Pith. sign in

REVIEW 4 major objections 4 minor 41 references

Perceive-Sample-Compress: Towards Real-Time 3D Gaussian Splatting

T0 review · 4 major / 4 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 This paper claims that 3D Gaussian Splatting can be made memory-efficient and real-time at large scales by combining a camera-aware level-of-detail pyramid with a Generalized Gaussian Mixed entropy coder.

desk verdict A workmanlike 3DGS compression/LOD paper with strong large-scale results, but the SPC equation contradicts its own text and needs a fix before the ablation can be fully trusted. read the letter →

arxiv 2508.04965 v1 pith:VJZPL7Q7 submitted 2025-08-07 cs.GR cs.CVcs.MM

Zijian Wang , Beizhen Zhao , Hao Wang This is my paper
classification cs.GRcs.CVcs.MM
keywords 3DGaussianSplattingLevel-of-DetailPyramidrepresentationSceneperceptioncompensationGeneralizedDistributionEntropycodingNovelviewsynthesisLarge-scalerendering
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 3D Gaussian Splatting, which normally stores scenes as large flat collections of Gaussians, can be reorganized into a hierarchy that is both cheaper to store and faster to render. The proposed Pyramid-GS pipeline works in three steps: a scene perception compensation that adjusts level-of-detail assignments using camera coverage and depth-of-field cues, a pyramid sampling representation built from Laplacian-pyramid voxelization, and a Generalized Gaussian Mixed compression model that entropy-codes the hierarchical Gaussians. The authors report competitive or better rendering quality than several compression and LOD baselines across four datasets, while using fewer Gaussian primitives and smaller memory footprints. This matters because storage size and scalability are the main barriers to deploying 3DGS in large-scale or memory-constrained settings.

What carries the argument

The load-bearing machinery is the three-part pipeline: scene perception compensation (Eqs. 1-4) computes a depth-variance compensation factor and a coverage-based visibility threshold to assign pyramid levels; Laplacian pyramid sampling (Eqs. 5-6) voxelizes the scene at multiple resolutions and stores residuals between levels; and the Generalized Gaussian Mixed entropy model (Eqs. 8-9) assigns β=1 to high-frequency residual features and β=2 to other parameters, producing the rate term in the end-to-end loss (Eq. 10). The GGD shape choice is what makes the compression estimates match the assumed statistics of the pyramid attributes.

What would settle it

Train a Pyramid-GS model on one Waymo scene, extract the quantized attribute tensors, and fit a Generalized Gaussian to each attribute's empirical histogram via maximum likelihood. If the fitted shape β for high-frequency residual features is far from 1, or for means, opacities, and colors is far from 2, the entropy model is miscalibrated. A direct check is to compare the predicted rate, the sum of $-\log_2 p_{GGD}$ over all attributes, against the actual compressed file size; a large gap would indicate the claimed memory savings are not realizable.

Watch

Extended reading notes

Core claim

The central claim is that a perceive-sample-compress design can simultaneously improve memory efficiency and rendering quality in 3D Gaussian Splatting. Scene perception compensation refines which pyramid level each anchor belongs to, using depth variance as a proxy for depth-of-field and camera coverage as a visibility prior. Pyramid sampling then organizes anchors into Laplacian-pyramid voxel levels, so coarse layers capture global structure and finer layers store residuals. The compression module applies a Generalized Gaussian Distribution with shape parameter β=1 (Laplace) to high-frequency residuals and β=2 (Gaussian) to other attributes, predicting per-attribute location and scale with

Load-bearing premise

The compression module assumes that Gaussian attributes follow a Generalized Gaussian Distribution with shape β=1 for high-frequency residuals and β=2 for other parameters; if the real trained attribute distributions differ, the entropy model is miscalibrated and the reported file sizes would not hold.

