Pith. sign in

REVIEW 3 major objections 6 minor 44 references

SAIL: Perceptual Quality-Aware Rate Control for Cloud Gaming

T0 review · 3 major / 6 minor · reviewed 2026-07-14 · grok-4.5

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

Pith's one-line read Cloud gaming can cut bandwidth nearly in half without losing visual quality by stopping at the perceptually lossless bitrate.

desk verdict Production QARC that actually ships: 44% bandwidth cut and lower latency on a live multi-million-user cloud-gaming platform, with the post-encode loop and CC co-design as the real systems contribution. read the letter →

arxiv 2607.11231 v1 pith:MGDX2TCI submitted 2026-07-13 cs.NI cs.MMeess.IV

Houde Qian , Chenglei Wu , Jiaxing Zhang , Rui-Xiao Zhang , Jing Wang , Meijia Song , Sijia Chen , Xiaozhong Xu
show 3 more authors
Zhi Wang Lifeng Sun Honghao Liu
This is my paper
classification cs.NIcs.MMeess.IV
keywords cloudgamingquality-awareratecontrolperceptuallylosslesspost-encodingVQAencoder-drivenqualityestimationcongestionco-designbandwidthsavingsNVENC
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

Cloud gaming platforms spend 30–60% of operating cost on bandwidth because they push every frame to the highest rate congestion control allows, chasing fidelity indistinguishable from local play. Past a frame-dependent “perceptually lossless” point, extra bits buy almost nothing the eye can notice, so the stream systematically wastes capacity. SAIL is a production rate-control loop that sits after the hardware encoder, estimates quality from free encoder statistics in under a millisecond, and caps each frame at that lossless threshold. Hybrid control keeps steady scenes thrifty while absorbing complexity spikes with a dynamically expanded buffer and an immediate bitrate shoot-up; a coordinated congestion-control change keeps the network from forgetting spare capacity. Deployed on a major commercial platform, SAIL cut bandwidth 44% and end-to-end latency 8% while preserving (and sometimes improving) user-rated quality across tens of millions of sessions.

What carries the argument

Post-encoding SAIL control loop: encoder-driven VQA (QP, SATD, motion vectors, intra ratio, frame size → distilled SQoE estimator with asymmetric loss and guardband) feeding hybrid rate control (asymmetric inverse-proportion bitrate steps + dynamic VBV expansion + shoot-up) and a modified congestion controller that probes capacity so under-sending does not shrink the estimated window.

What would settle it

A controlled subjective test in which quality is forced below the low-quality threshold for exactly one frame (versus two or more) during high-motion gameplay; if mean opinion scores fall significantly for the single-frame case, the recovery budget disappears and the reactive architecture cannot guarantee losslessness.

Watch

Extended reading notes

Core claim

Best-effort bitrate allocation in cloud gaming produces large amounts of redundant quality that can be removed by a post-encoding, frame-level quality-aware controller. Using only zero-cost encoder outputs, a lightweight distilled model can estimate a gaming-specific perceptual score accurately enough near the lossless threshold that hybrid reactive-plus-proactive rate control, plus network co-design, delivers 44% bandwidth savings and lower latency without degrading perceived quality at production scale.

Load-bearing premise

A single-frame quality drop below the low-quality threshold is essentially invisible to players, so the inherent one-frame lag of post-encode feedback is an acceptable recovery budget.

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

3 major / 6 minor

Summary. The paper presents SAIL, a production quality-aware rate control (QARC) system for cloud gaming that caps per-frame encoding bitrate at a perceptually lossless (p-lossless) threshold to eliminate redundant quality waste under best-effort allocation. SAIL uses a post-encoding loop with three components: (i) a lightweight encoder-driven VQA student distilled from a domain-specific SQoE teacher using zero-cost encoder statistics (QP, SATD, MVs, intra ratio, frame size); (ii) hybrid rate control combining asymmetric reactive bitrate steps (inverse-proportion model with EWMA-tracked scale α), dynamic VBV expansion from bandwidth headroom, and aggressive shoot-up on low-quality frames; and (iii) network-aware co-design with Pudica that probes capacity during under-sending to avoid CC underestimation. Motivated by large-scale measurements and user studies (p-lossless at SQoE q★=98.5; single-frame glitches below q_low=90 MOS-negligible), SAIL is fully deployed on the T platform. A month-long A/B evaluation (57k+ sessions) reports 44.27% bandwidth reduction, 8.37% lower E2E latency, improved engagement and stalls, with non-degraded (slightly better) quality/latency MOS versus the production default; offline CBR comparisons and component ablations support the design.

