Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

LEHA-CVQAD: Dataset To Enable Generalized Video Quality Assessment of Compression Artifacts

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A new 6,240-clip dataset with fused human quality labels shows that current video-quality metrics misjudge compression artifacts.

desk verdict A genuinely large compression-focused VQA dataset with a hidden test set and a practical bitrate-cost metric; the main weakness is the sparse MOS fusion that needs a cross-validation check. read the letter →

arxiv 2507.03990 v2 pith:7IMU53NF submitted 2025-07-05 cs.CV

Aleksandr Gushchin , Maksim Smirnov , Dmitriy Vatolin , Anastasia Antsiferova This is my paper
classification cs.CV
keywords videoqualityassessmentcompressionartifactssubjectivedatasetMOSpairwisecomparisonsrate-distortionalignmenterrorcodecevaluationbenchmark
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper introduces LEHA-CVQAD, a large-scale dataset for compression-oriented video quality assessment—the task of automatically predicting how good a compressed video looks to people. It contains 59 source videos encoded into 6,240 clips with 186 codec-preset variants spanning AVC, HEVC, VVC, VP9, and AV1, plus roughly 1.8 million pairwise comparisons and 1.5 thousand MOS ratings fused into a single quality scale. The authors' central claim is that this combination of professionally produced and user-generated content, together with an open/hidden split (1,963 open, 4,277 hidden clips), makes the dataset more diverse and harder than existing compressed-video benchmarks. They also introduce RDAE, a metric that expresses how many bits a quality metric wastes when used to tune a codec. Benchmarking 100 IQA/VQA metrics, they report that popular methods such as PSNR, SSIM, and VMAF show high RDAE and lower correlations with subjective scores, evidence that current metrics do not generalize across modern compression artifacts. If the claims hold, LEHA-CVQAD gives the field a training and evaluation platform for metrics that preserve bitrate-quality ordering.

What carries the argument

The load-bearing mechanism is the fusion model that unifies pairwise and ACR rating data. Pairwise votes are modeled with the Bradley-Terry observer model $F_{BT}(\Delta q, \beta) = 1/(1+\exp(-\Delta q/\beta))$, and each subject's MOS rating is modeled as $\pi_{ik} = q_i + \delta_k + \xi_{ik}$; assuming a linear relation $\omega_i = a \pi_i + b$ between the two latent scales, the paper maximizes the joint likelihood of the pairwise comparison matrix $C$ and the rating matrix $M$ to recover true quality $q_i$ for all clips. A second mechanism is RDAE, defined through under-prediction cost (UPC) and over-compression penalty (OCP): after mapping metric scores and MOS onto a shared perceptual scale with neural optimal transport, RDAE integrates the signed difference between subjective and predicted curves over bitrate, so a lower value means the metric preserves bitrate-quality ordering.

What would settle it

Collect direct quality ratings (ACR MOS) for a random sample of the clips that were not selected as highest, median, or lowest quality in their group, and compare these new ratings to the fused LEHA-CVQAD labels; systematic deviation from the linear pairwise-to-MOS mapping would show the unified scale is biased.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that LEHA-CVQAD is the first compressed-video quality dataset to combine this scale of codec variety with both ranking-based and rating-based subjective labels on one unified quality scale. It does so by collecting pairwise comparisons within each source-video group, converting them to Bradley-Terry and Elo scores, and then collecting MOS only on a representative subset of videos (highest, median, lowest quality per group plus the reference). A maximum-likelihood fusion procedure, adapted from previous work, assumes the latent quality scales behind the two experiments are linearly related and solves for a single quality score $q_i$ for every clip. The paper further claims that its open/hidden split lets the hidden part, dominated by proprietary codecs, serve as a blind test bed, and that its proposed RDAE metric turns quality prediction error into a bitrate cost: underestimation over-compresses, overestimation wastes bits.

Load-bearing premise

The load-bearing premise is that the quality scale built from pair-by-pair comparison votes is linearly related to the scale built from direct quality ratings, so the few directly-rated clips can be stretched to label all 6,240 videos.

Editorial extensions

If this is right

  • If LEHA-CVQAD's labels are trusted, VQA and IQA methods can be trained and tested on a unified quality scale covering 6,240 clips, including compression artifacts from AV1 and VVC, not just AVC.
  • Using RDAE, codec parameter tuning can be guided by a direct bitrate-waste signal: a metric with high RDAE either over-compresses videos or inflates file sizes when used as a quality proxy.
  • The hidden part of the dataset (4,277 clips made with proprietary codecs) provides a blind benchmark, so leaderboard results cannot be fitted to publicly available labels.
  • Because the open part includes UGC and professional content, full-reference and no-reference metric development can move beyond legacy datasets and toward realistic streaming conditions.