Editorial extensions

If this is right

  • If the central claim holds, large-scale driving and city scenes from Waymo and MatrixCity can be stored in 9.5 MB and 44.0 MB respectively, with rendering speed at or above 138 FPS on an RTX A6000.
  • On small object-centric scenes, the method reaches 29.39 PSNR on Mip-NeRF360 using 377k Gaussians, fewer than all compared baselines, and on Tanks & Temples it uses 317k Gaussians at the same PSNR as HAC++ with lower LPIPS.
  • The rate-distortion weight λ provides a practical quality-size slider: on Waymo, raising λ from 0.0005 to 0.004 reduces memory from 9.5 MB to 4.1 MB and increases FPS from 138 to 163 at a cost of roughly 0.5 dB PSNR.
  • The compression overhead is modest: encoding and decoding take 16.31 s and 28.95 s on Waymo, faster than the Context-GS and HAC++ numbers reported in the paper, so the pipeline is usable outside offline setting.

Reading between the lines

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

  • Editorial extension: the same perceive-sample-compress split could be applied to other primitive-based renderers; any representation with hierarchical LOD and quantized attributes can reuse the depth-variance level assignment and the GGD entropy model.
  • Editorial extension: because the pyramid already separates coarse structure from residuals, the bitstream could be made progressive, streaming coarse Gaussians first and adding residual levels on demand; the paper does not discuss this.
  • Editorial extension: the fixed β=1 for high-frequency residuals is a strong prior; a learned or per-scene fitted shape parameter could improve compression if real residual histograms are heavier- or lighter-tailed than Laplace. This is an empirical question the paper leaves open.
  • Editorial extension: the depth-of-field compensation is a proxy for visual importance; one could test it against saliency-based or gradient-based importance maps to see whether camera coverage is the right criterion for allocating Gaussians.
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 Pyramid-GS, a 3D Gaussian Splatting framework with three components: a scene perception compensation (SPC) algorithm that adjusts pyramid level assignment using camera coverage and depth-of-field variance (Eqs. 1–4), a Laplacian-pyramid/voxelization sampling scheme (Eqs. 5–7), and a Generalized Gaussian Mixed entropy model for compression (Eqs. 8–10). Experiments on Waymo, MatrixCity, Mip-NeRF360, and Tanks & Temples compare against seven baselines and report favorable PSNR/LPIPS and memory usage on large-scale scenes, with competitive real-time FPS; on Tanks & Temples the method is below Octree-GS in PSNR and SSIM. Ablation studies in Tables 3–5 attribute gains to the proposed components. No code is released.

Significance. If the results hold, the paper would be a practically useful contribution to large-scale 3DGS: the hierarchical representation reduces storage and active Gaussian counts on outdoor scenes while maintaining or improving quality, and the compression module has lower encoding/decoding times than HAC++. The broad comparison across four datasets is a strength. However, the central claims are weakened by an internal inconsistency in the SPC equations, under-specified pyramid construction, and the absence of code and statistical uncertainty. The manuscript needs substantial revision before the results can be accepted as stated.