Significance. If the deployment results hold, this is a high-impact systems contribution: it quantifies redundant quality cost in commercial cloud gaming (bandwidth 30–60% of OPEX) and delivers a practical, near-zero-overhead QARC that is compatible with black-box hardware encoders, ultra-low MTP latency, and congestion control. The month-long multi-city A/B with subjective MOS, engagement, and stall metrics, plus offline CBR comparisons and systematic ablations (VQA features/training, gain A, EWMA λ, fitting functions, VBV/step components), is stronger evidence than typical simulation-only QARC work. Credit is due for the post-encoding architecture that respects the one-to-one cost model, the domain-specific SQoE teacher grounded in platform user feedback, and the explicit CC co-design (Pudica*). The work is directly actionable for production platforms and advances the economic viability of cloud gaming at scale.

major comments (3)
  1. [§3.2.3 / Figure 7; §5.1.1] §3.2.3 / Figure 7 (Observation 3) is load-bearing: single-frame SQoE < 90 is treated as perceptually negligible, converting the inherent one-frame post-encode lag into an acceptable recovery budget for reactive shoot-up and VBV expansion. The controlled multi-genre MOS study supports this, and online results show a higher fraction of frames above q★ plus non-degraded Q-MOS. However, the manuscript does not report the distribution of quality-drop lengths (consecutive frames below q_low or q★) under SAIL in production, nor whether residual multi-frame or motion-correlated glitches remain under extreme spikes (e.g., FPS scene cuts). A short production analysis of drop-length CDFs and any correlation with MOS/engagement would make the p-lossless guarantee more falsifiable.
  2. [§5.1.1 / Figure 14 / Table 1] §5.1.1 / Figure 14 / Table 1: The headline 44.27% bandwidth and 8.37% latency gains are central. The CDFs and MOS averages are persuasive at scale, but the paper does not report confidence intervals, session-level variance, or breakdowns by game type (MOBA/SPG/RPG/FPS) and network (Ethernet/Wi-Fi) for the savings and latency reductions. Given that p-lossless bitrate demand varies sharply by scene (Figure 2) and the hybrid controller is content-adaptive, a stratified summary would confirm that the average is not driven by a subset of simple scenes and would strengthen the claim of broad production viability.
  3. [§4.3.1 / Eqs. (2)–(5); Appendix C] §4.3.1 Eqs. (2)–(5) and Appendix C: The inverse-proportion bitrate–SQoE model and EWMA-tracked α are the core of reactive stabilization. Figure 13 and the linear/log ablation (Figure 17) support the functional form offline, and A=1.8 / λ=0.7 are ablated. The manuscript should more clearly state the operating regime in which α tracking is assumed valid (how slowly complexity may drift before shoot-up/VBV dominate) and what happens when the sign-consistency gate in Eq. (5) repeatedly rejects updates after a complexity jump. A brief failure-mode discussion or production trace of α and step sizes around spikes would close this gap without changing the design.
minor comments (6)
  1. [Figure 14; Figure 15] Figures 14 and 15 label the proposed system as "BOAT" while the paper title and text use SAIL. This naming inconsistency should be corrected throughout captions and legends to avoid reader confusion about what was deployed.
  2. [Abstract; §5.1.1] Abstract and §1 state "serving tens of millions of users and accumulating billions of hours"; §5.1.1 reports a month-long evaluation with >57k sessions. Clarify which numbers refer to cumulative platform deployment of SAIL versus the controlled A/B window.
  3. [§3.2.1 / Figure 5] Figure 5 caption and text refer to "predicted SQoE/VMAF scores" and normalized DMOS; a short note on how DMOS was normalized to the 100-point scale used for q★=98.5 would improve reproducibility of the threshold.
  4. [§4.4] §4.4: The capacity-probing parameters (αR threshold, T_wd, ρ, burst/pace pattern) are only partially specified. Listing the production defaults (or ranges) would help replication of Pudica*.
  5. [Throughout] Typos / polish: "motivat ion" spacing artifacts in headings; "perceptually lossless (p-lossless)threshold" missing space (§1); "frame-dependentperceptually" (§ abstract body). Standard copy-edit pass recommended.
  6. [§5.3; Appendices D–E] Appendix D/E ablations are valuable; consider promoting a one-paragraph summary of the most sensitive parameters (A, μ, VBV) into the main §5 micro-benchmarks for readers who skip appendices.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: empirical systems paper whose central claims rest on independent large-scale deployment and MOS studies, not on self-referential derivation.

