Pith. sign in

REVIEW 4 major objections 5 minor 42 references

MD2G-Cast: Relay-Coordinated Multicast for Scalable Volumetric Streaming over MoQ

T0 review · 4 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read A relay control layer that groups overlapping viewers and selectively adds enhancement keeps 100-user volumetric streams under 40 ms delivery intervals.

desk verdict Solid systems paper with a careful matched ablation, but the central 'relay coordination and selective admission' claim is broader than the experiments can support. read the letter →

arxiv 2608.10020 v1 pith:ECOPDQF3 submitted 2026-08-09 eess.IV cs.MM

Ruonan Chai , Yisu Wang , Zili Meng , Dirk Kutscher This is my paper · ORCID
classification eess.IVcs.MM
keywords volumetricstreamingmulticastrelaycoordinationMediaoverQUICpointcloudvideoviewpointoverlapreinforcementlearningMoQ
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 sets out to show that multi-user volumetric streaming can be made scalable not by improving per-user rate adaptation, but by having relays coordinate what content is shared across users. The proposed system, MD2G-Cast, splits each point-cloud frame into a Base set and an Enhanced set, groups users whose fields of view overlap into multicast subscriptions, and only admits Enhanced delivery when a receiver's capability and bandwidth justify it. On a trace-driven emulation testbed with real MoQ processes, the paper reports that receiver-side P99 delivery intervals stay below 40 ms for 20 and 100 users across all seven access profiles, and that aggregate link load drops by about 27% relative to a clustering baseline at 100 users. A matched ablation indicates that the benefit comes from the relay coordination structure and selective admission, not from the particular reinforcement-learning optimizer. If correct, this reframes the design problem: transport and media stay fixed, and the control layer carries the scalability gain.

What carries the argument

The carrying mechanism is the relay control abstraction with feasibility enforced before optimization. Each second, the relay computes for every active user a compatibility feature $G(u;C)=\gamma_1 \bar{O}(u;C)+\gamma_2 D(u)+\gamma_3 B(u)$ over feasible candidate groups, where $\bar{O}$ is the directional FoV overlap, $D$ the normalized device capability, and $B$ the normalized bandwidth; candidate construction rejects invalid assignments before any policy runs, so a learned Proximal Policy Optimization (PPO) controller and a deterministic rule controller share the same interface. Decisions are scored by a clipped system utility $U_i = \mathrm{clip}_{[0,1]}(0.25 R_o + 0.60 R_q - 0.15 R_b)$, with $R_o$ grouping efficiency, $R_q$ delivered quality, and $R_b$ bandwidth inefficiency. A larger teacher model supervises a compact student model through KL matching on the Enhanced-admission branch, keeping per-decision inference around 0.04 ms. The same substrate, traces, and measurement procedure apply to MD2G-Cast, Heuristic, and Clustering, while Rolling retains independent HTTP/DASH delivery as a system-level reference.

What would settle it

Run the same MD2G-Cast controller at 100 heterogeneous users on a physical or independently built testbed with real wide-area links; if the receiver-side P99 delivery interval exceeds 40 ms or the link-load reduction over Clustering falls well below 27%, the quantitative claims fail. A second check: if a variant that only groups Base delivery and never admits Enhanced matches full MD2G-Cast utility, then selective Enhanced admission is not the active ingredient.

Watch

Extended reading notes

Core claim

The central claim is that a relay-side application-aware control layer, rather than per-client adaptation or a particular policy optimizer, is what makes multicast volumetric delivery scale. MD2G-Cast constructs a feasibility-preserving decision interface at MoQ relays: every control interval, the relay builds candidate multicast groups from directional FoV overlap, normalized device capability, and bandwidth adequacy; the policy then chooses a grouping state and a binary Enhanced-admission decision for each receiver. The media itself is prepared once as complementary Base and Enhanced V-PCC streams, so runtime control never re-encodes content. The paper reports that at 20 and 100 users the system keeps the P99 inter-arrival interval of useful media objects below 40 ms across all seven evaluated access profiles, achieves the highest or tied-highest mean system utility under homogeneous access and the highest under heterogeneous access, and reduces aggregate link load by about 27% relative to Clustering at 100 users. The matched relay-control ablation shows random feasible actions lower utility by increasing bandwidth inefficiency, while a deterministic rule controller matches PPO, supporting the interpretation that structured relay coordination and selective Enhanced admission constitute the contribution.

