Pith. sign in

REVIEW 3 major objections 4 minor 63 references

Video quality scores become far more accurate once models factor in the actual phone screen and lighting of the viewer.

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 · grok-4.5

2026-07-11 15:52 UTC pith:3LBZ356N

load-bearing objection Solid multi-device VQA dataset (300+ Android phones + metadata) plus a practical adapter that lifts Kendall on held-out models; the condition-pool claim is plausible but under-ablated. the 3 major comments →

arxiv 2607.04643 v1 pith:3LBZ356N submitted 2026-07-06 cs.CV

Learning Flexible Generalization in Video Quality Assessment by Bringing Device and Viewing Condition Distributions

Nikolay Safonov , Dmitriy S. Vatolin This is my paper
classification cs.CV
keywords video quality assessmentmulti-screen VQAviewing conditionsmobile devicesBlade-Chest modelcondition adaptationpairwise preferencescompression artifacts
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.

Standard video quality metrics treat every screen and room as the same, yet people see compression artifacts very differently depending on screen size, brightness, ambient light and display type. This paper shows that the gap can be closed by first collecting hundreds of thousands of pairwise preference votes across more than three hundred real Android phones, then recovering latent quality scores with a condition-aware aggregation model, and finally training a lightweight adapter that maps any existing metric plus viewing-condition features into a device-specific score. The resulting adapted metrics preserve human preference orderings much better than the original metrics, even on phone models never seen during training. Streaming services can therefore optimise bitrate ladders and encoding decisions for the actual devices their users hold rather than for a single laboratory monitor. The work supplies both the large multi-device dataset and the adaptation recipe needed to make that possible.

Core claim

Incorporating measured device and viewing-condition metadata into both the aggregation of human pairwise votes and the subsequent adaptation of objective metrics produces substantially higher Kendall rank correlation with real-user preferences than the same metrics used without any condition information. The gain holds across classical and modern full-reference and no-reference metrics and generalises to held-out phone models.

What carries the argument

Blade-Chest aggregation: latent quality scores qi are recovered jointly with two condition-conditioned networks (fc, fb) via EM so that the probability that video i beats j under observed conditions z is given by a sigmoid of the difference of blade-chest distances; a lightweight fully-connected adapter then maps any base-metric score plus a (possibly newly sampled) z vector into a condition-specific quality prediction.

Load-bearing premise

The latent quality numbers recovered by the Blade-Chest EM procedure are treated as true condition-independent scores, and synthetic viewing-condition vectors drawn from a hand-constrained uniform pool are assumed to produce valid soft training targets for the adapter.

What would settle it

Hold out an entire phone model (or an extreme brightness/ambient-light regime) never seen in training, retrain the adapters without any samples from that model or regime, and check whether the adapted metrics still improve Kendall correlation on the held-out votes; if the gain disappears or turns negative, the condition-pool generalisation claim fails.

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

If this is right

  • Streaming platforms can re-rank or re-encode the same video differently for low-end LCD phones versus high-brightness HDR flagships without collecting new subjective data for every device.
  • Existing laboratory-trained metrics such as VMAF can be reused rather than retrained from scratch simply by attaching the lightweight condition adapter.
  • Banding and other display-sensitive artefacts become measurable under the actual brightness and ambient light of end users instead of under fixed lab conditions.
  • Device manufacturers receive a quantitative signal of how their panel technology and peak-brightness choices alter perceived compression quality.
  • Future subjective tests can deliberately undersample rare device-condition combinations because the condition pool can later synthesise them.

Where Pith is reading between the lines

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

  • The same adapter architecture could be attached to live QoE estimators inside ABR controllers so that the bitrate ladder itself becomes device-aware in real time.
  • Because the condition vector is low-dimensional and hand-crafted, the method may transfer to non-Android platforms once analogous sensor and panel metadata are available.
  • The observed drop in perceived quality with larger screen diagonals for fixed VMAF scores supplies a simple, immediately usable rule-of-thumb for mobile-first encoding presets.