full rationale

SAIL is a production systems paper. Its strongest claims (44.27% bandwidth reduction, 8.37% E2E latency reduction, no quality degradation) are measured outcomes from a month-long A/B deployment on the T platform (Figure 14, Table 1) and offline CBR comparisons (Figure 15), not theoretical predictions forced by construction. The SQoE teacher is trained on independent user ratings; the student is a distilled lightweight regressor from zero-cost encoder statistics (QP, SATD, MV, Intra Ratio, frame size); the p-lossless threshold q★=98.5 and the +0.2 guardband are set from measured DMOS alignment and residual error distributions, respectively. The inverse-proportion bitrate-SQoE model (Appendix C) is an offline curve fit used only to supply a baseline step size that is then scaled online by an EWMA-adapted α driven by observed Δq; the adaptation equation is not tautological. Observation 3 (single-frame SQoE<90 is MOS-negligible) is itself an empirical MOS result, not a definition. The sole minor self-citation is to Pudica (overlapping authors) as the base congestion-control substrate that SAIL then modifies; this is ordinary engineering reuse and is not load-bearing for the quality-aware rate-control claims or the deployment numbers. No self-definitional loop, no fitted quantity re-labeled as an independent prediction, and no uniqueness theorem imported from prior author work appear in the derivation chain. Score 1 reflects only the non-load-bearing self-citation; the paper is otherwise self-contained against external benchmarks.

Assumptions & free parameters 7 free parameters · 4 assumptions · 2 invented entities

The central claim rests on empirical thresholds and control constants measured or ablated on the authors’ platform and user studies, plus the domain premise that hardware-encoder side-channels suffice for p-lossless decisions. No new physical entities; free parameters are the usual systems knobs.

free parameters (7)
  • p-lossless SQoE threshold q★ = 98.5
    Set to 98.5 from user-study DMOS mapping; all rate decisions target this value.
  • error-compensating guardband = 0.2
    Raised control target by measured average over-estimation of 0.2.
  • asymmetric loss weight μ = 8
    Penalty for over-estimation near threshold; chosen by ablation (Fig. 19).
  • bitrate-increase safety gain A = 1.8
    Multiplies recovery step; selected by quality/bandwidth trade-off ablation (Fig. 20).
  • EWMA smoothing λ for α = 0.7
    Tracks scene complexity; set to 0.7 after minor-sensitivity ablation (Fig. 21).
  • VBV expansion formula N·C/R and underflow 1.5× threshold = heuristic
    Heuristic buffer scaling and reset rule; no theoretical optimum claimed.
  • low-quality threshold q_low = 90
    SQoE 90 used to define glitch duration experiments and shoot-up trigger.
assumptions (4)
  • domain assumption Hardware encoder statistics (QP, SATD, MV counts, intra ratio, frame size) contain sufficient information to predict gaming perceptual quality near the p-lossless operating point.
    Foundation of the zero-cost VQA; validated by correlation plots and ablation but not proven for all future codecs or content.
  • domain assumption A single-frame quality drop below SQoE 90 is imperceptible to users across the tested game genres.
    Observation 3 / Figure 7; converts one-frame control lag into a usable recovery budget.
  • ad hoc to paper Bitrate–SQoE relationship can be locally approximated by an inverse-proportion function whose scale factor α varies slowly enough for EWMA tracking.
    Appendix C derivation and Figure 13; enables the analytic step-size formula.
  • domain assumption Congestion-control bandwidth estimate can be kept accurate by synthetic BUR compensation and occasional burst probing even when actual send rate is far below capacity.
    §4.4 adaptation of Pudica; required for shoot-up readiness.
invented entities (2)
  • SQoE (domain-specific DeepVQA teacher)
    purpose: Provide gaming-tuned ground-truth quality labels that VMAF fails to match.
    Trained on platform user feedback; independent of the student model but still proprietary.
  • SAIL hybrid controller (asymmetric step + dynamic VBV + shoot-up)
    purpose: Realize post-encode QARC under one-frame lag and non-stationary complexity.
    System-level invention; evaluated only inside the authors’ stack.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SAIL: Perceptual Quality-Aware Rate Control for Cloud Gaming." pith.science (2026). https://pith.science/paper/MGDX2TCI