Load-bearing premise

The load-bearing premise is that the trace-driven emulation with real MoQ processes faithfully reproduces the QUIC and MoQ behaviors that determine delivery intervals and system utility in real wide-area deployments, so that the reported P99 and link-load numbers transfer beyond the testbed.

Editorial extensions

If this is right

  • If the central claim is correct, operators can deploy MD2G-Cast without committing to a specific learned policy; a deterministic controller built on the same feasibility interface captures most of the benefit.
  • Because media preparation is separated from adaptive control, the Base and Enhanced streams are encoded once and published as ordinary MoQ objects, so the control layer works on top of an unmodified MoQ transport.
  • At scales up to 100 users, relay-side grouping keeps the delivery tail below 40 ms across heterogeneous access profiles, while per-client HTTP/DASH delivery saturates at the 500 ms reporting cap.
  • Because the 27% link-load reduction comes from replacing per-user redundancy with shared upstream delivery, the savings should grow as the number of users with overlapping fields of view increases.
  • The utility advantage is strongest under heterogeneous access, suggesting the coordination layer matters most exactly when receivers have mixed capabilities and bandwidths.

Reading between the lines

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

  • A direct test of the paper's design-contribution claim would be to replace the PPO policy with a simple threshold policy on the same compatibility features and compare utility in the same testbed; the paper's ablation predicts the two will stay close.
  • The fixed Base ratio $\rho=0.6$ is outside the optimizer, so varying it across scenes and access mixes could change the reuse-versus-refinement trade-off; the reported gains are established for this single partition.
  • The same group-and-admit structure could apply to other layered media with shared and optional dependencies, such as multi-view 360-degree video or neural-enhanced point-cloud streams, without changing the codec.
  • Because the evaluation keeps translational head positions aligned and varies only orientation, full 6DoF movement could reduce overlap and shrink the multicast benefit; the magnitude of the gain under free movement remains untested.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper presents MD2G-Cast, a relay-coordinated multicast framework over Media over QUIC (MoQ) for scalable multi-user volumetric streaming. The system partitions point-cloud frames into complementary Base and Enhanced point sets, and a relay-side control layer uses FoV overlap, receiver capability, and bandwidth headroom to form multicast groups for Base delivery and selectively admit Enhanced delivery. The grouping and admission problem is formulated as a sequential decision process and realized with PPO, with a distilled compact student model for the Enhanced-admission branch. The authors implement the system with real MoQ processes in a trace-driven Mininet testbed and compare it against three baselines: Heuristic and Clustering (same MoQ substrate) and Rolling (HTTP/DASH independent-delivery reference). They report that MD2G-Cast keeps the receiver-side P99 delivery interval below 40 ms at 20 and 100 users across seven access profiles, reduces aggregate link load by about 27% relative to Clustering at 100 users, attains the highest or tied-highest mean system utility under homogeneous access and the highest mean utility under heterogeneous access, and a matched ablation shows that random feasible actions reduce utility while deterministic control remains competitive with PPO.

Significance. The paper addresses an important scaling problem in volumetric streaming: avoiding repeated transmission of shared content across users with overlapping views. If the results hold, the main contribution is an application-aware relay control layer that separates media preparation from runtime control and demonstrates that relay coordination and selective Enhanced admission, rather than the particular RL optimizer, drive the benefit. The paper is careful in its experimental design: Heuristic and Clustering share the MoQ substrate and measurement procedure, Rolling is explicitly labeled as a system-level independent-delivery reference and not a transport-isolation baseline, and the matched ablation varies only the relay decision mechanism. The artifact is promised to be public, which supports reproducibility. These strengths make the paper a credible step toward practical multi-user volumetric streaming, provided the central claims are not overstated.