Reading between the lines

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

  • A natural check of the fusion step would be to collect ACR MOS on a random sample of clips outside the high/median/low selection; if their scores do not lie on the linear scale derived from pairwise comparisons, the unified labels for unrated clips would be biased.
  • RDAE as defined averages over groups with at least three bitrate points; extending it to continuous encoder parameter sweeps, such as CRF ladders, could turn it into a general codec-optimization objective rather than a benchmark score.
  • If the open part is used for training and the hidden part for testing, one could test whether retraining a metric on LEHA-CVQAD improves its hidden-part RDAE; a large improvement would support the claim that the dataset drives generalization to unseen codecs.
  • The bitrate ceiling of 4,000 kbps for FullHD means the dataset targets low-bitrate streaming; extending the same protocol to 4K/8K or higher bitrates could reveal whether current metrics' failures persist when artifacts are subtler.
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 LEHA-CVQAD, a compression-oriented video quality assessment dataset consisting of 59 source videos, 186 codec-preset variants, and 6,240 distorted clips. Subjective data comprise about 1.8M pairwise comparison responses and about 1.5k MOS ratings; the two are fused into a single quality scale using a Bradley-Terry-based adaptation of the method from [38]. The dataset is split into open and hidden parts to support blind benchmarking. The paper also introduces Rate-Distortion Alignment Error (RDAE), a metric that quantifies how well an objective VQA model preserves bitrate-quality ordering, and reports benchmark results for 100 IQA/VQA metrics. The central claims are that LEHA-CVQAD provides reliable quality labels for all 6,240 clips and that current popular metrics perform poorly on it in terms of both correlation and RDAE.

Significance. If the labels are reliable, LEHA-CVQAD would be a valuable resource: it is substantially larger than most compressed-video datasets, covers a wide range of codecs and presets, combines pairwise and MOS-style labels, and includes an open/hidden split that discourages benchmark overfitting. The diversity analysis in Section 5.1 is a useful contribution, and the split-half consistency check (median SRCC 0.992) is a strong sign that the collected MOS on the rated subset are stable. The paper is also honest about the practical motivation behind RDAE, and the proposed metric addresses a real need in codec tuning. However, the central reliability claim depends on the fusion procedure in Section 4.3, and that procedure is currently validated only on the fitted subset, so the dataset's usefulness for training and benchmarking is not yet fully established.

major comments (4)
  1. [§4.2–4.3, Eq. (3)–(5)] The central claim that reliable MOS-scale labels exist for all 6,240 clips rests on the fusion procedure, but MOS ratings are collected only for a selected subset of videos (highest, median, and lowest quality per reference group, plus the reference) and are then extrapolated to every clip through the assumed linear relation omega_i = a*pi_i + b. The paper validates this only by showing the fitted subset in Fig. 2(b); no held-out MOS clips are used to test predictive accuracy. The statement in Section 4.3 that the fused scores 'showed a high correlation with ranking scores across the entire dataset' is not supported by evidence from unrated clips. Since Table 2, Table 3, and all RDAE values depend on the fused labels, a failure of the linearity or extrapolation assumption would systematically bias the benchmark conclusions. Please add a held-out-MOS validation: collect MOS for a random sample of clips spanning the Bradley-Terry range, fit a and b on a training subset, and report prediction error on the held-out clips; alternatively, replace or supplement the global linear assumption with a nonparametric or local-calibration check.
  2. [§4.3, Eq. (4)] The likelihood in Eq. (4) is written as a product over all N videos and all J subjects, but MOS were collected only for a subset of clips per reference group. As written, the model appears to assume that every subject rated every video, which is inconsistent with the described ACR procedure. Please clarify how missing entries in the rating matrix M are handled and how the product over N is interpreted when most videos have no MOS observation. This is not a purely presentational issue, because the fusion objective in Eq. (5) depends on which videos actually enter the rating likelihood.
  3. [§5.2, Eq. (7) and Table 2] The RDAE computation first maps objective scores and MOS onto a shared perceptual scale using a Neural Optimal Transport map, then evaluates the integrated difference in Eq. (7). The paper does not state whether this map is fitted on the same data used to compute the reported RDAE values. If it is, the RDAE numbers partly measure calibration fit rather than intrinsic rate-distortion ordering ability. Please describe the OT fitting procedure, its inputs, and any regularization, and report a cross-validation or train/test scheme. Please also define the units of the reported RDAE values: the table header says 'Gb', but the integral in Eq. (7) is over a quality difference and a bitrate variable, so the units need to be explained.
  4. [Table 3 and §5.3] The benchmark comparison states that top metrics achieve lower SRCC and PLCC on LEHA-CVQAD than on other datasets, but the table reports no confidence intervals or significance tests for these differences. Given that the compared datasets differ in size, content type, and label protocol, please add bootstrap confidence intervals or a paired statistical test for at least the main comparisons, so readers can assess whether the claimed cross-dataset difficulty is reliable or within noise.
minor comments (4)
  1. [Figure 2 caption] The caption lists two panels labeled '(a)'; the third panel should be labeled '(c)'.
  2. [§5.2 / Table 2] The notation for bitrate b and the integral in Eq. (7) is not fully specified; please state the bitrate units and whether the integral is taken over discrete bitrate points or an interpolated curve.
  3. [§3 / Table 1] The phrase '186 codecs' is used interchangeably with '186 codec-preset variants'; please be consistent in the text so readers do not conflate distinct codecs with preset/CRF combinations.
  4. [§4.2] The reliability statistics (split-half SRCC 0.992 and intra-subject SRCC 0.973) are reported for the rated subset only; please note explicitly that they do not directly validate the extrapolated labels for unrated clips.

