Pith. sign in

REVIEW 3 major objections 5 minor 54 references

AdaDPCC: Adaptive Rate Control and Rate-Distortion-Complexity Optimization for Dynamic Point Cloud Compression

T0 review · 3 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read AdaDPCC claims a single slimmable dynamic point cloud codec that adaptively routes each frame to meet target bitrates, cutting bitrate by 5.81% and coding time by up to 44.6% against state-of-the-art baselines.

desk verdict Slimmable multi-route DPCC with a clever rate-control module, but the headline SOTA gain over patchDPCC rests on the original paper's numbers and is not verifiable from this submission. read the letter →

arxiv 2508.20741 v1 pith:VQFPGQZ2 submitted 2025-08-28 cs.MM

Chenhao Zhang , Wei Gao This is my paper
classification cs.MM
keywords dynamicpointcloudcompressionrate-distortion-complexityoptimizationslimmablenetworkratecontrolinter-framepredictionsparseconvolution
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 tries to show that dynamic point cloud compression can be made both rate-adaptive and complexity-adaptive in one trained model, without retraining per bitrate. It introduces a slimmable autoencoder with several nested coding routes, so a frame can be encoded through a lighter or heavier subnetwork depending on the bitrate budget and compute available. To handle temporal redundancy, it replaces expensive KNN-based motion search with a coarse-to-fine sparse-convolution motion estimation and compensation scheme that first captures block-level motion, then deconstructs geometric context for voxel-level refinement. A lightweight rate control module estimates the bitrate of each route and selects the route that lands closest to the target using a sliding-window bit allocation. The reported results show an average 5.81% BD-Rate reduction and 0.42 dB BD-PSNR gain over the previous best learning-based method, an average bitrate error of 0.40%, and up to 44.6% lower coding time than the KNN-based baseline.

What carries the argument

The load-bearing mechanism is the slimmable convolution operator, which prunes channel dimensions to create K overlapping sub-networks (routes) inside one encoder-decoder supernet. Each route takes a partial share of the computational load and yields a distinct rate-distortion trade-off, all jointly trained with a cumulative loss over decreasing Lagrange multipliers. Alongside this sits the coarse-to-fine inter-frame prediction module: it computes a coarse motion embedding with stride-2 sparse convolutions, then a fine embedding, and performs motion compensation by quantizing the motion vector to a 2x-downsampled grid, warping coordinates, and deconstructing geometric features from the refer

What would settle it

Train the same architecture but with each route trained as an independently tuned model (or with different lambda schedules) on the Owlii training set, then compare BD-Rate, BD-PSNR, and bitrate error on the 8iVFB sequences; if an independent or alternative schedule beats the proposed joint schedule on RD or rate control accuracy, the 'optimal RD at each complexity' claim fails. A second falsifier is to run the rate control module on an out-of-distribution dynamic point cloud sequence (e.g., LiDAR-like sparse frames) and check whether the average bitrate error stays near 0.40%; a large diverge

Watch

Extended reading notes

Core claim

The central claim is that rate-distortion-complexity optimization for dynamic point clouds is achievable within a single dynamic coding network. The paper's framework, AdaDPCC, trains a supernet with K slimmable coding routes jointly, each route using a subset of channels to produce a distinct RD trade-off while sharing parameters. The claim is that this joint training gives near-optimal RD performance at every complexity level, so the same model can cover variable bitrates. On the temporal side, the paper claims its coarse-to-fine inter-frame prediction - using coarse block-wise motion vectors, geometric deconstruction of the reference latent, and fine voxel-wise motion refinement - outperf

Load-bearing premise

The claim that jointly training subnets in one network achieves optimal rate-distortion performance at every complexity level rests on a heuristic schedule of decreasing Lagrange multipliers, not on a proof; if this schedule yields suboptimal routes, the reported RD trade-offs and route-selection behavior may not generalize to other sequences or training settings.