@misc{pith2026260711231,
  author       = {Pith},
  title        = {Pith review of: SAIL: Perceptual Quality-Aware Rate Control for Cloud Gaming},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MGDX2TCI}},
  note         = {Machine review of arXiv:2607.11231}
}
read the original abstract

Cloud gaming streams cloud-rendered frames under strict motion-to-photon latency, yet its at-scale viability is increasingly constrained by bandwidth cost: in our study of the T cloud gaming platform, bandwidth accounts for 30-60% of total operating expense. This high bandwidth consumption stems from a fidelity-first objective of making the stream perceptually indistinguishable from local gameplay. It drives production systems toward best-effort bitrate allocation that pushes the encoder to the highest rate allowed by congestion control. However, the bitrate-perception relationship saturates: beyond a frame-dependent perceptually lossless threshold, additional bits yield negligible perceptual improvement, creating systematic redundant quality that wastes bandwidth. We present SAIL, a production quality-aware rate control system with the goal of achieving perceptually lossless quality while avoiding unnecessary bandwidth waste. SAIL adopts a post-encoding architecture to enable millisecond-scale feedback at near-zero overhead. It comprises three key designs: (i) an encoder-driven quality assessment model that leverages zero-cost encoder outputs for real-time quality estimation; (ii) a hybrid rate control mechanism that balances steady-state adaptation with dynamic spike absorption; and (iii) a network-aware strategy that coordinates with congestion control to prevent capacity underestimation. SAIL has been fully deployed on the T cloud gaming platform and reduces bandwidth consumption by 44.27% and end-to-end latency by 8.37% without degrading perceived quality, serving tens of millions of users and accumulating billions of hours of total gameplay.

Figures

Figures reproduced from arXiv: 2607.11231 by the authors.

Figure 1
Figure 1. A typical cloud gaming system. To address this critical gap, we propose SAIL, a QARC solution tailored for cloud gaming at scale. Specifically de￾signed to meet the demands of large-scale deployment, SAIL achieves precise, real-time bitrate control through three syn￾ergistic components: (i) Encoder-Driven VQA (§ 4.2). Through an in-depth anal￾ysis of encoder outputs, we identify specific statistics that exhibit a st… view at source ↗
Figure 2
Figure 2. Comparison of video quality under differ [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. QARC architecture under different scenarios. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Computational latency of VMAF/CNN. 85.0 87.5 90.0 92.5 95.0 97.5 100.0 Predicted Score 80 85 90 95 100 105 DMOS (normalized) SQoE VMAF [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 8
Figure 8. Figure 8: Relatively big frame ratio under differ￾ent VBV buffer sizes. 2 4 6 8 10 VBV Buffer Size (# Frames) 0.0 2.5 5.0 7.5 10.0 Frame Ratio (%) MOBA SPG RPG FPS [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 10
Figure 10. Figure 10: SAIL system overview. reliability, we build our congestion control upon Pudica [34], a state-of-the-art solution tailored for cloud gaming’s system requirements. Pudica maintains accurate control by estimat￾ing the BUR and dynamically switching between Multiplica￾tive…
Figure 12
Figure 12. Figure 12: Intra ratio and motion vectors effectively [PITH_FULL_IMAGE:figures/full_fig_p008_12.png]
Figure 14
Figure 14. Figure 14: Online deployment performance of SAIL comparing to [PITH_FULL_IMAGE:figures/full_fig_p011_14.png]
Figure 15
Figure 15. Figure 15: Overall performance of SAIL comparing to CBR by offline simulation. [PITH_FULL_IMAGE:figures/full_fig_p011_15.png]
Figure 16
Figure 16. Figure 16: Evaluation of components in the rate control algorithm. [PITH_FULL_IMAGE:figures/full_fig_p012_16.png]
Figure 19
Figure 19. Figure 19: Ablation study of VQA penalty weight. selected to encompass a broad spectrum of visual content ranging from stable scenes to high-velocity motion. User Test Process. The study was conducted via a web￾based survey interface accessible exclusively through desk￾top PC br…
Figure 20
Figure 20. Figure 20: Ablation study of gain factor. 40.8 40.9 41.0 41.1 41.2 Average Frame Size (kB) 98.760 98.765 98.770 98.775 98.780 98.785 Average SQoE Score Better λ = 0.2 λ = 0.3 λ = 0.4 λ = 0.5 λ = 0.6 λ = 0.7 λ = 0.8 λ = 0.9 [PITH_FULL_IMAGE:figures/full_fig_p016_20.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