Circularity Check

1 steps flagged · score 6.0 of 10

Fused MOS for unrated clips is the Bradley-Terry posterior by construction; Fig. 2(b) 'validation' is self-referential, and full-set benchmarks use this model-extrapolated label as ground truth.

  1. fitted input called prediction [Section 4.3 Eq. (5), Fig. 2(b); Section 5.2 benchmark protocol]
    "Following the assumption of the linear relationship between random variables ω_i from the pairwise comparison experiment and the random variables π_i from the rating experiment, we can state: ω_i = a·π_i + b... As a result of fusing procedure, MOS labeled only on a part of the dataset, showed a high correlation with ranking scores across the entire dataset (Fig. 2 (b))."

    The fused q is the maximizer of Eq. (5), whose likelihood contains the full pairwise matrix C and the sparse MOS matrix M. For the 4,277 hidden and other unrated clips no M term exists, so q_i is determined by the same Bradley-Terry likelihood P(C|q,β) and Gaussian prior that produce the 'ranking scores' plotted on the x-axis of Fig. 2(b). The reported high correlation across the entire dataset is therefore a consistency property of fitting the same pairwise data twice, not an independent validation of the MOS scale.

full rationale

The paper's core artefacts are empirical: 6,240 compressed clips, 1.8M pairwise votes, 1.5k ACR MOS ratings, and a 100-metric benchmark. The fusion procedure is explicitly adapted from the external method of Pérez-Ortiz et al. [38], and the linearity assumption is attributed to that prior work, not to the authors' own results. The only self-citations are to the earlier CVQAD dataset, which is used as a baseline and motivation and is not load-bearing for any derived quantity. RDAE is a direct definition from MOS and model scores and has independent content. The circularity is confined to the fused full-set 'MOS' label: for unrated clips it is, up to prior regularization, the same Bradley-Terry posterior used as the ranking score, so the Fig. 2(b) correlation is by construction and the full-set benchmark numbers inherit that model output. This is a partial circularity in the validation/ground-truth construction, not a collapse of the entire derivation; the pairwise data, the collected MOS subset, and the open-part results remain real empirical evidence.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central quality labels depend on a chosen scale parameter (beta), three fitted fusion parameters (a,b,c), and a learned calibration mapping in RDAE. No new physical entities are introduced.