Editorial extensions

If this is right

  • A single trained dynamic point cloud codec can operate at several bitrate and complexity operating points, replacing per-target retraining and making deployment on bandwidth- or compute-constrained devices more practical.
  • Replacing KNN-based motion search with sparse-convolution coarse-to-fine prediction materially reduces inter-frame coding latency, which matters for real-time AR/VR and autonomous-driving pipelines.
  • Frame-level rate control with average bitrate error around 0.4% enables predictable bitrate compliance in streaming, where hitting a negotiated bandwidth is as important as reconstruction quality.
  • The same slimmable-route design could be transferred to other learned 3D compression tasks that need adaptive complexity, such as static point cloud geometry coding or LiDAR compression.
  • The ablations attribute roughly 3.4% BD-Rate reduction to coarse-to-fine inter prediction and 2.9% to the hyperprior context, indicating both temporal modeling and spatial context conditioning contribute to the reported gains.

Reading between the lines

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

  • The rate estimation module could plausibly be reused at a coarser granularity, such as scene-level or group-of-frames bitrate ladders, since it estimates each route's cost from frame content with negligible latency (about 0.01 s); the paper does not test that extension.
  • The paper selects routes based only on estimated bitrate, not estimated distortion; adding a per-route distortion estimator would be a natural testable extension that could make the allocation choice better match perceptual or MSE quality under a fixed budget.
  • The coarse-to-fine anchor-based motion compensation is a generic idea for sparse 3D data and might generalize beyond compression to point cloud flow or temporal interpolation; the paper reports only its use inside the codec.
  • Because the joint training strategy is an empirical heuristic, its claim of 'optimal RD performance at each complexity' is not proven; a fair ablation would train each route independently with the same lambda list and compare RD curves, which the paper does not include.
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 / 5 minor

Summary. The paper proposes AdaDPCC, a learned dynamic point cloud codec with slimmable multi-route encoder/decoder, coarse-to-fine motion estimation and compensation, conditional inter-frame coding, and a content-adaptive rate control module. The framework is designed to offer variable bitrate and computational complexity within a single model. On the MPEG 8iVFB dataset, the authors report an average BD-Rate reduction of 5.81% and a BD-PSNR improvement of 0.423 dB over patchDPCC, an average bitrate error of 0.40% against target rates from 0.15 to 0.29 bpp, and up to 44.6% coding time reduction compared to D-DPCC. The code is claimed to be available at an OpenI repository.

Significance. If fully substantiated, the multi-route RDCO design and the low-latency rate control module would be useful contributions to learned dynamic point cloud compression, and the release of code is a clear strength. The measured gains against the rerun baselines (D-DPCC, PCGCv2, V-PCC) are credible and internally consistent. However, the headline advantage over patchDPCC rests on externally reported numbers that cannot be independently verified, and the optimality claim for the joint route training is an unproven heuristic. These issues are central to the paper's main claims and need to be addressed before the comparison can be accepted.

major comments (3)
  1. [Experiments, 'Benchmark models'; Table 1] The headline 5.81% BD-Rate / 0.423 dB BD-PSNR advantage over patchDPCC is taken directly from the original paper because patchDPCC's source code is unavailable. The statement that 'identical testing conditions' are maintained is not verifiable. Differences in I-frame coding (the paper uses PCGCv2 with the same lambda), 10-bit quantization, RD sample points, or D1-PSNR computation could materially change BD-Rate values. Please either obtain and rerun patchDPCC under the authors' pipeline, provide independent confirmation, or clearly separate 'rerun baselines' from 'reported baseline' and temper the state-of-the-art claim.
  2. [Methodology, 'Dynamic Inference Routes'; Algorithm 1] The paper asserts that joint routes training 'achieves optimal RD performance at each level of complexity.' Algorithm 1 is a sequential decreasing-lambda heuristic with no proof and no comparison to alternatives such as independent route training or sandwich-rule sampling. Since the RDCO contribution and the route-selection behavior of the rate control module depend on this assertion, please add an ablation or revise the claim to 'near-optimal' or 'empirically effective.'
  3. [Methodology, 'Rate Control Module'] The Rate Estimation module is central to the 0.40% bitrate error claim, but the paper gives no architectural or training details: no input features, layer sizes, loss function, or per-route estimation accuracy. Figure 1 only shows generic FC/Conv blocks. This prevents reproduction and validation that the estimator is content-adaptive. Please specify this module explicitly or remove the equation-free description.
minor comments (5)
  1. [Algorithm 1] The Loss accumulator is not initialized to zero at the start of each outer loop; as written, losses accumulate across iterations. Add 'Loss ← 0' after line 4.
  2. [Eq. (12)] The convolution kernel is written as U(-1/2, -1/2); this should presumably be U(-1/2, 1/2) as in the accompanying text.
  3. [Implementation Details] The choices K=4 and lambda=[3,7,10,20] are said to be finalized 'through numerous trials'; a sensitivity study or a brief justification of these values would improve reproducibility.
  4. [Experimental Results] No error bars or multiple-seed results are reported. For the small BD-Rate margins over a non-rerun baseline, a single run is difficult to interpret; at minimum, report variance across training runs or state that this is standard practice.
  5. [Abstract / URL] The OpenI URL in the abstract contains a space ('Ada DPCC'); fix the link formatting.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; results are empirically measured and benchmarked against external methods.