major comments (4)
  1. [Abstract and Section 5.2, Table 4] The matched ablation does not isolate relay coordination or selective Enhanced admission as the active design ingredient. All four ablation variants (Full MD2G-Cast, Fixed grouping, Random feasible action, Rule-only controller) still perform relay-coordinated grouping and keep the same admission interface; only the policy realization changes. There is no same-substrate arm without relay coordination (e.g., independent MoQ delivery without group sharing) and no same-substrate ablation that toggles Enhanced admission between always-on and always-off. Therefore the claim that 'relay coordination and selective Enhanced admission' are the central design contribution is not causally supported by the presented experiments. The experiments support the weaker statements that PPO is not essential and that the tested controller outperforms two coordinated baselines on an author-defined metric.
  2. [Table 4] The utility differences reported in Table 4 are small relative to the dispersion across the three matched trials: Full MD2G-Cast is 0.716±0.004, Fixed grouping is 0.712±0.003, Random feasible action is 0.703±0.014, and Rule-only is 0.718±0.006. With n=3 and overlapping standard deviations, the differences (0.004 to 0.013) may be within noise, and no statistical significance test or confidence interval is provided. This weakens the claim that relay-control structure itself, rather than randomness, is the source of the reported benefit.
  3. [Section 3.2, Eq. (9) and Section 4.2, Eq. (11)] The system utility defined in Eq. (9) is used both as the training reward and as the primary evaluation metric. While this is not inherently circular, the ranking of strategies depends on the chosen weights (0.25, 0.60, -0.15) and penalty coefficients (0.2, 0.3) in Eq. (11). The paper does not report a sensitivity analysis with respect to these parameters, so it is unclear whether the 'highest utility' conclusions would remain under alternative reasonable choices. Since these parameters are free, the robustness of the central quantitative comparisons should be demonstrated.
  4. [Section 5, Testbed and topology] The headline numerical claims—P99 below 40 ms and the 27% link-load reduction—are derived entirely from a Mininet emulation with replayed bandwidth traces and real MoQ processes. The paper does not validate the emulation against a real wide-area deployment or analyze how Mininet's handling of QUIC/MoQ timers, loss, and congestion affects the delivery-interval metric. The paper should either temper the abstract's absolute-number emphasis or provide a validation/limitation discussion, because these numbers may not transfer quantitatively to real networks.
minor comments (5)
  1. [Figure 2] The caption and the figure text contain a typo: 'Video Sourse Server' should be 'Video Source Server'.
  2. [Section 5.1, Figure 5] The sentence 'All users consume the Red and Blacksequence' has a missing space; it should be 'Red and Black sequence'.
  3. [Figure 4] The abbreviation 'B+E' in the figure is not defined; the text uses 'Base-plus-Enhanced'. Please define it once for clarity.
  4. [Section 2.3, Eq. (5)] The values of the compatibility feature weights γ1, γ2, and γ3 used in the experiments are not reported. Specify these values in Section 5 or in a reproducibility section.
  5. [Section 5, Figure 5] The description in Section 5.1 says 'the lower bar segment ends at P95, the full bar height marks P99, and the blue curve shows P50,' but the figure shows a blue curve as well. Please clarify the legend and the visual encoding so that P50 is unambiguously identified.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the utility objective is author-defined but applied uniformly, and the matched ablation varies the policy over a fixed relay interface; no claim reduces to its inputs by construction.

full rationale

The paper's derivation chain is empirical rather than definitional. Equation (9) defines a system utility, and PPO is trained to optimize the same scalar objective, but this is not a circular prediction because the paper compares PPO against Heuristic, Clustering, Rule-only, Random-feasible, Fixed-grouping, and Rolling on the same utility; high utility is therefore not guaranteed by construction. The matched relay-control ablation (Section 5.2, Table 4) varies only the decision mechanism over a fixed feasible action space: Rule-only (0.718±0.006) is comparable to Full MD2G-Cast (0.716±0.004), which supports the 'optimizer is not essential' conclusion as an empirical finding. The claim that relay coordination and selective Enhanced admission are the central contribution is an interpretation of the design; the absence of a same-substrate non-coordinated arm limits how strongly the experiments isolate coordination, but that is an external-validity limitation, not a circular reduction. The paper's self-citation ([2]) appears only in related work and is not load-bearing; no uniqueness theorem is imported from the authors' previous work, and no ansatz is smuggled in via citation. The limitations stated in Section 7 (single V-PCC sequence, objective proxy, no subjective quality assessment) are acknowledged scope restrictions rather than hidden equivalences.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The central claims rest on several hand-chosen constants (Base ratio ρ=0.6, utility weights, penalty coefficients, control interval) and on the assumptions that MoQ relays behave as specified, V-PCC stitching works as expected, and the Mininet trace-driven testbed represents real network conditions. The paper is transparent about the proxy nature of the utility objective and the limited scope.