free parameters (3)
  • beta (Bradley-Terry scale) = 0.9102 (1/ln 3)
    Chosen by hand so a 1-unit quality gap corresponds to 75% preference; sets the unit of the quality scale.
  • a, b, c (fusion model parameters) = Not reported
    Estimated by maximum likelihood in Eq. 5 to map rating-scale scores to the pairwise scale.
  • Neural OT mapping parameters = Not reported
    Learned mappings used in RDAE to bring objective and subjective scores to a shared scale; fitted on the same data being evaluated.
assumptions (4)
  • domain assumption Ratings decompose as r_ik = q_i + d_k + xi_ik with independent normal components and zero-mean xi (Section 4.3, from Janowski and Pinson [20]).
    This subject model justifies the likelihood in Eq. 4 and is inherited from prior work.
  • domain assumption Pairwise preferences follow the Bradley-Terry/Gumbel model with scale beta, and the latent quality scales of pairwise and rating experiments are linearly related (omega_i = a*pi_i + b, Eq. 3).
    This linearity is the key linking assumption in the fusion; if false, the fused quality scale is biased. It is taken from Perez-Ortiz et al. [38].
  • standard math A Gaussian prior is placed on the true quality scores q (Eq. 6).
    Regularization for the optimization in Eq. 5; standard approach.
  • domain assumption MOS collected on a subset of videos (highest, median, lowest quality per group plus reference) can be extrapolated to all videos via the pairwise-derived scale.
    The MOS subset is used to anchor the global scale; this assumes the selected videos represent the quality range.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LEHA-CVQAD: Dataset To Enable Generalized Video Quality Assessment of Compression Artifacts." pith.science (2026). https://pith.science/paper/7IMU53NF

@misc{pith2026250703990,
  author       = {Pith},
  title        = {Pith review of: LEHA-CVQAD: Dataset To Enable Generalized Video Quality Assessment of Compression Artifacts},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7IMU53NF}},
  note         = {Machine review of arXiv:2507.03990}
}
read the original abstract

We propose the LEHA-CVQAD (Large-scale Enriched Human-Annotated Compressed Video Quality Assessment) dataset, which comprises 6,240 clips for compression-oriented video quality assessment. 59 source videos are encoded with 186 codec-preset variants, 1.8M pairwise, and 1.5k MOS ratings are fused into a single quality scale; part of the videos remains hidden for blind evaluation. We also propose Rate-Distortion Alignment Error (RDAE), a novel evaluation metric that quantifies how well VQA models preserve bitrate-quality ordering, directly supporting codec parameter tuning. Testing IQA/VQA methods reveals that popular VQA metrics exhibit high RDAE and lower correlations, underscoring the dataset challenges and utility. The open part and the results of LEHA-CVQAD are available at https://aleksandrgushchin.github.io/lcvqad/

Figures

Figures reproduced from arXiv: 2507.03990 by the authors.

Figure 1
Figure 1. Statistics (Range, Uniformity) of various complexity metrics for LEHA-CVQAD and other datasets. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. (a) Distribution of MOS in our subjective experiments; (b) Bradley-Terry scores Vs. Fused subjective score; (a) Visual [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

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. Full citation record

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

    cs.CV 2026-07 conditional novelty 7.0 of 10

    A 300+ device crowd-sourced VQA dataset plus Blade-Chest aggregation and a condition-adaptation MLP let standard metrics predict quality orderings under real mobile viewing conditions far better than unadapted baselines.

Reference graph

Works this paper leans on

51 extracted references · 30 canonical work pages · cited by 1 Pith paper

  1. [38]

    María Pérez-Ortiz, Aliaksei Mikhailiuk, Emin Zerman, Vedad Hulusic, Giuseppe Valenzise, and Rafał K. Mantiuk. From Pairwise Comparisons and Rating to a Unified Quality Scale. IEEE Transactions on Image Processing 29 (2020), 1139–1151. doi:10.1109/TIP.2019.2936103

    arXiv 2020

  2. [1]

    https://lmsys

    Chatbot Arena: Benchmarking LLMs in the Wild with Elo Ratings. https://lmsys. org/blog/2023-05-03-arena/

    work page 2023

  3. [2]

    Video compres- sion dataset and benchmark of learning-based video-quality metrics

    Anastasia Antsiferova, Sergey Lavrushkin, Maksim Smirnov, Aleksandr Gushchin, Dmitriy Vatolin, and Dmitriy Kulikov. Video compres- sion dataset and benchmark of learning-based video-quality metrics. In Advances in Neural Information Processing Systems , S. Koyejo, S. Mo- hamed, A. Agarwal, D. Belgrave, K. Cho, and A. Oh (Eds.), Vol. 35. 13814–13825. https...

    work page 2022

  4. [3]

    Video compression dataset and bench- mark of learning-based video-quality metrics

    Anastasia Antsiferova, Sergey Lavrushkin, Maksim Smirnov, Aleksandr Gushchin, Dmitriy Vatolin, and Dmitriy Kulikov. Video compression dataset and bench- mark of learning-based video-quality metrics. Advances in Neural Information Processing Systems 35 (2022), 13814–13825

    work page 2022

  5. [4]