full rationale

The paper's central claims are empirical: RD performance, coding time, and rate-control error are all measured end-to-end on a held-out dataset (8iVFB) after training on Owlii. The rate-control module's bitrate error is a direct measurement of prediction accuracy against actual coded bitrates, not a fitted parameter renamed as a prediction. The BD-Rate/BD-PSNR comparison against patchDPCC uses numbers taken from the original patchDPCC paper because its code is unavailable; this is a comparability/verifiability concern, not circularity, since the baseline is external and not derived from the present method's assumptions. The joint training strategy's claim of 'optimal RD performance at each level of complexity' is an unsupported heuristic assertion, but it is not a self-referential derivation. Self-citations in related work (e.g., Zhang and Gao 2024) are background and not load-bearing: no uniqueness theorem or key architectural choice is justified solely by the authors' prior work. The method's equations (e.g., sliding-window bit allocation and route selection) are standard operational definitions, not inputs masquerading as outputs. Therefore, no circular step can be exhibited with the required specificity, and the appropriate score is 0.

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

The paper's core contribution is architectural; its principal free parameters are tuning choices, and the main untested premise is the optimality of the joint training strategy.

free parameters (4)
  • Lagrange multiplier list = [3, 7, 10, 20]
    Chosen through numerous trials to define the RD trade-offs of the four coding routes; directly controls the RD points reported.
  • Number of routes K = 4
    Selected by experimentation; more routes increase flexibility but also training and inference cost.
  • Sliding window length SW = not specified
    Used in Eq. 14 for frame-level bitrate allocation; its value is never stated, yet it affects rate control stability.
  • Post-training lambda_0 schedule = not specified
    Algorithm 1 says post-train by gradually decreasing lambda_0, but the schedule is unspecified.
assumptions (4)
  • ad hoc to paper Joint sequential training of subnets with decreasing lambda yields near-optimal RD at each complexity level
    Invoked in the Dynamic Inference Routes section; this is the central design assumption and is not proven.
  • domain assumption Training on Owlii dataset generalizes to MPEG 8iVFB
    Standard practice in learned PCC; implicitly assumed in Experiments.
  • domain assumption Rate estimation network accurately predicts bitrates for unseen content
    The rate control module relies on this; empirically tested, but the network architecture is only sketched.
  • domain assumption PCGCv2 I-frame encoding with the same lambda provides a fair reference
    The setup encodes I-frames with PCGCv2; any bias in PCGCv2 affects the comparison.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AdaDPCC: Adaptive Rate Control and Rate-Distortion-Complexity Optimization for Dynamic Point Cloud Compression." pith.science (2026). https://pith.science/paper/VQFPGQZ2

@misc{pith2026250820741,
  author       = {Pith},
  title        = {Pith review of: AdaDPCC: Adaptive Rate Control and Rate-Distortion-Complexity Optimization for Dynamic Point Cloud Compression},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VQFPGQZ2}},
  note         = {Machine review of arXiv:2508.20741}
}
read the original abstract