44 extracted references · 3 linked inside Pith

  1. [1]

    Ahmad Alhilal, Tristan Braud, Bo Han, and Pan Hui. 2022. Nebula: Reliable low-latency video transmission for mobile cloud gaming. In Proceedings of the ACM Web Conference 2022. 3407–3417

    2022

  2. [2]

    Bonree. 2024. Bonree Data - Bonree One. https://www.bonree.com/. (2024)

    2024

  3. [3]

    Leo Breiman. 2001. Random forests.Machine learning45, 1 (2001), 5–32

    2001

  4. [4]

    Tianqi Chen. 2016. XGBoost: A Scalable Tree Boosting System.Cornell University(2016)

    2016

  5. [5]

    Lark Kwon Choi and Alan C Bovik. 2016. Flicker sensitive motion tuned video quality assessment. In2016 IEEE southwest symposium on image analysis and interpretation (SSIAI). IEEE, 29–32

    2016

  6. [6]

    Lark Kwon Choi and Alan Conrad Bovik. 2018. Video quality assess- ment accounting for temporal visual masking of local flicker.Signal Processing: image communication67 (2018), 182–198

    2018

  7. [7]

    Andrea Di Domenico, Gianluca Perna, Martino Trevisan, Luca Vassio, and Danilo Giordano. 2021. A network analysis on cloud gaming: Stadia, GeForce Now and PSNow.Network1, 3 (2021), 247–260

    2021

  8. [8]

    Sebastian Flinck Lindström, Markus Wetterberg, and Niklas Carlsson

Show all 44 references
  1. [9]

    In2020 IEEE/ACM 13th International Conference on Utility and Cloud Computing (UCC)

    Cloud Gaming: A QoE Study of Fast-paced Single-player and Multiplayer Gaming. In2020 IEEE/ACM 13th International Conference on Utility and Cloud Computing (UCC). 34–45. https://doi.org/10.1109/ UCC48980.2020.00023

    2020

  2. [10]

    Yashuang Guo, Qinghai Yang, Jiayi Liu, and Kyung Sup Kwak. 2017. Quality-aware streaming in heterogeneous wireless networks.IEEE Transactions on Wireless Communications16, 12 (2017), 8162–8174

    2017

  3. [11]

    Tianchi Huang, Rui-Xiao Zhang, Chao Zhou, and Lifeng Sun. 2018. QARC: Video quality aware rate control for real-time video streaming based on deep reinforcement learning. InProceedings of the 26th ACM international conference on Multimedia. 1208–1216

    2018

  4. [12]

    Tianchi Huang, Chao Zhou, Rui-Xiao Zhang, Chenglei Wu, Xin Yao, and Lifeng Sun. 2019. Comyco: Quality-aware adaptive video stream- ing via imitation learning. InProceedings of the 27th ACM international conference on multimedia. 429–437

    2019

  5. [13]