major comments (4)
  1. [Scene Perception Compensation] The text states that 'a larger σz,j leads to a smaller fdepth, effectively making objects appear closer... finer levels,' but Eq. (1) defines fdepth = 1 + α·max(0, σz/σz,thresh − 1), which increases with σz. Since d′ij = dij·fdepth and Lij = log2(Dstd/d′ij), a larger σz produces a larger adjusted distance and hence a smaller L (or, if level ordering is inverted, the stated rationale fails). The level-to-resolution mapping is never defined, and no code is released, so the reader cannot determine which variant was actually used. This is load-bearing because Table 3 attributes a 1.06 dB PSNR gain and a 33k-primitive reduction to SPC. Please correct Eq. (1) or the text, and specify the integer level mapping.
  2. [Pyramid Sampling Representation] The notation is ambiguous and under-specified: V′l, Rl, Downsample, Upsample are not precisely defined, and Eq. (5) uses V′l−1 = Rl−1 ∩ P and Rl−1 = Downsample(V′l, ρl−1) while Eq. (6) redefines Rl = Upsample(V′l−1, ρl). It is unclear how residuals are computed, how levels are numbered, and how the MLP in Eq. (7) is conditioned on voxel features. The paper only states that the number of levels L is 'automatically set according to the dataset.' Because the pyramid is a core contribution, this vagueness prevents reproducibility and verification.
  3. [Generalized Gaussian Mixed Compression] The choice of β=1 for high-frequency residual features and β=2 for other parameters is justified only by a qualitative illustration (Fig. 2). No distributional fits, likelihood comparisons, or ablations over β are provided. Since the entropy model directly determines the reported storage sizes, the claim that GGD with fixed β provides a 'statistically sound foundation' is not established. Please add quantitative distribution analysis or an ablation comparing fixed β values to a learned or non-parametric density.
  4. [Table 3] The ablation results are internally puzzling: the 'w/o SPC&PS' row (PSNR 27.90, SSIM 0.846) is better than 'w/o SPC' (PSNR 27.12, SSIM 0.833), and 'w/o PS' (27.95) is only marginally better than removing both. This is inconsistent with the claim that SPC provides a clear benefit. The interactions between components need to be discussed, and ideally the ablation should include standard errors over multiple runs.
minor comments (4)
  1. [Tables 1–2] All quantitative results are reported without error bars or multiple seeds; given the small differences between some methods (e.g., Mip-NeRF360 PSNR 29.39 vs. 29.35), the significance of the reported gains is unclear.
  2. [Table 2 / Conclusion] On Tanks & Temples, the method's PSNR (24.32) is tied with HAC++ and below Octree-GS (24.68), and SSIM (0.847) is below several baselines. The text calls this 'near-optimal' and the abstract claims 'high visual quality'; this mismatch should be addressed honestly.
  3. [Implementation Details] The text refers to the 'Laplacian Mixed Model' instead of 'Generalized Gaussian Mixed model'; also 'superparameter' should be 'hyperparameter', and 'the most best results' is ungrammatical.
  4. [General] The paper would benefit from a pseudocode algorithm for the complete pipeline, a precise definition of 'current level' in the visibility test, and a description of how the integer pyramid level is derived from Lij.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: Pyramid-GS is an empirical engineering system; reported gains come from external benchmarks, not from fitting or self-citation.

full rationale

I walked the claimed derivation chain: scene perception compensation, pyramid sampling, and GGD-based compression. None of these components derives its reported result from its own inputs by construction. The compression model assumes pre-set GGD shape parameters (β=1 for residuals, β=2 for others) but this is an explicit modeling assumption, not a fitted parameter renamed as a prediction; the reported PSNR/Memory numbers come from end-to-end training and external benchmark evaluation. No load-bearing self-citation was found: references to HAC++ and Scaffold-GS are prior external works, and there is no invocation of a uniqueness theorem from the authors' own previous papers. The pyramid equations define a standard multi-scale aggregation; they are recursive but not circular. One internal inconsistency exists in the Scene Perception Compensation section: the text states that a larger σ_z,j should lead to a smaller f_depth ("effectively making objects appear closer ... finer levels"), while Eq. (1) defines f_depth as increasing in σ_z,j for σ_z,j > σ_z,thresh. This is a real algorithmic correctness/reproducibility concern that should be fixed, but it is not circularity: the reported gains in Table 3 are empirical and are not derived from Eq. (1). Therefore the circularity score is 0.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The framework depends on several manually chosen hyperparameters (α, σz_thresh, β, λ, Dstd, L). These are not derived and could be dataset-specific. The domain assumptions about Gaussian attribute distributions and the level-partition formula are empirical heuristics. No new physical entities are introduced.