free parameters (6)
  • Base sampling ratio ρ = 0.6
    Fixed at 0.6 across all access profiles and not adapted online (Section 2.2, Eq 1). Determines how much content is shared as Base versus optional Enhanced, affecting all utility and link-load results.
  • Compatibility feature weights γ1, γ2, γ3 = not reported
    Weights in Eq (5) summing to 1 combine FoV overlap, device capability, and bandwidth into the grouping feature. Values are not given in the paper, so the exact feature is not reproducible.
  • Utility objective weights (0.25, 0.60, -0.15) = 0.25 / 0.60 / -0.15
    Eq (9) defines the system utility used for evaluation and ablation. These hand-chosen coefficients determine what counts as a good outcome.
  • Quality penalty coefficients (0.2, 0.3) = 0.2 / 0.3
    Eq (11) applies delivery-interval and stall penalties inside the quality term. Values are chosen without external justification.
  • Relay control interval = 1.0 s
    Relay decisions are updated every 1.0 s (Section 5.1, Section 3.2). The interval is a design choice affecting responsiveness and stability.
  • KL distillation coefficient λ_d = not reported
    Eq (10) uses λ_d to weight teacher-guided KL matching for the Enhanced-admission branch. The value is not reported.
assumptions (6)
  • domain assumption MoQ provides object-oriented publish/subscribe and relay-assisted delivery over QUIC.
    The design builds on MoQ's relay capabilities as described in [8,9]; if those capabilities differ, the relay control layer may not realize the claimed sharing.
  • domain assumption V-PCC independently encodes the complementary Base and Enhanced point sets as separate atlas streams with lossless index stitching.
    Section 2.2 assumes V-PCC [26] supports the described split-and-stitch reconstruction; the quality gain (Table 3) is measured on this coding, not on a combined bitstream.
  • domain assumption The 8i Red and Black sequence and CWI head-motion traces represent realistic 6DoF viewing dynamics.
    Evaluation uses one content sequence and orientation-only head traces with aligned positions (Section 5). The authors acknowledge this limits generality.
  • domain assumption Mininet with replayed bandwidth traces and real MoQ processes faithfully reproduces the network conditions that determine delivery intervals.
    All absolute numbers (P99<40ms, link load) come from this testbed (Section 5, Testbed and topology).
  • ad hoc to paper The utility objective in Eq (9) is a valid proxy for system performance.
    The paper states it is an objective proxy and does not replace subjective quality assessment (Section 7). The central 'highest utility' claim depends on this definition.
  • domain assumption Directional FoV overlap as patch-set intersection (Eq 2) captures reusable demand among receivers.
    The grouping relies on patch-level overlap; if patch visibility does not reflect actual content reuse, grouping efficiency may differ.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MD2G-Cast: Relay-Coordinated Multicast for Scalable Volumetric Streaming over MoQ." pith.science (2026). https://pith.science/paper/ECOPDQF3

@misc{pith2026260810020,
  author       = {Pith},
  title        = {Pith review of: MD2G-Cast: Relay-Coordinated Multicast for Scalable Volumetric Streaming over MoQ},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ECOPDQF3}},
  note         = {Machine review of arXiv:2608.10020}
}
abstract

Volumetric streaming remains difficult to scale because receivers with overlapping fields of view are often served independently, causing repeated transmission of shared content. We present MD2G-Cast, a relay-coordinated multicast framework over Media over QUIC with an application-aware control layer for scalable multi-user volumetric delivery. MD2G-Cast jointly uses viewing overlap, receiver capability, and bandwidth conditions to form reusable multicast groups, share common Base content, and selectively admit Enhanced delivery. We formulate grouping and Enhanced admission as a sequential control problem, realize it with Proximal Policy Optimization (PPO), and train a compact relay model with teacher guidance for Enhanced admission. We implement MD2G-Cast with real MoQ processes and evaluate it with real access and 6DoF viewing traces for up to 100 users. At 20 and 100 users, MD2G-Cast keeps the receiver-side $P_{99}$ delivery interval below 40 ms across all seven access profiles, while Rolling reaches the 500 ms reporting cap in most cases. Across the evaluated user scales, MD2G-Cast achieves the highest or tied-highest mean system utility under homogeneous access and the highest mean utility under heterogeneous access, while reducing aggregate link load by about 27% relative to Clustering at 100 users. A matched relay-control ablation separates the control structure from its optimizer, showing that random feasible actions reduce utility while deterministic control remains competitive with PPO. Together, the results support relay coordination and selective Enhanced admission, rather than a particular policy optimizer, as the central design contribution.