    Nabajeet Barman, Emmanuel Jammeh, Seyed Ali Ghorashi, and Maria G. Martini. No-Reference Video Quality Estimation Based on Machine Learning for Passive Gaming Video Streaming Applications. IEEE Access 7 (2019), 74511–74527. doi:10. 1109/ACCESS.2019.2920477

    work page arXiv 2019

  6. [5]

    Elo Uncovered: Robustness and Best Practices in Language Model Evaluation

    Meriem Boubdir, Edward Kim, Beyza Ermis, Sara Hooker, and Marzieh Fadaee. Elo Uncovered: Robustness and Best Practices in Language Model Evaluation. arXiv:2311.17295 [cs.CL] https://arxiv.org/abs/2311.17295

    arXiv

  7. [6]

    Ralph Allan Bradley and Milton E. Terry. Rank Analysis of Incomplete Block Designs: I. The Method of Paired Comparisons.Biometrika 39, 3/4 (1952), 324–345. http://www.jstor.org/stable/2334029

    arXiv 1952

  8. [7]

    A Subjective Study for the Design of Multi-resolution ABR Video Streams with the VP9 Codec

    Chao Chen, Sasi Inguva, Andrew Rankin, and Anil Kokaram. A Subjective Study for the Design of Multi-resolution ABR Video Streams with the VP9 Codec. Electronic Imaging 28, 2 (2016), 1–1. doi:10.2352/ISSN.2470-1173.2016.2.VIPC-235

    arXiv 2016

Show all 51 references
  1. [8]

    Redundant representation for network video streaming using reconstructed P-frames and SP-frames

    Gene Cheung and Wai-tian Tan. Redundant representation for network video streaming using reconstructed P-frames and SP-frames. In2010 IEEE International Conference on Acoustics, Speech and Signal Processing . 2322–2325. doi:10.1109/ ICASSP.2010.5496025

  2. [9]

    Gonzalez, and Ion Stoica

    Wei-Lin Chiang, Lianmin Zheng, Ying Sheng, Anastasios Nikolas Angelopoulos, Tianle Li, Dacheng Li, Hao Zhang, Banghua Zhu, Michael Jordan, Joseph E. Gonzalez, and Ion Stoica. Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference. arXiv:2403.04132 [cs.AI]

    arXiv

  3. [10]

    Ward, and Alan F

    Iya Chivileva, Philip Lynch, Tomás E. Ward, and Alan F. Smeaton. Measur- ing the Quality of Text-to-Video Model Outputs: Metrics and Dataset. ArXiv abs/2309.08009 (2023). https://api.semanticscholar.org/CorpusID:262012984

    2023 arXiv

  4. [11]

    Averaging Correlations: Expected Values and Bias in Combined Pearson rs and Fisher’s z Transformations

    David Corey, William Dunlap, and Michael Burke. Averaging Correlations: Expected Values and Bias in Combined Pearson rs and Fisher’s z Transformations. Journal of General Psychology - J GEN PSYCHOL 125 (07 1998), 245–261. doi:10. 1080/00221309809595548

    1998

  5. [12]

    De Simone, M

    F. De Simone, M. Tagliasacchi, M. Naccari, S. Tubaro, and T. Ebrahimi. A H.264/AVC video database for the evaluation of quality metrics. In 2010 IEEE International Conference on Acoustics, Speech and Signal Processing . 2430–2433. doi:10.1109/ICASSP.2010.5496296

    2010

  6. [13]

    IC9600: A Benchmark Dataset for Automatic Image Complexity Assessment

    Tinglei Feng, Yingjie Zhai, Jufeng Yang, Jie Liang, Deng-Ping Fan, Jing Zhang, Ling Shao, and Dacheng Tao. IC9600: A Benchmark Dataset for Automatic Image Complexity Assessment. IEEE Transactions on Pattern Analysis and Machine Intelligence 01 (2023), 1–17. doi:10.1109/TPAMI.2...