free parameters (5)
  • α (distance compensation coefficient) = 0.7
    Used in Eq. 1 to scale depth-of-field compensation; set by hand, not derived.
  • σz_thresh (depth-of-field standard deviation threshold) = 50.0
    Determines when depth compensation activates; manually chosen.
  • β (coverage weight) = 0.5
    Controls the normalized coverage score in Eq. 3; set by hand.
  • λ (rate-distortion trade-off) = 0.0005
    Balances rendering quality and compression in Eq. 10; set uniformly for all experiments.
  • Dstd (standard reference distance) = not specified
    Maps adjusted distance to pyramid level in Eq. 2; arbitrary reference, no justification.
assumptions (5)
  • domain assumption High-frequency pyramid residuals follow a Laplace distribution (GGD β=1) and other parameters follow a Gaussian distribution (GGD β=2).
    Stated in 'Generalized Gaussian Mixed Compression' and illustrated qualitatively in Fig. 2, but never validated with distributional fits on actual data.
  • domain assumption Camera coverage and depth-of-field variance correlate with visual importance of anchors.
    Basis of the scene perception compensation; plausible but not justified beyond qualitative examples.
  • ad hoc to paper The formula Lij = log2(Dstd / d′ij) with integer mapping produces a good pyramid-level partition.
    Introduced without comparison to alternative level-assignment functions or theoretical justification.
  • ad hoc to paper The visible threshold update τnew = (1+C)τold improves level selection stability.
    Heuristic update rule with no convergence or optimality analysis.
  • domain assumption MLP networks can predict Gaussian parameters from voxel coordinates.
    Standard in 3DGS literature, used in Scaffold-GS and HAC++; accepted without new evidence.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Perceive-Sample-Compress: Towards Real-Time 3D Gaussian Splatting." pith.science (2026). https://pith.science/paper/VJZPL7Q7

@misc{pith2026250804965,
  author       = {Pith},
  title        = {Pith review of: Perceive-Sample-Compress: Towards Real-Time 3D Gaussian Splatting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VJZPL7Q7}},
  note         = {Machine review of arXiv:2508.04965}
}
read the original abstract

Recent advances in 3D Gaussian Splatting (3DGS) have demonstrated remarkable capabilities in real-time and photorealistic novel view synthesis. However, traditional 3DGS representations often struggle with large-scale scene management and efficient storage, particularly when dealing with complex environments or limited computational resources. To address these limitations, we introduce a novel perceive-sample-compress framework for 3D Gaussian Splatting. Specifically, we propose a scene perception compensation algorithm that intelligently refines Gaussian parameters at each level. This algorithm intelligently prioritizes visual importance for higher fidelity rendering in critical areas, while optimizing resource usage and improving overall visible quality. Furthermore, we propose a pyramid sampling representation to manage Gaussian primitives across hierarchical levels. Finally, to facilitate efficient storage of proposed hierarchical pyramid representations, we develop a Generalized Gaussian Mixed model compression algorithm to achieve significant compression ratios without sacrificing visual fidelity. The extensive experiments demonstrate that our method significantly improves memory efficiency and high visual quality while maintaining real-time rendering speed.

Figures

Figures reproduced from arXiv: 2508.04965 by the authors.

Figure 1
Figure 1. Visualization of pyramid level Compared with Scaffold-GS, our method can achieve higher PSNR scores with fewer Gaussian primitives. 2024a) adopt context-aware approaches that explicitly incor￾porate hierarchical relationships among anchors and Gaus￾sians to enhance representational capacity. HAC++ (Chen et al. 2025) further introduces hashed features as priors for entropy coding and demonstrates its effectiveness in… view at source ↗
Figure 2
Figure 2. Framework of Pyramid-GS. We begin by sampling the input sparse point cloud. We propose a scene perception compensation algorithm, which take the camera coverage and depth of field accounts for pyramid level partition. Then we design a hierarchical pyramid sampling representation to set up pyramid structure. Finally, we utilize compression algorithm for efficient storage while maintaining real-time rendering speed an… view at source ↗
Figure 3
Figure 3. Comparison Results. Visual differences are highlighted with yellow insets for better clarity. Our approach consis￾tently outperforms other models on different scenes, demonstrating advantages in challenging scenarios. Best viewed in color [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Effect of Scene Perception Compensation. While relying solely on camera distance can work well in most situations, in some extreme scenes it may fail and re￾sult in poor rendering results. This is because that too many key points are ignored by the algorithm. Through o…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

41 extracted references · 24 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]

    S.; Qamar, M.; Bae, S.-H.; and Tartaglione, E

    Ali, M. S.; Qamar, M.; Bae, S.-H.; and Tartaglione, E. 2024. Trimming the fat: Efficient compression of 3d gaussian splats through pruning. arXiv preprint arXiv:2406.18214

    arXiv 2024

  4. [4]