Figures

Figures reproduced from arXiv: 2608.10020 by the authors.

Figure 1
Figure 1. Pipeline contrast between DASH and MoQ for multi [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. System architecture of MD2G-Cast. MoQ relays running the network transmission module, and User Terminals (UTs). 2.2 Video Source Server (VSS) For each source frame, let X𝑓 = {𝑥1, . . . , 𝑥𝑛 } denote its ordered point set. For all reported experiments, the Base ratio is fixed at 𝜌 = 0.6 across access profiles and is not adapted online. With 𝑚 = round(𝜌𝑛), a deterministic index-stride sampler selects a Base index set … view at source ↗
Figure 4
Figure 4. Visual comparison of Base-only and Base-plus [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: Receiver-side delivery interval at 20 and 100 users. The lower bar segment ends at [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Enhanced exposure across user scales under homo [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 8
Figure 8. Figure 8: Mean system utility across homogeneous and het [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 33 canonical work pages

  1. [1]

    Chenghong Bian, Yulin Shao, and Deniz Gündüz. 2024. Wireless point cloud trans- mission. In2024 IEEE 25th International Workshop on Signal Processing Advances in Wireless Communications (SPA WC). IEEE, 851–855

    work page 2024

  2. [2]

    Ruonan Chai, Yixiang Zhu, Xinjiao Li, Jiawei Li, Zili Meng, and Dirk Kutscher

    work page

  3. [3]

    Das, Aruna Balasubramanian, and Dimitris Samaras

    Mallesham Dasari, Kumara Kahatapitiya, Samir R. Das, Aruna Balasubramanian, and Dimitris Samaras. 2022. Swift: Adaptive Video Streaming with Layered Neural Codecs. In19th USENIX Symposium on Networked Systems Design and Implementation. USENIX Association, 103–118

    work page 2022

  4. [4]

    Nianzhen Gao, Yiming Wang, Xinhai Hua, Ting Bi, and Tao Jiang. 2024. 360- Degree Video Multicast Scheduling Optimization in Orthogonal and Non- Orthogonal Scenarios.IEEE Internet of Things Journal(2024)

    work page 2024

  5. [5]

    Jianping Gou, Baosheng Yu, Stephen J Maybank, and Dacheng Tao. 2021. Knowl- edge distillation: A survey.International Journal of Computer Vision129, 6 (2021), 1789–1819

    2021

  6. [6]

    Yongjie Guan, Xueyu Hou, Nan Wu, Bo Han, and Tao Han. 2023. MetaStream: Live Volumetric Content Capture, Creation, Delivery, and Rendering in Real Time. InProceedings of the 29th Annual International Conference on Mobile Computing and Networking. 425–439. doi:10.1145/3570361.3592530

    arXiv 2023

  7. [7]

    Zafer Gurel, Tugce Erkilic Civelek, and Ali C. Begen. 2024. This Is the Way: Prioritization in Media-over-QUIC Transport. InProceedings of the ACM Mile- High Video Conference. doi:10.1145/3638036.3640280

    work page arXiv 2024

  8. [8]

    Zafer Gurel, Tugce Erkilic Civelek, Deniz Ugur, Yigit K Erinc, and Ali C Begen

    work page

Show all 42 references
  1. [9]

    Zafer Gurel, Tugce Erkilic Civelek, Atakan Bodur, Senem Bilgin, Deniz Yeniceri, and Ali C. Begen. 2023. Media over QUIC: Initial Testing, Findings and Results. InProceedings of the 14th ACM Multimedia Systems Conference(Vancouver, BC, Canada)(MMSys ’23). Association for Comput...

