Zhu, Mengqiu; Yu, Lingjie; Wang, Zongbiao; Ke, Zhenxia; Zhi, Chao. A survey on objective evaluation of image sharpness. Applied Sciences2023,13(4), 2652

    work page

  2. [2]

    Progress in blind image quality assessment: A brief review.Mathematics2023,11(12), 2766

    Yang, Pei; Sturtz, Jordan; Qingge, Letu. Progress in blind image quality assessment: A brief review.Mathematics2023,11(12), 2766

    work page

  3. [3]

    Review of quality assessment algorithms on the realistic blurred image database (bid2011).International Conference on Signal and Image Processing2023, pp

    Yu, Shaode and Gu, Jiacheng and Wang, Jiayi and Jin, Mingxue and Zhu, Bing. Review of quality assessment algorithms on the realistic blurred image database (bid2011).International Conference on Signal and Image Processing2023, pp. 450–454

    work page

  4. [4]

    Image quality assessment: from error visibility to structural similarity.IEEE Transactions on Image Processing2004,13(4), 600–612

    Wang, Zhou; Bovik, Alan C.; Sheikh, Hamid R.; Simoncelli, Eero P . Image quality assessment: from error visibility to structural similarity.IEEE Transactions on Image Processing2004,13(4), 600–612

    work page

  5. [5]

    FSIM: A feature similarity index for image quality assessment.IEEE Transactions on Image Processing2011,20(8), 2378–2386

    Zhang, Lin; Zhang, Lei; Mou, Xuanqin; Zhang, David. FSIM: A feature similarity index for image quality assessment.IEEE Transactions on Image Processing2011,20(8), 2378–2386

    work page

  6. [6]

    Edge preservation ratio for image sharpness assessment.World Congress on Intelligent Control and Automation2016,1(1), 1377–1381

    Chen, Luming; Jiang, Fan; Zhang, Hefang; Wu, Shibin; Yu, Shaode; Xie, Yaoqin. Edge preservation ratio for image sharpness assessment.World Congress on Intelligent Control and Automation2016,1(1), 1377–1381

    work page

  7. [7]

    Evaluation of realistic blurring image quality by using a shallow convolutional neural network.IEEE International Conference on Information and Automation2017, pp

    Li, Yaoqing and Wang, Zhaoyang and Dai, Guangzhe and Wu, Shibin and Yu, Shaode and Xie, Yaoqin. Evaluation of realistic blurring image quality by using a shallow convolutional neural network.IEEE International Conference on Information and Automation2017, pp. 853–857

    work page

  8. [8]

    Perceptual video quality assessment: A survey.Science China Information Sciences2024,67(11), 211301

    Min, Xiongkuo; Duan, Huiyu; Sun, Wei; Zhu, Yucheng; Zhai, Guangtao. Perceptual video quality assessment: A survey.Science China Information Sciences2024,67(11), 211301

    work page

  9. [9]

    No-reference image quality assessment in the spatial domain.IEEE Transactions on Image Processing2012,21(12), 4695–4708

    Mittal, A.; Moorthy, A.K.; Bovik, A.C. No-reference image quality assessment in the spatial domain.IEEE Transactions on Image Processing2012,21(12), 4695–4708

    work page

  10. [10]

    completely blind

    Mittal, A.; Soundararajan, R.; Bovik, A.C. Making a “completely blind” image quality analyzer.IEEE Signal Processing Letters2013, 20(3), 209–212

    work page

  11. [11]

    Blind image quality assessment: A natural scene statistics approach in the DCT domain

    Saad, M.A.; Bovik, A.C.; Charrier, C. Blind image quality assessment: A natural scene statistics approach in the DCT domain. IEEE Transactions on Image Processing2012,21(8), 3339–3352

    work page

  12. [12]

    No-reference image quality assessment based on spatial and spectral entropies.Signal Processing: Image Communication2014,29(8), 856–863

    Liu, L.; Liu, B.; Huang, H.; Bovik, A.C. No-reference image quality assessment based on spatial and spectral entropies.Signal Processing: Image Communication2014,29(8), 856–863

    work page

  13. [13]

    A hybrid indicator for realistic blurred image quality assessment.Journal of Visual Communication and Image Representation2023,94, 103848