Dynamic point cloud compression (DPCC) is crucial in applications like autonomous driving and AR/VR. Current compression methods face challenges with complexity management and rate control. This paper introduces a novel dynamic coding framework that supports variable bitrate and computational complexities. Our approach includes a slimmable framework with multiple coding routes, allowing for efficient Rate-Distortion-Complexity Optimization (RDCO) within a single model. To address data sparsity in inter-frame prediction, we propose the coarse-to-fine motion estimation and compensation module that deconstructs geometric information while expanding the perceptive field. Additionally, we propose a precise rate control module that content-adaptively navigates point cloud frames through various coding routes to meet target bitrates. The experimental results demonstrate that our approach reduces the average BD-Rate by 5.81% and improves the BD-PSNR by 0.42 dB compared to the state-of-the-art method, while keeping the average bitrate error at 0.40%. Moreover, the average coding time is reduced by up to 44.6% compared to D-DPCC, underscoring its efficiency in real-time and bitrate-constrained DPCC scenarios. Our code is available at https://git.openi.org.cn/OpenPointCloud/Ada_DPCC.

Figures

Figures reproduced from arXiv: 2508.20741 by the authors.

Figure 1
Figure 1. Overview of the proposed method: (a) Rate Control Module. It content-adaptively allocates coding routes for point [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 3
Figure 3. Illustration of the coarse-to-fine inter-frame prediction. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. D1-PSNR RD curves of different methods on 8iVFB dataset. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: Average encoding time comparison between the proposed coarse-to-fine (c2f) inter-frame prediction and the 3DAWI [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Ablation study on RD performance: comparison of [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

54 extracted references · 52 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    A.; and Tian, D

    Ahn, J.; Pang, J.; Lodhi, M. A.; and Tian, D. 2023. DDA-Net: Deep Distribution-Aware Network for Point Cloud Compression. In IEEE International Symposium on Circuits and Systems, 1--5

    work page 2023

  4. [4]

    Akhtar, A.; Li, Z.; and Auwera, G. V. D. 2024. Inter-Frame Compression for Dynamic Point Cloud Geometry Coding. IEEE Trans. Image Process., 33: 584--594

    work page 2024

  5. [5]

    A.; Minnen, D.; Singh, S.; Johnston, N.; Agustsson, E.; Hwang, S

    Ball \' e , J.; Chou, P. A.; Minnen, D.; Singh, S.; Johnston, N.; Agustsson, E.; Hwang, S. J.; and Toderici, G. 2021. Nonlinear Transform Coding. IEEE J. Sel. Top. Signal Process. , 15(2): 339--353

    work page 2021

  6. [6]

    J.; and Johnston, N

    Ball \' e , J.; Minnen, D.; Singh, S.; Hwang, S. J.; and Johnston, N. 2018. Variational image compression with a scale hyperprior. In Int. Conf. Learn. Represent

    work page 2018

  7. [7]

    Cao, K.; Xu, Y.; Lu, Y.; and Wen, Z. 2018. Owlii Dynamic Human Mesh Sequence Dataset . Presented at the 122th MPEG Meeting, ISO/IEC JTC1/SC29/WG11 m42816

    work page 2018

  8. [8]

    A.; Koroteev, M.; and Krivokuca, M

    Chou, P. A.; Koroteev, M.; and Krivokuca, M. 2020. A Volumetric Approach to Point Cloud Compression - Part I: Attribute Compression. IEEE Trans. Image Process., 29: 2203--2216

    work page 2020

Show all 54 references
  1. [9]

    B.; Gwak, J.; and Savarese, S

    Choy, C. B.; Gwak, J.; and Savarese, S. 2019. 4D Spatio-Temporal ConvNets: Minkowski Convolutional Neural Networks. In IEEE Conf. Comput. Vis. Pattern Recog., 3075--3084

    2019

  2. [10]

    Cui, M.; Long, J.; Feng, M.; Li, B.; and Kai, H. 2023. OctFormer: Efficient Octree-Based Transformer for Point Cloud Compression with Local Enhancement. In Williams, B.; Chen, Y.; and Neville, J., eds., AAAI, 470--478