    Avidan, S.; and Shashua, A. 1997. Novel view synthesis in tensor space. In Proceedings of IEEE computer society conference on computer vision and pattern recognition, 1034--1040. IEEE

    work page 1997

  5. [5]

    T.; Mildenhall, B.; Verbin, D.; Srinivasan, P

    Barron, J. T.; Mildenhall, B.; Verbin, D.; Srinivasan, P. P.; and Hedman, P. 2022. Mip-nerf 360: Unbounded anti-aliased neural radiance fields. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 5470--5479

    2022

  6. [6]

    Chen, J.; Ye, W.; Wang, Y.; Chen, D.; Huang, D.; Ouyang, W.; Zhang, G.; Qiao, Y.; and He, T. 2024 a . Gigags: Scaling up planar-based 3d gaussians for large scene surface reconstruction. arXiv preprint arXiv:2409.06685

    arXiv 2024

  7. [7]

    Chen, Y.; Wu, Q.; Lin, W.; Harandi, M.; and Cai, J. 2024 b . Hac: Hash-grid assisted context for 3d gaussian splatting compression. In European Conference on Computer Vision, 422--438. Springer

    work page 2024

  8. [8]

    Chen, Y.; Wu, Q.; Lin, W.; Harandi, M.; and Cai, J. 2025. Hac++: Towards 100x compression of 3d gaussian splatting. arXiv preprint arXiv:2501.12255

    arXiv 2025

Show all 41 references
  1. [9]

    H.; and Kautz, J

    Choi, I.; Gallo, O.; Troccoli, A.; Kim, M. H.; and Kautz, J. 2019. Extreme view synthesis. In Proceedings of the IEEE/CVF international conference on computer vision, 7781--7790

    2019

  2. [10]

    Cui, J.; Cao, J.; Zhao, F.; He, Z.; Chen, Y.; Zhong, Y.; Xu, L.; Shi, Y.; Zhang, Y.; and Yu, J. 2024. Letsgo: Large-scale garage modeling and rendering via lidar-assisted gaussian primitives. ACM Transactions on Graphics (TOG), 43(6): 1--18

    2024

  3. [11]

    G.; and Knausg rd, K

    Dalal, A.; Hagen, D.; Robbersmyr, K. G.; and Knausg rd, K. M. 2024. Gaussian splatting: 3D reconstruction and novel view synthesis: A review. IEEE Access, 12: 96797--96820

    2024

  4. [12]

    David, L.; Reddy, M.; Cohen, J.; Varshney, A.; Watson, B.; and Huebner, R. 2003. Level of detail for 3D graphics

    2003

  5. [13]

    Fan, Z.; Wang, K.; Wen, K.; Zhu, Z.; Xu, D.; Wang, Z.; et al. 2024. Lightgaussian: Unbounded 3d gaussian compression with 15x reduction and 200+ fps. Advances in neural information processing systems, 37: 140138--140158

    2024

  6. [14]

    Girish, S.; Gupta, K.; and Shrivastava, A. 2024. Eagles: Efficient accelerated 3d gaussians with lightweight encodings. In European Conference on Computer Vision, 54--71. Springer

    2024

  7. [15]

    Huang, B.; Yu, Z.; Chen, A.; Geiger, A.; and Gao, S. 2024. 2d gaussian splatting for geometrically accurate radiance fields. In ACM SIGGRAPH 2024 conference papers, 1--11

    2024

  8. [16]

    Kerbl, B.; Kopanas, G.; Leimk \"u hler, T.; and Drettakis, G. 2023. 3D Gaussian splatting for real-time radiance field rendering. ACM Trans. Graph., 42(4): 139--1

    2023

  9. [17]

    Kerbl, B.; Meuleman, A.; Kopanas, G.; Wimmer, M.; Lanvin, A.; and Drettakis, G. 2024. A hierarchical 3d gaussian representation for real-time rendering of very large datasets. ACM Transactions on Graphics (TOG), 43(4): 1--15

    2024

  10. [18]