    Yu, Shaode and Wang, Jiayi and Gu, Jiacheng and Jin, Mingxue and Ma, Yunling and Yang, Lijuan and Li, Jianguang. A hybrid indicator for realistic blurred image quality assessment.Journal of Visual Communication and Image Representation2023,94, 103848

    work page

  14. [14]

    IEEE Transactions on Image Processing 26(5), 2274–2285 (2017)

    Bosse, S.; Maniry, D.; Müller, K.-R.; Wiegand, T.; Samek, W. Deep neural networks for no-reference and full-reference image quality assessment.IEEE Transactions on Image Processing2018,27(1), 206–219. https://doi.org/10.1109/TIP .2017.2760518

    work page doi:10.1109/tip 2017

  15. [16]

    A shallow convolutional neural network for blind image sharpness assessment.PloS one2017,12(5), e0176632

    Yu, Shaode and Wu, Shibin and Wang, Lei and Jiang, Fan and Xie, Yaoqin and Li, Leida. A shallow convolutional neural network for blind image sharpness assessment.PloS one2017,12(5), e0176632

    work page

  16. [17]

    CNN-GRNN for image sharpness assessment

    Yu, Shaode and Jiang, Fan and Li, Leida and Xie, Yaoqin. CNN-GRNN for image sharpness assessment. InAsian Conference on Computer Vision; 2016; pp. 50–61

    work page 2016

  17. [18]

    From Patches to Pictures (PaQ-2-PiQ): Mapping the Perceptual Space of Picture Quality

    Ying, Z.; Niu, H.; Gupta, P .; Mahajan, D.; Ghadiyaram, D.; Bovik, A.C. From Patches to Pictures (PaQ-2-PiQ): Mapping the Perceptual Space of Picture Quality. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR); 2020; pp. 3575–3585

    work page 2020

  18. [19]

    Exploring Kolmogorov-Arnold networks for realistic image sharpness assessment

    Yu, Shaode and Chen, Ze and Yang, Zhimu and Gu, Jiacheng and Feng, Bizu and Sun, Qiurui. Exploring Kolmogorov-Arnold networks for realistic image sharpness assessment. InIEEE International Conference on Acoustics, Speech and Signal Processing; 2025; pp. 1–5

    work page 2025

  19. [20]

    Taylor expansion-based Kolmogorov–Arnold network for blind image quality assessment.Journal of Visual Communication and Image Representation2025,12(5), 104571

    Chen, Ze and Yu, Shaode. Taylor expansion-based Kolmogorov–Arnold network for blind image quality assessment.Journal of Visual Communication and Image Representation2025,12(5), 104571

    work page

  20. [21]

    Zeqing He, Zhibo Wang, Zhixuan Chu, Huiyu Xu, Rui Zheng, Kui Ren, and Chun Chen

    You, Junyong; Korhonen, Jari. Transformer for image quality assessment.IEEE International Conference on Image Processing (ICIP) 2021, 1389–1393. https://doi.org/10.1109/ICIP42928.2021.9506075

    work page doi:10.1109/icip42928.2021.9506075 2021

  21. [22]

    Walk in the cloud: Learning curves for point clouds shape analysis, pp

    Ke, J.; Wang, Q.; Wang, Y.; Milanfar, P .; Yang, F. MUSIQ: Multi-scale image quality transformer. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV); 2021; pp. 5148–5157. https://doi.org/10.1109/ICCV48922.2021.00510

    work page doi:10.1109/iccv48922.2021.00510 2021

  22. [23]

    Dual-branch vision transformer for blind image quality assessment.Journal of Visual Communication and Image Representation2023,94, 103850

    Lee, Se-Ho; Kim, Seung-Wook. Dual-branch vision transformer for blind image quality assessment.Journal of Visual Communication and Image Representation2023,94, 103850. https://doi.org/10.1016/j.jvcir.2023.103850

    work page doi:10.1016/j.jvcir.2023.103850 2023

  23. [24]

    CLIP-Guided Attribute Aware Pretraining for Generalizable Image Quality Assessment.arXiv e-prints2024, arXiv:2406.01020