    2023

  3. [11]

    d'Eon, E.; Harrison, B.; Myers, T.; and Chou, P. A. 2017. 8i Voxelized Full Bodies: A Voxelized Point Cloud Dataset . ISO/IEC JTC1/SC29 Joint WG11/WG1 (MPEG/JPEG) input document WG11M40059/WG1M74006

    2017

  4. [12]

    Fan, T.; Gao, L.; Xu, Y.; Li, Z.; and Wang, D. 2022. D-DPCC: Deep Dynamic Point Cloud Compression via 3D Motion Prediction. In Raedt, L. D., ed., IJCAI, 898--904

    2022

  5. [13]

    Fu, C.; Li, G.; Song, R.; Gao, W.; and Liu, S. 2022. OctAttention: Octree-Based Large-Scale Contexts Model for Point Cloud Compression. In AAAI, 625--633

    2022

  6. [14]

    Gao, L.; Fan, T.; Wan, J.; Xu, Y.; Sun, J.; and Ma, Z. 2021. Point Cloud Geometry Compression Via Neural Graph Sampling. In IEEE Int. Conf. Image Process., 3373--3377. IEEE

    2021

  7. [15]

    Goyal, V. K. 2001. Theoretical foundations of transform coding. IEEE Signal Process. Mag. , 18(5): 9--21

    2001

  8. [16]

    He, Y.; Ren, X.; Tang, D.; Zhang, Y.; Xue, X.; and Fu, Y. 2022. Density-preserving Deep Point Cloud Compression. In IEEE Conf. Comput. Vis. Pattern Recog., 2323--2332. IEEE

    2022

  9. [17]

    Hu, Z.; and Xu, D. 2023. Complexity-guided Slimmable Decoder for Efficient Deep Video Compression. In IEEE Conf. Comput. Vis. Pattern Recog., 14358--14367

    2023

  10. [18]

    Huang, L.; Wang, S.; Wong, K.; Liu, J.; and Urtasun, R. 2020. OctSqueeze: Octree-Structured Entropy Model for LiDAR Compression. In IEEE Conf. Comput. Vis. Pattern Recog., 1310--1320

    2020

  11. [19]

    Huang, T.; and Liu, Y. 2019. 3D Point Cloud Geometry Compression on Deep Learning. In Amsaleg, L.; Huet, B.; Larson, M. A.; Gravier, G.; Hung, H.; Ngo, C.; and Ooi, W. T., eds., ACM Int. Conf. Multimedia, 890--898

    2019

  12. [20]

    Li, Z.; Wang, W.; Wang, Z.; and Lei, N. 2024. Point Cloud Compression via Constrained Optimal Transport. CoRR, abs/2403.08236

    2024 arXiv

  13. [21]

    Liang, Z.; and Liang, F. 2022. TransPCC: Towards Deep Point Cloud Compression via Transformers. In Oria, V.; Sapino, M. L.; Satoh, S.; Kerherv \' e , B.; Cheng, W.; Ide, I.; and Singh, V. K., eds., International Conference on Multimedia Retrieval, 1--5

    2022

  14. [22]

    Liu, L.; Hu, Z.; and Zhang, J. 2023. PCHM-Net: A New Point Cloud Compression Framework for Both Human Vision and Machine Vision. In Int. Conf. Multimedia and Expo, 1997--2002

    2023

  15. [23]

    Mekuria, R.; Blom, K.; and C \' e sar, P. 2017. Design, Implementation, and Evaluation of a Point Cloud Codec for Tele-Immersive Video. IEEE Trans. Circuit Syst. Video Technol., 27(4): 828--842

    2017

  16. [24]

    Minnen, D.; Ball \' e , J.; and Toderici, G. 2018. Joint Autoregressive and Hierarchical Priors for Learned Image Compression. In Adv. Neural Inform. Process. Syst., 10794--10803

    2018

  17. [25]

    T.; Quach, M.; Valenzise, G.; and Duhamel, P

    Nguyen, D. T.; Quach, M.; Valenzise, G.; and Duhamel, P. 2021. Lossless Coding of Point Cloud Geometry Using a Deep Generative Model. IEEE Trans. Circuit Syst. Video Technol., 31(12)

    2021

  18. [26]