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 / 4 minor

Summary. The paper introduces a large-scale crowd-sourced multi-device VQA dataset of ~250k pairwise preference judgments collected on 300+ Android devices, with rich metadata on display properties (size, resolution, type, brightness) and ambient conditions. It aggregates the votes with a Blade-Chest model (Eq. 1–3, Appendix A) that conditions on a five-dimensional viewing vector z to recover latent quality scores qi, then trains per-metric lightweight MLPs that map an existing VQA prediction plus z to a condition-specific quality estimate. Soft targets for the adapters are generated by sampling synthetic z from a hand-constrained uniform condition pool and passing them through the learned fc/fb networks. On held-out source videos and three frequent phone models, the adapted metrics show substantial Kendall-rank gains over their unadapted counterparts (Fig. 4, Table 3).

Significance. If the claimed condition-aware generalization holds, the work supplies both a uniquely realistic multi-screen mobile dataset and a practical, lightweight adaptation recipe that streaming services could use for device-specific rate-distortion optimization. The public release of data and code is a clear strength. The Blade-Chest aggregation plus synthetic condition pool is a novel and potentially reusable mechanism for sparse multi-condition subjective data. These contributions would meaningfully advance VQA beyond laboratory MOS toward real-world heterogeneous viewing.

major comments (3)
  1. [Section 6, Fig. 4] Section 6 and Fig. 4: the sole reported performance measure is Kendall rank correlation against the same raw, noisy pairwise votes used to fit Blade-Chest. No correlation of the adapted scores with the recovered qi, with Bradley-Terry aggregates, or with the desktop-monitor reference scores is provided. Because the evaluation target is itself the noisy preference data, it remains unclear whether the large gains reflect genuine condition-aware quality prediction or simply a better ranking of the fixed video set under vote noise.
  2. [Section 5, Table 3] Section 5 and the condition-pool construction: there is no ablation that freezes, zeros, or randomly permutes the z channel while keeping the soft targets fixed. Without this control it is impossible to isolate how much of the Kendall improvement is attributable to the claimed viewing-condition pathway versus the MLP simply learning a more flexible ranking function from the Blade-Chest soft labels. The OOD phone-model hold-outs (Table 3) still share the identical video content and similar z ranges, so they do not close this gap.
  3. [Section 5, Limitations] Section 5 and Limitations: the adapters are never compared against the natural baseline of fine-tuning the original VQA models (or their final layers) directly on the proposed dataset without an explicit condition input. Given that the authors themselves note that full retraining was left for future work, the present gains cannot yet be attributed specifically to the multi-screen adaptation strategy rather than to domain adaptation on the new content.
minor comments (4)
  1. [Abstract, §1] Abstract and Introduction contain repeated grammatical slips (“more than different 300 Android devices”, “First off all we collect”). A careful proof-reading pass is needed.
  2. [Figure 2] Figure 2 caption and surrounding text mix “LCD/LED” with “HDR/SDR” without clarifying whether the display-type label is self-reported or inferred; a short note on how display technology was obtained would help reproducibility.
  3. [§4–5] The five-dimensional z vector is described only informally; an explicit listing of the exact features, their normalization, and the hand-picked physical constraints used for the condition pool should appear in the main text or Appendix B.
  4. [Table 1] Table 1 lists “Proposed … 300+ Crowd. 10,000 240K” while the text claims 250k valid annotations; the discrepancy should be reconciled.

Circularity Check

1 steps flagged

Minor fitted-input step in the condition-pool soft targets; central Kendall gains remain empirical and not forced by construction.

specific steps
  1. fitted input called prediction [Section 5 (Conditions adaptation) and Figure 3]
    "Now, when we have obtained subjective scores qi and trained the networks fc(·) and fb(·) on the real predictions, we can use their combination to expand the training set with simulated samples. ... Conditions z are sampled from a uniform distribution and passed to fc(·) and fb(·) together with a randomly selected pair of distorted versions of the same source video. This yields the probability that video i is of higher quality than video j under the given conditions z. ... The video quality metric predictions for the selected videos and the conditions are processed through the adaptation module"

    The soft targets that supervise the adaptation MLP are generated from the already-fitted Blade-Chest latent scores and networks; the adapter therefore learns to reproduce a ranking already extracted from the same pairwise votes. While subsequent Kendall evaluation is performed on raw human votes (not the soft targets) and on held-out content/devices, the training signal itself is a fitted quantity rather than an independent first-principles label, creating a mild fitted-input-to-prediction loop.

full rationale

The paper's derivation is largely self-contained and empirical: a new multi-device pairwise dataset is collected, Blade-Chest (external citation) is fitted via EM to recover latent qi and condition-aware fc/fb, a lightweight MLP adapter is then trained on soft pairwise probabilities generated from those fitted networks under synthetically sampled z, and the resulting adapted metrics are evaluated by Kendall rank correlation against held-out raw human votes (and OOD phone models). No equation equates a reported gain to a fitted constant, no uniqueness theorem is imported from the authors, and the evaluation target is the original noisy votes rather than the soft targets themselves. The only mild circularity is that the adapter's training signal is produced by a model already fitted to the same vote distribution; this is ordinary two-stage fitting rather than a definitional tautology, so the score remains low.