    Xiangjie Huang, Jiayang Xu, Haiping Wang, Hebin Yu, Sandesh Dhawaskar Sathyanarayana, Shu Shi, and Zili Meng

  6. [14]

    InProceedings of the ACM SIGCOMM 2025 Conference

    ACE: Sending Burstiness Control for High-Quality Real-time Communication. InProceedings of the ACM SIGCOMM 2025 Conference. 1182–1198

    2025

  7. [15]

    J Stuart Hunter. 1986. The exponentially weighted moving average. Journal of quality technology18, 4 (1986), 203–210

    1986

  8. [16]

    Michael Jarschel, Daniel Schlosser, Sven Scheuring, and Tobias Hoßfeld

  9. [17]

    In2011 Fifth International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing

    An evaluation of QoE in cloud gaming based on subjective tests. In2011 Fifth International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing. IEEE, 330–335

  10. [18]

    Zhidong Jia, Yihang Zhang, Qingyang Li, and Xinggong Zhang. 2024. Tackling Bit-Rate Variation of RTC Through Frame-Bursting Conges- tion Control. In2024 IEEE 32nd International Conference on Network Protocols (ICNP). IEEE, 1–11

    2024

  11. [19]

    Woojae Kim, Jongyoo Kim, Sewoong Ahn, Jinwoo Kim, and Sanghoon Lee. 2018. Deep video quality assessor: From spatio-temporal visual sensitivity to a convolutional neural aggregation network. InProceed- ings of the European conference on computer vision (ECCV). 219–234

    2018

  12. [20]

    Diederik P Kingma. 2014. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980(2014)

    2014 arXiv

  13. [21]

    Kyungmin Lee, David Chu, Eduardo Cuervo, Johannes Kopf, Yury Degt- yarev, Sergey Grizan, Alec Wolman, and Jason Flinn. 2015. Outatime: Using speculation to enable low-latency continuous interaction for mobile cloud gaming. InProceedings of the 13th Annual International Confere...

    2015

  14. [22]

    Zhi Li, Anne Aaron, Ioannis Katsavounidis, Anush Moorthy, and Megha Manohara. 2016. Toward A Practical Perceptual Video Quality Metric. (2016). https://netflixtechblog.com/ toward-a-practical-perceptual-video-quality-metric-653f208b9652

    2016

  15. [23]

    Yuanxin Liu, Shicheng Li, Yi Liu, Yuxiang Wang, Shuhuai Ren, Lei Li, Sishuo Chen, Xu Sun, and Lu Hou. 2024. Tempcompass: Do video llms really understand videos?arXiv preprint arXiv:2403.00476(2024)

    2024 arXiv

  16. [24]

    Zili Meng, Tingfeng Wang, Yixin Shen, Bo Wang, Mingwei Xu, Rui Han, Honghao Liu, Venkat Arun, Hongxin Hu, and Xue Wei. 2023. Enabling High Quality Real-Time Communications with Adaptive Frame-Rate. In20th USENIX Symposium on Networked Systems Design and Implementation (NSDI 23...

    2023

  17. [25]

    https://www.usenix.org/conference/nsdi23/presentation/meng

  18. [26]

    Anush Krishna Moorthy and Alan Conrad Bovik. 2010. Efficient video quality assessment along temporal trajectories.IEEE transactions on circuits and systems for video technology20, 11 (2010), 1653–1658

    2010

  19. [27]

    Omar Mossad, Khaled Diab, Ihab Amer, and Mohamed Hefeeda. 2021. DeepGame: Efficient Video Encoding for Cloud Gaming. InProceedings of the 29th ACM International Conference on Multimedia (MM ’21). Association for Computing Machinery, New York, NY, USA, 1387–

    2021

  20. [28]

    https://doi.org/10.1145/3474085.3475594

    doi

  21. [29]

    Nvidia. 2026. NVENC Video Encoder API Programming Guide. https://docs.nvidia.com/video-technologies/video-codec-sdk/13.0/ nvenc-video-encoder-api-prog-guide/index.html. (2026)

    2026

  22. [30]