    Pan, Z.; Xiao, M.; Han, X.; Yu, D.; Zhang, G.; and Liu, Y. 2024. patchDPCC: A Patchwise Deep Compression Framework for Dynamic Point Clouds. In AAAI, 4406--4414

    2024

  19. [27]

    Qi, Y.; He, Y.; Qi, X.; Zhang, Y.; and Yang, G. 2023. Dynamic Snake Convolution based on Topological Geometric Constraints for Tubular Structure Segmentation. In Int. Conf. Comput. Vis., 6047--6056

    2023

  20. [28]

    Quach, M.; Valenzise, G.; and Dufaux, F. 2019. Learning Convolutional Transforms for Lossy Point Cloud Geometry Compression. In IEEE Int. Conf. Image Process., 4320--4324

    2019

  21. [29]

    Quach, M.; Valenzise, G.; and Dufaux, F. 2020. Improved Deep Point Cloud Geometry Compression. In International Workshop on Multimedia Signal Processing, 1--6

    2020

  22. [30]

    Que, Z.; Lu, G.; and Xu, D. 2021. VoxelContext-Net: An Octree Based Framework for Point Cloud Compression. In IEEE Conf. Comput. Vis. Pattern Recog., 6042--6051

    2021

  23. [31]

    A.; Cohen, R

    Schwarz, S.; Preda, M.; Baroncini, V.; Budagavi, M.; C \' e sar, P.; Chou, P. A.; Cohen, R. A.; Krivokuca, M.; Lasserre, S.; Li, Z.; Llach, J.; Mammou, K.; Mekuria, R.; Nakagami, O.; Siahaan, E.; Tabatabai, A. J.; Tourapis, A. M.; and Zakharchenko, V. 2019. Emerging MPEG Stand...

    2019

  24. [32]

    Song, R.; Fu, C.; Liu, S.; and Li, G. 2023. Efficient Hierarchical Entropy Model for Learned Point Cloud Compression. In IEEE Conf. Comput. Vis. Pattern Recog., 14368--14377

    2023

  25. [33]

    Tang, H.; Yang, S.; Liu, Z.; Hong, K.; Yu, Z.; Li, X.; Dai, G.; Wang, Y.; and Han, S. 2023. TorchSparse++: Efficient Training and Inference Framework for Sparse Convolution on GPUs. In Proceedings of the 56th Annual IEEE/ACM International Symposium on Microarchitecture , 225--239

    2023

  26. [34]

    Tao, L.; Gao, W.; Li, G.; and Zhang, C. 2023. AdaNIC: Towards Practical Neural Image Compression via Dynamic Transform Routing. In Int. Conf. Comput. Vis., 16833--16842

    2023

  27. [35]

    A.; and Frossard, P

    Thanou, D.; Chou, P. A.; and Frossard, P. 2016. Graph-Based Compression of Dynamic 3D Point Cloud Sequences. IEEE Trans. Image Process., 25(4): 1765--1778

    2016

  28. [36]

    Veit, A.; and Belongie, S. J. 2020. Convolutional Networks with Adaptive Inference Graphs. Int. J. Comput. Vis., 128(3): 730--741

    2020

  29. [37]

    Wang, J.; Ding, D.; Chen, H.; and Ma, Z. 2023 a . Dynamic Point Cloud Geometry Compression Using Multiscale Inter Conditional Coding. CoRR, abs/2301.12165

    2023 arXiv

  30. [38]

    Wang, J.; Ding, D.; Li, Z.; Feng, X.; Cao, C.; and Ma, Z. 2023 b . Sparse Tensor-Based Multiscale Representation for Point Cloud Geometry Compression. IEEE Trans. Pattern Anal. Mach. Intell., 45(7)

    2023

  31. [39]

    Wang, J.; Ding, D.; Li, Z.; and Ma, Z. 2021 a . Multiscale Point Cloud Geometry Compression. In Bilgin, A.; Marcellin, M. W.; Serra - Sagrist \` a , J.; and Storer, J. A., eds., Data Compression Conference, 73--82

    2021

  32. [40]

    Wang, J.; Zhu, H.; Liu, H.; and Ma, Z. 2021 b . Lossy Point Cloud Geometry Compression via End-to-End Learning. IEEE Trans. Circuit Syst. Video Technol., 31(12): 4909--4923