Axiom & Free-Parameter Ledger

3 free parameters · 3 axioms · 2 invented entities

The central empirical claim rests on the Blade-Chest generative model of pairwise preferences, the sufficiency of a five-dimensional hand-crafted condition vector, the validity of EM recovery of latent scores, and the usefulness of uniform sampling over a constrained condition pool. Network widths, activation choices and physical bounds on the pool are free design decisions; no new physical entities are postulated.

free parameters (3)
  • adaptation MLP depth and width = depth=4, width=64
    Fixed at depth 4, hidden size 64 with tanh/sigmoid activations; chosen by the authors without reported search.
  • hand-picked physical constraints on condition pool
    Uniform sampling of z is restricted by unspecified but ‘physically motivated’ bounds on size, brightness, luminance etc.; these bounds directly affect the training distribution of the adapter.
  • Blade-Chest embedding dimension and network architecture for fc/fb
    Fully-connected nets with tanh activations; exact widths and latent dimension of blade/chest vectors are design choices that influence the recovered qi.
axioms (3)
  • domain assumption Pairwise preference probability is a sigmoid of the difference of Euclidean distances between condition-conditioned blade and chest embeddings (Eq. 1).
    Taken from Chen & Joachims 2016 and assumed to hold for video quality under varying displays.
  • ad hoc to paper A five-dimensional vector (physical size, resolution, brightness, ambient luminance, display type) is a sufficient statistic for the relevant viewing-condition effects.
    Stated in Section 4; no ablation shows that additional factors (refresh rate, bit-depth, viewing distance) are unnecessary.
  • domain assumption Latent quality scores qi recovered by EM are comparable across videos and can serve as ground-truth targets once conditions are re-sampled.
    Core of the two-stage training pipeline in Section 5 and Appendix A.
invented entities (2)
  • viewing-conditions pool no independent evidence
    purpose: Augment sparse real annotations by sampling synthetic condition vectors so the adapter can be trained on combinations never observed in the crowd-sourced data.
    Introduced in Section 5; independent evidence is limited to the held-out phone-model generalization experiments.
  • condition-adaptation module (per-metric MLP) no independent evidence
    purpose: Map a base metric score plus a condition vector z to a condition-specific quality prediction without retraining the base metric.
    Lightweight fully-connected network trained separately for each metric; its utility is demonstrated only inside this paper’s evaluation protocol.

pith-pipeline@v1.1.0-grok45 · 25828 in / 2706 out tokens · 29115 ms · 2026-07-11T15:52:36.015680+00:00 · methodology

0 comments
read the original abstract

Video quality assessment (VQA) plays a critical role in optimizing video delivery systems. While numerous objective metrics have been proposed to approximate human perception, the perceived quality strongly depends on viewing conditions and display characteristics. Factors such as ambient lighting, display brightness, and resolution significantly influence the visibility of distortions. In this work, we address the question of the multi-screen quality assessment on mobile devices, as this area still tends to be under-covered. We introduce a first large-scale subjective dataset collected across more than different 300 Android devices, accompanied by metadata on viewing conditions and display properties. We propose a strategy for aggregated score extraction and adaptation of VQA models to device-specific quality estimation. Our results demonstrate that incorporating device and context information enables more accurate and flexible quality prediction, offering new opportunities for fine-grained optimization in streaming services. Ultimately, this work advances the development of perceptual quality models that bridge the gap between laboratory evaluations and the diverse conditions of real-world media consumption. We made the dataset and the code available at https://videoprocessing.github.io/device-viewing-conditions.

Figures

Figures reproduced from arXiv: 2607.04643 by Dmitriy S. Vatolin, Nikolay Safonov.