    2023

  2. [10]

    Bo Han, Yu Liu, and Feng Qian. 2020. ViVo: Visibility-Aware Mobile Volumetric Video Streaming. InProceedings of the 26th Annual International Conference on Mobile Computing and Networking. 1–13. doi:10.1145/3372224.3380888

    2020

  3. [11]

    Yakun Huang, Boyuan Bai, Yuanwei Zhu, Xiuquan Qiao, Xiang Su, Lei Yang, and Ping Zhang. 2023. ISCom: Interest-aware Semantic Communication Scheme for Point Cloud Video Streaming on Metaverse XR Devices.IEEE Journal on Selected Areas in Communications(2023)

    2023

  4. [12]

    Han-Rong Lai, Ru-Jun Wang, Chih-Hang Wang, De-Nian Yang, Wen-Tsuen Chen, and Jang-Ping Sheu. 2024. Socially-Aware Tile-Based Point Cloud Multicast with Registration. InICC 2024-IEEE International Conference on Communications. IEEE, 527–532

    2024

  5. [13]

    Kyungjin Lee, Juheon Yi, Youngki Lee, Sunghyun Choi, and Young Min Kim

  6. [14]

    Jie Li, Huiyu Wang, Zhi Liu, Pengyuan Zhou, Xianfu Chen, Qiyue Li, and Richang Hong. 2023. Toward Optimal Real-Time Volumetric Video Streaming: A Rolling Optimization and Deep Reinforcement Learning Based Approach.IEEE Transac- tions on Circuits and Systems for Video Technolog...

    2023

  7. [15]

    Jie Li, Cong Zhang, Zhi Liu, Richang Hong, and Han Hu. 2022. Optimal volu- metric video streaming with hybrid saliency based tiling.IEEE Transactions on Multimedia25 (2022), 2939–2953

    2022

  8. [16]

    Zhicheng Liang, Junhua Liu, Mallesham Dasari, and Fangxin Wang. 2024. Fumos: Neural compression and progressive refinement for continuous point cloud video streaming.IEEE Transactions on Visualization and Computer Graphics30, 5 (2024), 2849–2859

    2024

  9. [17]

    Hai Lin and Xianfu Chen. 2022. Transformer-driven multi-agent deep reinforce- ment learning based point cloud video transmissions. InProceedings of the 1st Workshop on Digital Twin & Edge AI for Industrial IoT. 25–30

    2022

  10. [18]

    Hai Lin, Bo Zhang, Yangjie Cao, Zhi Liu, and Xianfu Chen. 2021. A deep rein- forcement learning approach for point cloud video transmissions. In2021 IEEE 94th Vehicular Technology Conference (VTC2021-Fall). IEEE, 1–5

    2021

  11. [19]

    Junhua Liu, Boxiang Zhu, Fangxin Wang, Yili Jin, Wenyi Zhang, Zihan Xu, and Shuguang Cui. 2023. CaV3: Cache-assisted Viewport Adaptive Volumetric Video Streaming. InIEEE VR

    2023

  12. [20]

    Yu Liu, Bo Han, Feng Qian, Arvind Narayanan, and Zhi-Li Zhang. 2022. Vues: Practical Mobile Volumetric Video Streaming Through Multiview Transcoding. InProceedings of the 28th Annual International Conference on Mobile Computing and Networking. 514–527. doi:10.1145/3495243.3517027

    2022

  13. [21]

    Yu Liu, Puqi Zhou, Zejun Zhang, Anlan Zhang, Bo Han, Zhenhua Li, and Feng Qian. 2024. Muv2: scaling up multi-user mobile volumetric video streaming via content hybridization and sharing. InProceedings of the 30th annual International Conference on mobile computing and Networki...

    2024

  14. [22]

    Moatasim Mahmoud, Stamatia Rizou, Andreas S Panayides, Nikolaos V Kantartzis, George K Karagiannidis, Pavlos I Lazaridis, and Zaharias D Zaharis. 2023. A sur- vey on optimizing mobile delivery of 360°videos: Edge caching and multicasting. IEEE Access(2023)

    2023

  15. [23]

    Rui Meireles, António Rodrigues, Andrei Stanciu, Ana Aguiar, and Peter Steenkiste. 2020. Exploring Wi-Fi Network Diversity for Vehicle-To-Infrastructure Communication. In2020 IEEE Vehicular Networking Conference (VNC). 1–8. doi:10.1109/VNC51378.2020.9318407

    2020

  16. [24]

    Cristina Perfecto, Mohammed S Elbamby, Javier Del Ser, and Mehdi Bennis

  17. [25]

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov

  18. [26]

    Chou, Robert A

    Sebastian Schwarz, Marius Preda, Vittorio Baroncini, Madhukar Budagavi, Pablo Cesar, Philip A. Chou, Robert A. Cohen, Maja Krivokuća, Sébastien Lasserre, Zhu Li, Joan Llach, Khaled Mammou, Rufael Mekuria, Ohji Nakagami, Ernes- tasia Siahaan, Ali Tabatabai, Alexis M. Tourapis, ...