    2023

  7. [14]

    In-capture mobile video distortions: A study of subjective behavior and objective algorithms

    Deepti Ghadiyaram, Janice Pan, Alan C Bovik, Anush Krishna Moorthy, Prasanjit Panda, and Kai-Chieh Yang. In-capture mobile video distortions: A study of subjective behavior and objective algorithms. IEEE Transactions on Circuits and Systems for Video Technology 28, 9 (2017), 2061–2077

    2017

  8. [15]

    Measuring Colourfulness in Natural Images

    David Hasler and Sabine Suesstrunk. Measuring Colourfulness in Natural Images. Proceedings of SPIE - The International Society for Optical Engineering 5007 (06 2003), 87–95. doi:10.1117/12.477378

    2003 doi

  9. [16]

    Best Practices for QoE Crowdtesting: QoE Assessment With Crowdsourcing

    Tobias Hossfeld, Christian Keimel, Matthias Hirth, Bruno Gardlo, Julian Habigt, Klaus Dieopold, and Phuoc Tran-Gia. Best Practices for QoE Crowdtesting: QoE Assessment With Crowdsourcing. Multimedia, IEEE Transactions on 16 (02 2014), 541–558. doi:10.1109/TMM.2013.2291663

    2014

  10. [17]

    The Konstanz natural video database (KoNViD-1k)

    Vlad Hosu, Franz Götz-Hahn, Mohsen Jenadeleh, Hanhe Lin, Hui Men, Tamás Szirányi, Shujun Li, and Dietmar Saupe. The Konstanz natural video database (KoNViD-1k). 1–6. doi:10.1109/QoMEX.2017.7965673

    2017

  11. [18]

    Recommended encoding settings for IBM Watson Media

    IBM. Recommended encoding settings for IBM Watson Media. https://support.video.ibm.com/hc/en-us/articles/207852117-Internet- connection-and-recommended-encoding-settings. Accessed: 2022-08-12

    2022

  12. [19]

    Recommended encoding settings for Twitch streaming

    Twitch Inc. Recommended encoding settings for Twitch streaming. https:// stream.twitch.tv/encoding/. Accessed: 2022-08-12

    2022

  13. [20]

    The accuracy of subjects in a quality experiment: A theoretical subject model

    Lucjan Janowski and Margaret Pinson. The accuracy of subjects in a quality experiment: A theoretical subject model. IEEE Transactions on Multimedia 17, 12 (2015), 2210–2224

    2015

  14. [21]

    BVI-CC: A Dataset for Research on Video Compression and Quality Assessment

    Angeliki Katsenou, Fan Zhang, Mariana Afonso, Goce Dimitrov, and David Bull. BVI-CC: A Dataset for Research on Video Compression and Quality Assessment. Frontiers in Signal Processing 2 (04 2022), 874200. doi:10.3389/frsip.2022.874200

    2022

  15. [22]

    The TUM high definition video datasets

    Christian Keimel, Arne Redl, and Klaus Diepold. The TUM high definition video datasets. In 2012 Fourth international workshop on quality of multimedia experience. IEEE, 97–102

    2012

  16. [23]

    Kernel neural optimal transport

    Alexander Korotin, Daniil Selikhanovych, and Evgeny Burnaev. Kernel neural optimal transport. arXiv preprint arXiv:2205.15269 (2022)

    2022 arXiv

  17. [24]

    Norm-in-norm loss with faster convergence and better performance for image quality assessment

    Dingquan Li, Tingting Jiang, and Ming Jiang. Norm-in-norm loss with faster convergence and better performance for image quality assessment. InProceedings of the 28th ACM International Conference on Multimedia . 789–797

  18. [25]

    Unified quality assessment of in- the-wild videos with mixed datasets training

    Dingquan Li, Tingting Jiang, and Ming Jiang. Unified quality assessment of in- the-wild videos with mixed datasets training. International Journal of Computer Vision 129, 4 (2021), 1238–1257

    2021

  19. [26]

    UGC-VIDEO: perceptual quality assessment of user-generated videos

    Yang Li, Shengbin Meng, Xinfeng Zhang, Shiqi Wang, Yue Wang, and Siwei Ma. UGC-VIDEO: perceptual quality assessment of user-generated videos. arXiv:1908.11517 [cs.MM] https://arxiv.org/abs/1908.11517

    1908 arXiv

  20. [27]

    UGC-VIDEO: Perceptual Quality Assessment of User-Generated Videos

    Yang Li, Shengbin Meng, Xinfeng Zhang, Shiqi Wang, Yue Wang, and Siwei Ma. UGC-VIDEO: Perceptual Quality Assessment of User-Generated Videos. In 2020 IEEE Conference on Multimedia Information Processing and Retrieval (MIPR). 35–38. doi:10.1109/MIPR49039.2020.00015

    2020

  21. [28]