Figure 1
Figure 1. Figure 1: Depiction of the multi-screen video quality assessment task problem. The perceived quality strongly depends on viewing conditions and display characteristics. Videos with visibly different quality on the high-end devices may be perceived with the same quality on the low-end ones, and existing VQA metrics do not take that into account. streaming, and content delivery. The goal of VQA models is to estimate h… view at source ↗
Figure 2
Figure 2. Figure 2: Top row: Bradley–Terry (Bradley & Terry, 1952) ag￾gregated scores for subsets of devices grouped by screen size (the smallest quarter, the two middle quarters, and the largest quarter) and comparing them with aggregated scores obtained on desk￾top devices with large screens by Pearson (PLCC), Spearman (SROCC) and Kendall (KROCC) correlations; Middle row: correlations separately for LED and LCD displays; Bo… view at source ↗
Figure 3
Figure 3. Figure 3: The training framework scheme: from the video set a random pair is sampled, and viewing conditions are selected from a distribution. The video quality metric predictions for the selected videos and the viewing conditions are processed through the adap￾tation module, while the estimated subjective scores together with the viewing conditions are processed through the match function using fc(·) and fb(·). Alt… view at source ↗
Figure 4
Figure 4. Figure 4: Kendall rank correlations for the original metrics and their adapted counterparts. Gain represents how score improved after metric training (green = positive, red = negative). models. The condition pool addresses this by enabling sam￾pling of viewing conditions independently from the original annotations, effectively augmenting the training process and improving generalization, including to conditions not … view at source ↗
Figure 5
Figure 5. Figure 5: Relationship between the estimated quality of the VMAF adaptation module and the viewing conditions, predictions for several fixed VMAF levels with varying the screen diagonal (all other parameters are held constant). Five source videos, along with all their distorted versions, were held out as the testing set. We first applied both clas￾sical and modern neural network–based image and video quality metrics… view at source ↗
Figure 6
Figure 6. Figure 6: Distribution of the phone models frequency. C. Evaluation 16 [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Distribution of the phone models frequency. Top-10 models cover almost 20% of the votes [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Distribution of device heights. 18 [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Distribution of screen brightness levels. 19 [PITH_FULL_IMAGE:figures/full_fig_p019_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Distribution of the ambient lightning. AVQT LPIPS SSIMULACRA FSIM PAQ2PIQ DISTS UNIQUE VMAF DBCNN CLIP-IQA-PLUS DOVER COMPRESSED-VQA VMAF NEG LI2022 VIF MDTVSFA HAARPSI TOPIQ-NR SSIM RANK-IQA TOPIQ NLPD PSNR MS-SSIM KONCEPT LINEARITY VIDEVAL VSFA MUSIQ NIQE BRISQUE CLIP-IQA 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 KROCC KROCC gain for adapted metrics (Phone model 1) KROCC (original) Gain (positive) Gain (negative)… view at source ↗
Figure 11
Figure 11. Figure 11: Kendall rank correlations for the original metrics and their adapted counterparts for Xiaomi Redmi Note 13. Gain represents how score improved after metric training (green = positive, red = negative) 20 [PITH_FULL_IMAGE:figures/full_fig_p020_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Kendall rank correlations for the original metrics and their adapted counterparts for Samsung Galaxy A55. Gain represents how score improved after metric training (green = positive, red = negative) SSIMULACRA LPIPS FSIM UNIQUE DOVER AVQT VMAF CLIP-IQA-PLUS LI2022 DISTS VIF VMAF NEG PAQ2PIQ DBCNN HAARPSI RANK-IQA COMPRESSED-VQA MDTVSFA SSIM TOPIQ-NR TOPIQ VSFA KONCEPT LINEARITY NLPD PSNR MS-SSIM MUSIQ VIDE… view at source ↗
Figure 13
Figure 13. Figure 13: Kendall rank correlations for the original metrics and their adapted counterparts for Xiaomi Redmi Note 8 Pro. Gain represents how score improved after metric training (green = positive, red = negative) 21 [PITH_FULL_IMAGE:figures/full_fig_p021_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.