    Kwon, D.; Kim, D.; Ki, S.; Jo, Y.; Lee, H.-E.; Kim, S.J. CLIP-Guided Attribute Aware Pretraining for Generalizable Image Quality Assessment.arXiv e-prints2024, arXiv:2406.01020. https://doi.org/10.48550/arXiv.2406.01020

    work page doi:10.48550/arxiv.2406.01020

  24. [25]

    Enhancing Descriptive Image Quality Assessment with a Large-Scale Multi-Modal Dataset.arXiv e-prints2025, arXiv:2405.18842

    You, Z.; Gu, J.; Cai, X.; Li, Z.; Zhu, K.; Dong, C.; Xue, T. Enhancing Descriptive Image Quality Assessment with a Large-Scale Multi-Modal Dataset.arXiv e-prints2025, arXiv:2405.18842. https://doi.org/10.48550/arXiv.2405.18842

    work page doi:10.48550/arxiv.2405.18842

  25. [26]

    Q-Insight: Understanding Image Quality via Visual Reinforcement Learning.arXiv e-prints2025, arXiv:2503.22679

    Li, W.; Li, G.; Zhao, T.; Li, C.; Sun, W.; Wang, H.; Zheng, B.; Chen, C.; Min, X.; Zhai, G. Q-Insight: Understanding Image Quality via Visual Reinforcement Learning.arXiv e-prints2025, arXiv:2503.22679. https://doi.org/10.48550/arXiv.2503.22679

    work page doi:10.48550/arxiv.2503.22679

  26. [27]

    Visualquality-r1: Reasoning-inducedimagequalityassessmentviareinforcement learning to rank.arXiv:2505.14460, 2025

    Wu, T.; Zhang, J.; Wang, Y.; Ren, T.; Zeng, K.; Zhang, W.; Jia, J.; Chen, Z.; Zhang, Z.; Min, X.; Zhai, G. VisualQuality-R1: Reasoning-Induced Image Quality Assessment via Reinforcement Learning to Rank.arXiv e-prints2025, arXiv:2505.14460. https://doi.org/10.48550/arXiv.2505.14460

    work page doi:10.48550/arxiv.2505.14460

  27. [28]

    High Resolution Image Quality Database.arXiv2024, arXiv:2401.16087

    Huang, H.; Wan, Q.; Korhonen, J. High Resolution Image Quality Database.arXiv2024, arXiv:2401.16087. https://doi.org/10.485 50/arXiv.2401.16087

    work page arXiv

  28. [29]

    Image Intrinsic Scale Assessment: Bridging the Gap Between Quality and Resolution

    Hosu, V .; Agnolucci, L.; Iso, D.; Saupe, D. Image Intrinsic Scale Assessment: Bridging the Gap Between Quality and Resolution. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV); 2025; pp. 12863–12872

    work page 2025

  29. [30]

    UHD-IQA Benchmark Database: Pushing the Boundaries of Blind Photo Quality Assessment

    Hosu, V .; Agnolucci, L.; Wiedemann, O.; Iso, D.; Saupe, D. UHD-IQA Benchmark Database: Pushing the Boundaries of Blind Photo Quality Assessment. InComputer Vision—ECCV 2024 Workshops, Proceedings, Part IX; Springer Nature Switzerland: Cham, Switzerland, 2024; pp. 467–482. https://doi.org/10.1007/978-3-031-91838-4_28

    work page doi:10.1007/978-3-031-91838-4_28 2024

  30. [31]

    AIM 2024 Challenge on UHD Blind Photo Quality Assessment

    Hosu, V .; Conde, M.V .; Agnolucci, L.; Barman, N.; Zadtootaghaj, S.; Timofte, R.; Sun, W.; Zhang, W.; Cao, Y.; Cao, L.; Jia, J.; Chen, Z.; Zhang, Z.; Min, X.; Zhai, G.; et al. AIM 2024 Challenge on UHD Blind Photo Quality Assessment. InComputer Vision—ECCV 2024 Workshops, Proceedings; Springer Nature: Cham, Switzerland, 2025; pp. 261–286. https://doi.org...

    work page doi:10.1007/978-3-031-91856-8_16 2024

  31. [32]