  19. [27]

    Shishir Subramanyam, Irene Viola, Alan Hanjalic, and Pablo Cesar. 2020. User centered adaptive streaming of dynamic point clouds with low complexity tiling. InProceedings of the 28th ACM international conference on multimedia. 3669–3677

    2020

  20. [28]

    Taming the latency in multi-user VR 360°: A QoE-aware deep learning- aided multicast framework.IEEE Transactions on Communications68, 4 (2020), 2491–2508

    2020

  21. [29]

    Mengsi Yeh, Chih-Hang Wang, De-Nian Yang, Ji-Tang Lee, and Wanjiun Liao

  22. [30]

    2022.{YuZu}:{Neural- Enhanced} volumetric video streaming

    Anlan Zhang, Chendong Wang, Bo Han, and Feng Qian. 2022.{YuZu}:{Neural- Enhanced} volumetric video streaming. In19th USENIX Symposium on Networked Systems Design and Implementation (NSDI 22). 137–154

    2022

  23. [31]

    Ding Zhang, Bo Han, Parth H Pathak, and Haoliang Wang. 2021. Innovating Multi-user Volumetric Video Streaming through Cross-layer Design. InACM HotNets

    2021

  24. [32]

    Ding Zhang, Puqi Zhou, Bo Han, and Parth Pathak. 2022. M5: Facilitating multi- user volumetric content delivery with multi-lobe multicast over mmWave. In Proceedings of the 20th ACM Conference on Embedded Networked Sensor Systems. 31–46

    2022

  25. [33]

    Zhilong Zhang, Minyin Zeng, Mingzhe Chen, Danpu Liu, Walid Saad, Shuguang Cui, and H Vincent Poor. 2021. Joint user grouping, version selection, and band- width allocation for live video multicasting.IEEE Transactions on Communications 70, 1 (2021), 350–365

    2021

  26. [34]

    Soushi Ueno, Takuya Fujihashi, Toshiaki Koike-Akino, and Takashi Watanabe

  27. [37]

    IEEE Transactions on Mobile Computing21, 8 (2020), 2909–2921

    Mobile proxy caching for multi-view 3D videos with adaptive view selection. IEEE Transactions on Mobile Computing21, 8 (2020), 2909–2921

    2020

  28. [42]

    Yuanwei Zhu, Yakun Huang, Xiuquan Qiao, Zhijie Tan, Boyuan Bai, Huadong Ma, and Schahram Dustdar. 2022. A semantic-aware transmission with adaptive control scheme for volumetric video service.IEEE Transactions on Multimedia25 (2022), 7160–7172

    2022

  29. [2017]

    Proximal policy optimization algorithms.arXiv preprint arXiv:1707.06347 (2017)

    2017 arXiv

  30. [2019]

    doi:10.1109/JETCAS.2018.2885981

    Emerging MPEG Standards for Point Cloud Compression.IEEE Journal on Emerging and Selected Topics in Circuits and Systems9, 1 (2019), 133–148. doi:10.1109/JETCAS.2018.2885981

    2019

  31. [2020]

    InProceedings of the 26th Annual International Conference on Mobile Computing and Networking

    GROOT: a real-time streaming system of high-fidelity volumetric videos. InProceedings of the 26th Annual International Conference on Mobile Computing and Networking. 1–14

  32. [2023]

    Point Cloud Soft Multicast for Untethered XR Users.IEEE Transactions on Multimedia25, 8 (2023), 7185–7195

    2023

  33. [2024]

    low-latency dash: a deathmatch testbed

    Media-over-quic transport vs. low-latency dash: a deathmatch testbed. In Proceedings of the 15th ACM Multimedia Systems Conference. 448–452

  34. [2025]

    InProceedings of the 33rd ACM International Conference on Multimedia

    INDS: Incremental Named Data Streaming for Real-Time Point Cloud Video. InProceedings of the 33rd ACM International Conference on Multimedia. 12102–12110

Pith tools

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