Reference graph

Works this paper leans on

63 extracted references · 4 linked inside Pith

  1. [1]

    264/AVC video quality assessment dataset , author=

    MCL-JCV: a JND-based H. 264/AVC video quality assessment dataset , author=. 2016 IEEE international conference on image processing (ICIP) , pages=. 2016 , organization=

    2016

  2. [2]

    Journal of Visual Communication and Image Representation , volume=

    VideoSet: A large-scale compressed video quality dataset based on JND measurement , author=. Journal of Visual Communication and Image Representation , volume=. 2017 , publisher=

    2017

  3. [3]

    2016 IEEE international symposium on broadband multimedia systems and broadcasting (BMSB) , pages=

    SJTU 4K video subjective quality dataset for content adaptive bit rate estimation without encoding , author=. 2016 IEEE international symposium on broadband multimedia systems and broadcasting (BMSB) , pages=. 2016 , organization=

    2016

  4. [4]

    2018 16th Annual Workshop on Network and Systems Support for Games (NetGames) , pages=

    GamingVideoSET: a dataset for gaming video streaming applications , author=. 2018 16th Annual Workshop on Network and Systems Support for Games (NetGames) , pages=. 2018 , organization=

    2018

  5. [5]

    Advances in Neural Information Processing Systems , volume=

    Video compression dataset and benchmark of learning-based video-quality metrics , author=. Advances in Neural Information Processing Systems , volume=

  6. [6]

    2024 16th International Conference on Quality of Multimedia Experience (QoMEX) , pages=

    AVT-VQDB-UHD-1-HDR: an open video quality dataset for quality assessment of UHD-1 HDR videos , author=. 2024 16th International Conference on Quality of Multimedia Experience (QoMEX) , pages=. 2024 , organization=

    2024

  7. [7]

    2023 15th International Conference on Quality of Multimedia Experience (QoMEX) , pages=

    A subjective dataset for multi-screen video streaming applications , author=. 2023 15th International Conference on Quality of Multimedia Experience (QoMEX) , pages=. 2023 , organization=

    2023

  8. [8]

    2024 IEEE International Conference on Visual Communications and Image Processing (VCIP) , pages=

    Multi-Screen Effects on Quality Assessment: Investigating Banding Metrics Inconsistencies , author=. 2024 IEEE International Conference on Visual Communications and Image Processing (VCIP) , pages=. 2024 , organization=

    2024

  9. [9]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    The unreasonable effectiveness of deep features as a perceptual metric , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  10. [10]

    IEEE transactions on pattern analysis and machine intelligence , volume=

    Image quality assessment: Unifying structure and texture similarity , author=. IEEE transactions on pattern analysis and machine intelligence , volume=. 2020 , publisher=

    2020

  11. [11]

    2016 , url=

    Toward A Practical Perceptual Video Quality Metric , author=. 2016 , url=

    2016

  12. [12]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Exploring video quality assessment on user generated contents from aesthetic and technical perspectives , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  13. [13]

    IEEE Transactions on Image Processing , volume=

    HDR or SDR? A subjective and objective study of scaled and compressed videos , author=. IEEE Transactions on Image Processing , volume=. 2024 , publisher=

    2024

  14. [14]

    IEEE Access , volume=

    No-reference video quality estimation based on machine learning for passive gaming video streaming applications , author=. IEEE Access , volume=. 2019 , publisher=

    2019

  15. [15]

    2020 IEEE Conference on Multimedia Information Processing and Retrieval (MIPR) , pages=

    UGC-VIDEO: Perceptual quality assessment of user-generated videos , author=. 2020 IEEE Conference on Multimedia Information Processing and Retrieval (MIPR) , pages=. 2020 , organization=

    2020

  16. [16]

    2019 IEEE International Symposium on Multimedia (ISM) , pages=

    AVT-VQDB-UHD-1: A large scale video quality database for UHD-1 , author=. 2019 IEEE International Symposium on Multimedia (ISM) , pages=. 2019 , organization=

    2019

  17. [17]

    ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    Subjective and objective quality assessment of mobile gaming video , author=. ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2022 , organization=

    2022

  18. [18]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    MD-VQA: Multi-dimensional quality assessment for UGC live videos , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  19. [19]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Kvq: Kwai video quality assessment for short-form videos , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  20. [20]

    arXiv preprint arXiv:2507.03990 , year=

    LEHA-CVQAD: Dataset To Enable Generalized Video Quality Assessment of Compression Artifacts , author=. arXiv preprint arXiv:2507.03990 , year=

    Pith/arXiv arXiv

  21. [21]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Rich features for perceptual quality assessment of UGC videos , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  22. [22]

    Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages=

    Subjective and objective video quality assessment of high dynamic range sports content , author=. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages=

  23. [23]

    2025 , organization =

    BrightVQ: Quality Assessment for User-Generated HDR Videos , author =. 2025 , organization =

    2025

  24. [24]

    2022 IEEE international conference on image processing (ICIP) , pages=

    Subjective assessment of high dynamic range videos under different ambient conditions , author=. 2022 IEEE international conference on image processing (ICIP) , pages=. 2022 , organization=

    2022

  25. [25]

    arXiv preprint arXiv:2505.21831 , year=

    HDRSDR-VQA: A Subjective Video Quality Dataset for HDR and SDR Comparative Evaluation , author=. arXiv preprint arXiv:2505.21831 , year=

    Pith/arXiv arXiv

  26. [26]

    IEEE Journal on Emerging and Selected Topics in Circuits and Systems , volume=

    Subjective and objective quality assessment of compressed screen content images , author=. IEEE Journal on Emerging and Selected Topics in Circuits and Systems , volume=. 2016 , publisher=

    2016

  27. [27]

    IEEE Transactions on Image Processing , volume=

    CVD2014—A database for evaluating no-reference video quality assessment algorithms , author=. IEEE Transactions on Image Processing , volume=. 2016 , publisher=

    2016

  28. [28]

    Journal of Visual Communication and Image Representation , volume=

    MCL-V: A streaming video quality assessment database , author=. Journal of Visual Communication and Image Representation , volume=. 2015 , publisher=

    2015

  29. [29]

    IEEE Transactions on Image Processing , volume=

    Study of temporal effects on subjective video quality of experience , author=. IEEE Transactions on Image Processing , volume=. 2017 , publisher=

    2017

  30. [30]

    IEEE Network , volume=

    A generic quantitative relationship between quality of experience and quality of service , author=. IEEE Network , volume=. 2010 , publisher=

    2010

  31. [31]

    2012 Fourth international workshop on quality of multimedia experience , pages=

    The TUM high definition video datasets , author=. 2012 Fourth international workshop on quality of multimedia experience , pages=. 2012 , organization=

    2012

  32. [32]

    IEEE Access , year=

    Assessing quality differences for 8K/UHD-2 and 4K/UHD-1 HDR video based on viewing distance , author=. IEEE Access , year=

  33. [33]

    arXiv preprint arXiv:2401.11485 , year=

    ColorVideoVDP: A visual difference predictor for image, video and display distortions , author=. arXiv preprint arXiv:2401.11485 , year=

    Pith/arXiv arXiv

  34. [34]

    arXiv preprint arXiv:2304.13625 , year=

    HDR-VDP-3: A multi-metric for predicting image differences, quality and contrast distortions in high dynamic range and regular content , author=. arXiv preprint arXiv:2304.13625 , year=

    Pith/arXiv arXiv

  35. [35]

    Signal Processing: Image Communication , volume=

    HDR-VQM: An objective quality measure for high dynamic range video , author=. Signal Processing: Image Communication , volume=. 2015 , publisher=

    2015

  36. [36]

    2025 IEEE International Conference on Multimedia and Expo Workshops (ICMEW) , pages=

    ICME 2025 Generalizable HDR and SDR Video Quality Measurement Grand Challenge , author=. 2025 IEEE International Conference on Multimedia and Expo Workshops (ICMEW) , pages=. 2025 , organization=

    2025

  37. [37]

    The method of paired comparisons , author=

    Rank analysis of incomplete block designs: I. The method of paired comparisons , author=. Biometrika , volume=. 1952 , publisher=

    1952

  38. [38]

    1978 , publisher=

    The Rating of Chessplayers, Past and Present , author=. 1978 , publisher=

    1978

  39. [39]

    Proceedings of the ninth acm international conference on web search and data mining , pages=

    Modeling intransitivity in matchup and comparison data , author=. Proceedings of the ninth acm international conference on web search and data mining , pages=

  40. [40]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Tabnet: Attentive interpretable tabular learning , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  41. [41]

    2021 , howpublished =

    Pranav Sodhani , title =. 2021 , howpublished =

    2021

  42. [42]

    IEEE transactions on Image Processing , volume=

    FSIM: A feature similarity index for image quality assessment , author=. IEEE transactions on Image Processing , volume=. 2011 , publisher=