    Assessing UHD Image Quality from Aesthetics, Distortions, and Saliency

    Sun, W.; Zhang, W.; Cao, Y.; Cao, L.; Jia, J.; Chen, Z.; Zhang, Z.; Min, X.; Zhai, G.-T. Assessing UHD Image Quality from Aesthetics, Distortions, and Saliency. InComputer Vision—ECCV 2024 Workshops, Proceedings; Springer Nature: Cham, Switzerland, 2025; pp. 109–126. https://doi.org/10.1007/978-3-031-91856-8_7

    work page doi:10.1007/978-3-031-91856-8_7 2024

  32. [33]

    Super-Resolved Pseudo Reference in Dual-Branch Embedding for Blind Ultra-High-Definition Image Quality Assessment.Electronics2025,14(17), 3447

    Gu, J.; Meng, Q.; Zhao, S.; Wang, Y.; Yu, S.; Sun, Q. Super-Resolved Pseudo Reference in Dual-Branch Embedding for Blind Ultra-High-Definition Image Quality Assessment.Electronics2025,14(17), 3447. https://doi.org/10.3390/electronics14173447

    work page doi:10.3390/electronics14173447

  33. [34]

    GraphIQA: Learning Distortion Graph Representations for Blind Image Quality Assessment.IEEE Transactions on Multimedia2022,25, 2912–2925

    Sun, S.; Yu, T.; Xu, J.; Zhou, W.; Chen, Z. GraphIQA: Learning Distortion Graph Representations for Blind Image Quality Assessment.IEEE Transactions on Multimedia2022,25, 2912–2925. https://doi.org/10.1109/TMM.2022.3152942

    work page doi:10.1109/tmm.2022.3152942 2022

  34. [35]

    No-reference Image Quality Assessment via Non-local Dependency Modeling

    Jia, S.; Chen, B.; Li, D.; Wang, S. No-reference Image Quality Assessment via Non-local Dependency Modeling. InProceedings of the IEEE International Workshop on Multimedia Signal Processing (MMSP); 2022. https://github.com/SuperBruceJia/NLNet-IQA

    work page 2022

  35. [36]

    Blind Image Quality Assessment via Adaptive Graph Attention.IEEE Transactions on Circuits and Systems for Video Technology2024,34(10), 10299–10309

    Wang, H.; Liu, J.; Tan, H.; Lou, J.; Liu, X.; Zhou, W.; Liu, H. Blind Image Quality Assessment via Adaptive Graph Attention.IEEE Transactions on Circuits and Systems for Video Technology2024,34(10), 10299–10309. https://doi.org/10.1109/TCSVT.2024.3405789

    work page doi:10.1109/tcsvt.2024.3405789 2024

  36. [37]

    Image Quality Assessment by Enabling Inter-Patch Message Passing via Graph Convolutional Networks.Neural Computing and Applications2025,37, 6145–6161

    Liu, Y.; Guo, J. Image Quality Assessment by Enabling Inter-Patch Message Passing via Graph Convolutional Networks.Neural Computing and Applications2025,37, 6145–6161. https://doi.org/10.1007/s00521-024-10893-w

    work page doi:10.1007/s00521-024-10893-w

  37. [38]

    A Survey on Graph Representation Learning Methods.ACM Transactions on Intelligent Systems and Technology2024,15(1), 19:1–19:55

    Khoshraftar, S.; An, A. A Survey on Graph Representation Learning Methods.ACM Transactions on Intelligent Systems and Technology2024,15(1), 19:1–19:55. https://doi.org/10.1145/3633518

    work page doi:10.1145/3633518

  38. [39]

    Semi-Supervised Classification with Graph Convolutional Networks

    Kipf, T.N.; Welling, M. Semi-Supervised Classification with Graph Convolutional Networks. InInternational Conference on Learning Representations (ICLR); 2017. https://doi.org/10.48550/arXiv.1609.02907. https://doi.org/10.3390/app1010000 Version May 22, 2026 submitted toAppl. Sci. 20 of 20

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1609.02907 2017

  39. [40]

    Zhang, H

    Zhang, S.; Tong, H.; Xu, J.; Maciejewski, R. Graph Convolutional Networks: A Comprehensive Review.Computational Social Networks2019,6(1), 11. https://doi.org/10.1186/s40649-019-0069-y

    work page doi:10.1186/s40649-019-0069-y

  40. [41]