    2021

  33. [41]

    Wang, W.; Dai, J.; Chen, Z.; Huang, Z.; Li, Z.; Zhu, X.; Hu, X.; Lu, T.; Lu, L.; Li, H.; Wang, X.; and Qiao, Y. 2023 c . InternImage: Exploring Large-Scale Vision Foundation Models with Deformable Convolutions. In IEEE Conf. Comput. Vis. Pattern Recog., 14408--14419

    2023

  34. [42]

    Xia, S.; Fan, T.; Xu, Y.; Hwang, J.; and Li, Z. 2023. Learning Dynamic Point Cloud Compression via Hierarchical Inter-frame Block Matching. In El - Saddik, A.; Mei, T.; Cucchiara, R.; Bertini, M.; Vallejo, D. P. T.; Atrey, P. K.; and Hossain, M. S., eds., ACM Int. Conf. Multim...

    2023

  35. [43]

    Xie, L.; and Gao, W. 2024 a . LearningPCC: A PyTorch Library for Learning-Based Point Cloud Compression. In ACM Int. Conf. Multimedia

    2024

  36. [44]

    Xie, L.; and Gao, W. 2024 b . PCHMVision: An Open-Source Library of Point Cloud Compression for Human and Machine Vision. In ACM Int. Conf. Multimedia

    2024

  37. [45]

    Xie, L.; Gao, W.; Zheng, H.; and Li, G. 2024 a . ROI-Guided Point Cloud Geometry Compression Towards Human and Machine Vision. In ACM Int. Conf. Multimedia

    2024

  38. [46]

    Xie, L.; Gao, W.; Zheng, H.; and Li, G. 2024 b . SPCGC: Scalable Point Cloud Geometry Compression for Machine Vision . In IEEE International Conference on Robotics and Automation, 594--595

    2024

  39. [47]

    Xie, L.; Gao, W.; Zheng, H.; and Ye, H. 2024 c . Semantic-Aware Visual Decomposition for Point Cloud Geometry Compression. In 2024 Data Compression Conference (DCC), 595--595

    2024

  40. [48]

    Xie, L.; Mu, X.; and Gao, W. 2024. PKU-DPCC: A New Dataset for Dynamic Point Cloud Compression. In APSIPA Transactions on Signal and Information Processing

    2024

  41. [49]

    H.; Li, Z.; and Li, G

    Yan, W.; Shao, Y.; Liu, S.; Li, T. H.; Li, Z.; and Li, G. 2019. Deep AutoEncoder-based Lossy Geometry Compression for Point Clouds. CoRR, abs/1905.03691

    2019 arXiv

  42. [50]

    Yang, C.; Wang, X.; Yao, L.; Long, G.; and Xu, G. 2024. Dyformer: A dynamic transformer-based architecture for multivariate time series classification. Inf. Sci., 656: 119881

    2024

  43. [51]

    Yang, F.; Herranz, L.; Cheng, Y.; and Mozerov, M. G. 2021. Slimmable Compressive Autoencoders for Practical Neural Image Compression. In IEEE Conf. Comput. Vis. Pattern Recog

    2021

  44. [52]

    Zhang, C.; and Gao, W. 2024. Learned Rate Control for Frame-Level Adaptive Neural Video Compression via Dynamic Neural Network. In Eur. Conf. Comput. Vis. Springer

    2024

  45. [53]

    Zhang, J.; Chen, T.; Ding, D.; and Ma, Z. 2023. YOGA: Yet Another Geometry-based Point Cloud Compressor. In El - Saddik, A.; Mei, T.; Cucchiara, R.; Bertini, M.; Vallejo, D. P. T.; Atrey, P. K.; and Hossain, M. S., eds., ACM Int. Conf. Multimedia, 9070--9081. ACM

    2023

  46. [54]

    Zuo, D.; Yu, P.; Huang, R.; Huang, Y.; Sun, W.; and Liang, F. 2023. Diverse Context Model for Large-Scale Dynamic Point Cloud Compression. In VCIP, 1--5

    2023

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.