    2011

  43. [43]

    2023 , version =

    Jon Sneyers , title =. 2023 , version =

    2023

  44. [44]

    Sergey Kastryulin and Dzhamil Zakirov and Denis Prokopenko , year=

  45. [45]

    IEEE Transactions on image processing , volume=

    Image information and visual quality , author=. IEEE Transactions on image processing , volume=. 2006 , publisher=

    2006

  46. [46]

    The thrity-seventh asilomar conference on signals, systems & computers, 2003 , volume=

    Multiscale structural similarity for image quality assessment , author=. The thrity-seventh asilomar conference on signals, systems & computers, 2003 , volume=. 2003 , organization=

    2003

  47. [47]

    IEEE Transactions on Image Processing , year=

    Chen, Chaofeng and Mo, Jiadi and Hou, Jingwen and Wu, Haoning and Liao, Liang and Sun, Wenxiu and Yan, Qiong and Lin, Weisi , title=. IEEE Transactions on Image Processing , year=

  48. [48]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Exploring clip for assessing the look and feel of images , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  49. [49]

    IEEE Transactions on Circuits and Systems for Video Technology , volume=

    Blindly Assess Quality of In-the-Wild Videos via Quality-aware Pre-training and Motion Perception , author=. IEEE Transactions on Circuits and Systems for Video Technology , volume=

  50. [50]

    IEEE Transactions on Image Processing , volume =

    Uncertainty-aware blind image quality assessment in the laboratory and wild , author =. IEEE Transactions on Image Processing , volume =

  51. [51]

    IEEE Transactions on Circuits and Systems for Video Technology , volume=

    Blind Image Quality Assessment Using A Deep Bilinear Convolutional Neural Network , author=. IEEE Transactions on Circuits and Systems for Video Technology , volume=

  52. [52]

    2021 IEEE International Conference on Multimedia & Expo Workshops (ICMEW) , pages=

    Deep learning based full-reference and no-reference quality assessment models for compressed ugc videos , author=. 2021 IEEE International Conference on Multimedia & Expo Workshops (ICMEW) , pages=. 2021 , organization=

    2021

  53. [53]

    IEEE Transactions on Image Processing , year=

    UGC-VQA: Benchmarking blind video quality assessment for user generated content , author=. IEEE Transactions on Image Processing , year=

  54. [54]

    International Journal of Computer Vision , volume=

    Unified quality assessment of in-the-wild videos with mixed datasets training , author=. International Journal of Computer Vision , volume=. 2021 , publisher=

    2021

  55. [55]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    From patches to pictures (PaQ-2-PiQ): Mapping the perceptual space of picture quality , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  56. [56]

    Proceedings of the IEEE international conference on computer vision , pages=

    Rankiqa: Learning from rankings for no-reference image quality assessment , author=. Proceedings of the IEEE international conference on computer vision , pages=

  57. [57]

    IEEE Transactions on Image Processing , volume=

    KonIQ-10k: An ecologically valid database for deep learning of blind image quality assessment , author=. IEEE Transactions on Image Processing , volume=. 2020 , publisher=

    2020

  58. [58]

    IEEE Transactions on image processing , volume=

    No-reference image quality assessment in the spatial domain , author=. IEEE Transactions on image processing , volume=. 2012 , publisher=

    2012

  59. [59]

    Proceedings of the 28th ACM International conference on multimedia , pages=

    Norm-in-norm loss with faster convergence and better performance for image quality assessment , author=. Proceedings of the 28th ACM International conference on multimedia , pages=

  60. [60]

    Proceedings of the 27th ACM international conference on multimedia , pages=

    Quality assessment of in-the-wild videos , author=. Proceedings of the 27th ACM international conference on multimedia , pages=

  61. [61]

    AAAI , year =

    Wang, Jianyi and Chan, Kelvin CK and Loy, Chen Change , title =. AAAI , year =

  62. [62]

    ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    Boosting Image Quality Assessment Performance: Unsupervised Score Fusion by Deep Maximum a Posteriori Estimation , author=. ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2024 , organization=

    2024

  63. [63]

    Advances in Neural Information Processing Systems , volume =

    A Unified Approach to Interpreting Model Predictions , author =. Advances in Neural Information Processing Systems , volume =. 2017 , publisher =

    2017