    Graph Attention Networks

    Veliˇ ckovi´ c, P .; Cucurull, G.; Casanova, A.; Romero, A.; Liò, P .; Bengio, Y. Graph Attention Networks. InInternational Conference on Learning Representations (ICLR); 2018. https://openreview.net/forum?id=rJXMpikCZ

    work page 2018

  41. [42]

    How Powerful are Graph Neural Networks? InInternational Conference on Learning Representations (ICLR); 2019

    Xu, K.; Hu, W.; Leskovec, J.; Jegelka, S. How Powerful are Graph Neural Networks? InInternational Conference on Learning Representations (ICLR); 2019. https://openreview.net/forum?id=ryGs6iA5Km

    work page 2019

  42. [43]

    Inductive Representation Learning on Large Graphs

    Hamilton, W.L.; Ying, Z.; Leskovec, J. Inductive Representation Learning on Large Graphs. InAdvances in Neural Information Processing Systems 30 (NeurIPS); 2017. https://papers.nips.cc/paper/6703-inductive-representation-learning-on-large-graphs

    work page 2017

  43. [44]

    Xu, J.; Zhou, W.; Chen, Z. Blind Omnidirectional Image Quality Assessment with Viewport Oriented Graph Convolutional Networks.IEEE Transactions on Circuits and Systems for Video Technology2021,31(5), 1724–1737. https://doi.org/10.1109/TCSVT. 2020.3015186

    work page doi:10.1109/tcsvt 2020

  44. [45]

    2023/120

    Xie, W.; Bian, T.; Wang, M. kgMBQA: Quality Knowledge Graph-driven Multimodal Blind Image Assessment. InProceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence (IJCAI); 2025; pp. 4245–4253. https://doi.org/10.24963/ijcai .2025/473

    work page doi:10.24963/ijcai 2025

  45. [46]

    Deep residual learning for image recognition,

    He, K.; Zhang, X.; Ren, S.; Sun, J. Deep Residual Learning for Image Recognition. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR); 2016; pp. 770–778. https://doi.org/10.1109/CVPR.2016.90

    work page doi:10.1109/cvpr.2016.90 2016

  46. [47]

    Belkin and P

    Belkin, M.; Niyogi, P . Laplacian Eigenmaps for Dimensionality Reduction and Data Representation.Neural Computation2003, 15(6), 1373–1396. https://doi.org/10.1162/089976603321780317

    work page doi:10.1162/089976603321780317

  47. [48]

    Diffusion Maps.Applied and Computational Harmonic Analysis2006,21(1), 5–30

    Coifman, R.R.; Lafon, S. Diffusion Maps.Applied and Computational Harmonic Analysis2006,21(1), 5–30. https://doi.org/10.1016/ j.acha.2006.04.006

    work page 2006

  48. [49]

    Attention-based Deep Multiple Instance Learning

    Ilse, M.; Tomczak, J.M.; Welling, M. Attention-based Deep Multiple Instance Learning. InInternational Conference on Machine Learning (ICML); 2018; pp. 2127–2136

    work page 2018

  49. [50]

    ImageNet: A Large-Scale Hierarchical Image Database

    Deng, J.; Dong, W.; Socher, R.; Li, L.-J.; Li, K.; Fei-Fei, L. ImageNet: A Large-Scale Hierarchical Image Database. In2009 IEEE Conference on Computer Vision and Pattern Recognition (CVPR); IEEE: 2009; pp. 248–255. https://doi.org/10.1109/CVPR.2009.5206 848

    work page doi:10.1109/cvpr.2009.5206 2009

  50. [51]

    Image Quality Assessment Using Contrastive Learning

    Madhusudana, P .C., Birkbeck, N., Wang, Y., Adsumilli, B., Bovik, A.C. Image Quality Assessment Using Contrastive Learning. IEEE Transactions on Image Processing2022,31(1), 4149–4161

    work page

  51. [52]

    Arniqa: Learning distortion manifold for image quality assessment.Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision; 2024; pp

    Agnolucci, Lorenzo and Galteri, Leonardo and Bertini, Marco and Del Bimbo, Alberto. Arniqa: Learning distortion manifold for image quality assessment.Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision; 2024; pp. 189–198. https://doi.org/10.3390/app1010000

    work page doi:10.3390/app1010000 2024