    Nvidia. 2026. NVIDIA Video Codec SDK. https://developer.nvidia.cn/ video-codec-sdk. (2026)

    2026

  23. [31]

    Nvidia. 2026. System Requirements. https://www.nvidia.com/en-us/ geforce-now/system-reqs/#windows-pc. (2026)

    2026

  24. [32]

    Karl Pearson. 1896. VII. Mathematical contributions to the theory of evolution.—III. Regression, heredity, and panmixia.Philosophical Transactions of the Royal Society of London. Series A, containing papers of a mathematical or physical character187 (1896), 253–318

  25. [33]

    Yanyuan Qin, Shuai Hao, Krishna R Pattipati, Feng Qian, Subhabrata Sen, Bing Wang, and Chaoqun Yue. 2019. Quality-aware strategies for optimizing ABR video streaming QoE and reducing data usage. In Proceedings of the 10th ACM multimedia systems conference. 189–200

    2019

  26. [34]

    Devdeep Ray, Connor Smith, Teng Wei, David Chu, and Srinivasan Seshan. 2022. Sqp: Congestion control for low-latency interactive 13 ArXiv, 2026, Online Qian et al. video streaming.arXiv preprint arXiv:2207.11857(2022)

    2022 arXiv

  27. [35]

    Ivan Slivar, Lea Skorin-Kapov, and Mirko Suznjevic. 2016. Cloud gaming QoE models for deriving video encoding adaptation strategies. InProceedings of the 7th international conference on multimedia systems. 1–12

    2016

  28. [36]

    Alex J Smola and Bernhard Schölkopf. 2004. A tutorial on support vector regression.Statistics and computing14, 3 (2004), 199–222

    2004

  29. [37]

    Charles Spearman. 1961. The proof and measurement of association between two things. (1961)

    1961

  30. [38]

    Statista. 2024. Cloud Gaming - Worldwide. https://www.statista.com/ outlook/amo/media/games/cloud-gaming/worldwide. (2024)

    2024

  31. [39]

    Shibo Wang, Shusen Yang, Xiao Kong, Chenglei Wu, Longwei Jiang, Chenren Xu, Cong Zhao, Xuesong Yang, Jianjun Xiao, Xin Liu, Changxi Zheng, Jing Wang, and Honghao Liu. 2024. Pudica: Toward Near- Zero Queuing Delay in Congestion Control for Cloud Gaming. In21st USENIX Symposium ...

    2024

  32. [40]

    Zhou Wang, Ligang Lu, and Alan C Bovik. 2004. Video quality assess- ment based on structural distortion measurement.Signal processing: Image communication19, 2 (2004), 121–132

    2004

  33. [41]

    Saman Zadtootaghaj, Steven Schmidt, and Sebastian Möller. 2018. Mod- eling Gaming QoE: Towards the Impact of Frame Rate and Bit Rate on Cloud Gaming. In2018 Tenth International Conference on Quality of Multimedia Experience (QoMEX). 1–6. https://doi.org/10.1109/QoMEX. 2018.8463416

    2018 doi

  34. [42]

    Fan Zhang and David R Bull. 2015. A perception-based hybrid model for video quality assessment.IEEE Transactions on Circuits and Systems for Video Technology26, 6 (2015), 1017–1028

    2015

  35. [43]

    Yuankang Zhao, Qinghua Wu, Gerui Lv, Furong Yang, Jiuhai Zhang, Feng Peng, Yanmei Liu, Zhenyu Li, Ying Chen, Hongyu Guo, et al. 2024. JitBright: towards Low-Latency Mobile Cloud Rendering through Jitter Buffer Optimization. InProceedings of the 34th edition of the Workshop on ...

    2024

  36. [44]

    Yuhan Zhou, Tingfeng Wang, Liying Wang, Nian Wen, Rui Han, Jing Wang, Chenglei Wu, Jiafeng Chen, Longwei Jiang, Shibo Wang, Hong- hao Liu, and Chenren Xu. 2024. AUGUR: Practical Mobile Multipath Transport Service for Low Tail Latency in Real-Time Streaming. In21st USENIX Sympo...

    2024

Pith tools

Reviewed July 14, 2026 · model on record in the stance chip above.