    VMAF: The journey continues

    Zhi Li, Christos Bampis, Julie Novak, Anne Aaron, Kyle Swanson, Anush Moorthy, and JD Cock. VMAF: The journey continues. Netflix Technology Blog 25 (2018)

    2018

  22. [29]

    MCL-V: A streaming video quality assessment database

    Joe Yuchieh Lin, Rui Song, Chi-Hao Wu, TsungJung Liu, Haiqiang Wang, and C-C Jay Kuo. MCL-V: A streaming video quality assessment database. Journal of Visual Communication and Image Representation 30 (2015), 1–9

    2015

  23. [30]

    Joe Yuchieh Lin, Rui Song, Chi-Hao Wu, TsungJung Liu, Haiqiang Wang, and C.-C. Jay Kuo. MCL-V: A streaming video quality assessment database. Journal of Visual Communication and Image Representation 30 (2015), 1–9. doi:10.1016/j. jvcir.2015.02.012

    2015 doi

  24. [31]

    Least squares quantization in PCM.IEEE transactions on information theory 28, 2 (1982), 129–137

    Stuart Lloyd. Least squares quantization in PCM.IEEE transactions on information theory 28, 2 (1982), 129–137

    1982

  25. [32]

    Kvq: Kwai video quality assessment for short-form videos

    Yiting Lu, Xin Li, Yajing Pei, Kun Yuan, Qizhi Xie, Yunpeng Qu, Ming Sun, Chao Zhou, and Zhibo Chen. Kvq: Kwai video quality assessment for short-form videos. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 25963–25973

  26. [33]

    A Study of High Frame Rate Video Formats

    Alex Mackin, Fan Zhang, and David Bull. A Study of High Frame Rate Video Formats. IEEE Transactions on Multimedia PP (11 2018), 1–1. doi:10.1109/TMM. 2018.2880603

    2018

  27. [34]

    Subjective and Objective Quality Assessment of High Frame Rate Videos

    Pavan Madhusudana, Xiangxu Yu, Neil Birkbeck, Yilin Wang, Balu Adsumilli, and Alan Bovik. Subjective and Objective Quality Assessment of High Frame Rate Videos. IEEE Access PP (07 2021), 1–1. doi:10.1109/ACCESS.2021.3100462

    2021

  28. [35]

    Madhusudana, Xiangxu Yu, Neil Birkbeck, Yilin Wang, Balu Adsumilli, and Alan C

    Pavan C. Madhusudana, Xiangxu Yu, Neil Birkbeck, Yilin Wang, Balu Adsumilli, and Alan C. Bovik. Subjective and Objective Quality Assessment of High Frame Rate Videos. IEEE Access 9 (2021), 108069–108082. doi:10.1109/access.2021. 3100462

    2021 doi

  29. [36]

    On Thurstone’s Model for Paired Comparisons and Ranking Data

    Albert Maydeu-Olivares. On Thurstone’s Model for Paired Comparisons and Ranking Data. InNew Developments in Psychometrics, H. Yanai, A. Okada, K. Shige- masu, Y. Kano, and J. J. Meulman (Eds.). 519–526

  30. [37]

    CVD2014—A Database for Evaluating No-Reference Video Quality Assessment Algorithms

    Mikko Nuutinen, Toni Virtanen, Mikko Vaahteranoksa, Tero Vuori, Pirkko Oitti- nen, and Jukka Häkkinen. CVD2014—A Database for Evaluating No-Reference Video Quality Assessment Algorithms. IEEE Transactions on Image Processing 25, 7 (2016), 3073–3086. doi:10.1109/TIP.2016.2562513

    2016

  31. [39]

    Kalpana Seshadrinathan, Rajiv Soundararajan, Alan Conrad Bovik, and Lawrence K. Cormack. Study of Subjective and Objective Quality Assess- ment of Video. IEEE Transactions on Image Processing 19, 6 (2010), 1427–1441. doi:10.1109/TIP.2010.2042111

    2010

  32. [40]

    Large-Scale Study of Perceptual Video Quality

    Zeina Sinno and Alan Conrad Bovik. Large-Scale Study of Perceptual Video Quality. IEEE Transactions on Image Processing 28, 2 (2019), 612–627. doi:10.1109/ TIP.2018.2869673

    2019

  33. [41]

    ViS3: An algorithm for video quality assessment via analysis of spatial and spatiotemporal slices

    Phong Vu and Damon Chandler. ViS3: An algorithm for video quality assessment via analysis of spatial and spatiotemporal slices. Journal of Electronic Imaging 23 (02 2014), 013016. doi:10.1117/1.JEI.23.1.013016

    2014 doi

  34. [42]

    Haiqiang Wang, Ioannis Katsavounidis, Jiantong Zhou, Jeonghoon Park, Shawmin Lei, Xin Zhou, Man-On Pun, Xin Jin, Ronggang Wang, Xu Wang, Yun Zhang, Jiwu Huang, Sam Kwong, and C.-C. Jay Kuo. VideoSet. J. Vis. Comun. Image Represent. 46, C (July 2017), 292–302. doi:10.1016/j.jvc...

    2017 doi

  35. [43]

    YouTube UGC Dataset for Video Compression Research

    Yilin Wang, Sasi Inguva, and Balu Adsumilli. YouTube UGC Dataset for Video Compression Research. In 2019 IEEE 21st International Workshop on Multimedia Signal Processing (MMSP). doi:10.1109/mmsp.2019.8901772

    2019

  36. [45]

    Fast-vqa: Efficient end-to-end video quality assessment with fragment sampling

    Haoning Wu, Chaofeng Chen, Jingwen Hou, Liang Liao, Annan Wang, Wenxiu Sun, Qiong Yan, and Weisi Lin. Fast-vqa: Efficient end-to-end video quality assessment with fragment sampling. In European conference on computer vision . Springer, 538–554. Conference acronym ’ACM MM, Octo...

    2025

  37. [46]

    Exploring Video Quality Assess- ment on User Generated Contents from Aesthetic and Technical Perspectives

    Haoning Wu, Erli Zhang, Liang Liao, Chaofeng Chen, Jingwen Hou, Annan Wang, Wenxiu Sun, Qiong Yan, and Weisi Lin. Exploring Video Quality Assess- ment on User Generated Contents from Aesthetic and Technical Perspectives. arXiv:2211.04894 [cs.CV] https://arxiv.org/abs/2211.04894

    arXiv

  38. [47]

    Towards Explainable In-the-Wild Video Quality Assessment: A Database and a Language-Prompted Approach

    Haoning Wu, Erli Zhang, Liang Liao, Chaofeng Chen, Jingwen Hou, Annan Wang, Wenxiu Sun, Qiong Yan, and Weisi Lin. Towards Explainable In-the-Wild Video Quality Assessment: A Database and a Language-Prompted Approach. In Proceedings of the 31st ACM International Conference on M...

  39. [48]

    Q-Align: Teaching LMMs for Visual Scoring via Discrete Text-Defined Levels

    Haoning Wu, Zicheng Zhang, Weixia Zhang, Chaofeng Chen, Liang Liao, Chunyi Li, Yixuan Gao, Annan Wang, Erli Zhang, Wenxiu Sun, Qiong Yan, Xiongkuo Min, Guangtao Zhai, and Weisi Lin. Q-Align: Teaching LMMs for Visual Scoring via Discrete Text-Defined Levels. arXiv:2312.17090 [c...

    arXiv

  40. [50]

    Patch-VQ: ‘Patching Up’ the Video Quality Problem

    Zhenqiang Ying, Maniratnam Mandal, Deepti Ghadiyaram, and Alan Bovik. Patch-VQ: ‘Patching Up’ the Video Quality Problem. In2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . 14014–14024. doi:10.1109/ cvpr46437.2021.01380

  41. [51]

    Recommended encoding settings for Youtube

    YouTube. Recommended encoding settings for Youtube. https://support.google. com/youtube/answer/2853702. Accessed: 2022-08-12

    2022

  42. [52]

    Bampis, Balu Adsumilli, and Alan C

    Xiangxu Yu, Neil Birkbeck, Yilin Wang, Christos G. Bampis, Balu Adsumilli, and Alan C. Bovik. Predicting the Quality of Compressed Videos With Pre- Existing Distortions. IEEE Transactions on Image Processing 30 (2021), 7511–7526. doi:10.1109/tip.2021.3107213

    2021

  43. [53]

    MD-VQA: Multi-dimensional quality assessment for UGC live videos

    Zicheng Zhang, Wei Wu, Wei Sun, Danyang Tu, Wei Lu, Xiongkuo Min, Ying Chen, and Guangtao Zhai. MD-VQA: Multi-dimensional quality assessment for UGC live videos. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 1746–1755. Received 30 may 2025

    2025

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.