    Knapitsch, A.; Park, J.; Zhou, Q.-Y.; and Koltun, V. 2017. Tanks and temples: Benchmarking large-scale scene reconstruction. ACM Transactions on Graphics (ToG), 36(4): 1--13

    2017

  11. [19]

    C.; Rho, D.; Sun, X.; Ko, J

    Lee, J. C.; Rho, D.; Sun, X.; Ko, J. H.; and Park, E. 2024. Compact 3d gaussian representation for radiance field. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 21719--21728

    2024

  12. [20]

    Li, Y.; Jiang, L.; Xu, L.; Xiangli, Y.; Wang, Z.; Lin, D.; and Dai, B. 2023. Matrixcity: A large-scale city dataset for city-scale neural rendering and beyond. In Proceedings of the IEEE/CVF International Conference on Computer Vision, 3205--3215

    2023

  13. [21]

    Liu, X.; Wu, X.; Zhang, P.; Wang, S.; Li, Z.; and Kwong, S. 2024 a . Compgs: Efficient 3d scene representation via compressed gaussian splatting. In Proceedings of the 32nd ACM International Conference on Multimedia, 2936--2944

    2024

  14. [22]

    Liu, Y.; Luo, C.; Fan, L.; Wang, N.; Peng, J.; and Zhang, Z. 2024 b . Citygaussian: Real-time high-quality large-scale scene rendering with gaussians. In European Conference on Computer Vision, 265--282. Springer

    2024

  15. [23]

    Lombardi, S.; Simon, T.; Saragih, J.; Schwartz, G.; Lehrmann, A.; and Sheikh, Y. 2019. Neural volumes: Learning dynamic renderable volumes from images. arXiv preprint arXiv:1906.07751

    2019 arXiv

  16. [24]

    Lu, T.; Yu, M.; Xu, L.; Xiangli, Y.; Wang, L.; Lin, D.; and Dai, B. 2024. Scaffold-gs: Structured 3d gaussians for view-adaptive rendering. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 20654--20664

    2024

  17. [25]

    Milef, N.; Seyb, D.; Keeler, T.; Nguyen-Phuoc, T.; Bo z i c , A.; Kondguli, S.; and Marshall, C. 2025. Learning fast 3D gaussian splatting rendering using continuous level of detail. In Computer Graphics Forum, e70069. Wiley Online Library

    2025

  18. [26]

    P.; Koohpayegani, S

    Navaneet, K.; Meibodi, K. P.; Koohpayegani, S. A.; and Pirsiavash, H. 2023. Compact3d: Compressing gaussian splat radiance field models with vector quantization. arXiv preprint arXiv:2311.18159, 2(3)

    2023 arXiv

  19. [27]

    Niedermayr, S.; Stumpfegger, J.; and Westermann, R. 2024. Compressed 3d gaussian splatting for accelerated novel view synthesis. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 10349--10358

    2024

  20. [28]

    Papantonakis, P.; Kopanas, G.; Kerbl, B.; Lanvin, A.; and Drettakis, G. 2024. Reducing the memory footprint of 3d gaussian splatting. Proceedings of the ACM on Computer Graphics and Interactive Techniques, 7(1): 1--17

    2024

  21. [29]

    Penner, E.; and Zhang, L. 2017. Soft 3d reconstruction for view synthesis. ACM Transactions on Graphics (TOG), 36(6): 1--11

    2017

  22. [30]

    Ren, K.; Jiang, L.; Lu, T.; Yu, M.; Xu, L.; Ni, Z.; and Dai, B. 2024. Octree-gs: Towards consistent real-time rendering with lod-structured 3d gaussians. arXiv preprint arXiv:2403.17898

    2024 arXiv

  23. [31]

    Riegler, G.; and Koltun, V. 2020. Free view synthesis. In European conference on computer vision, 623--640. Springer

    2020

  24. [32]

    S.; Son, H

    Seo, Y.; Choi, Y. S.; Son, H. S.; and Uh, Y. 2024. Flod: Integrating flexible level of detail into 3d gaussian splatting for customizable rendering. arXiv preprint arXiv:2408.12894

    2024 arXiv

  25. [33]

    Sun, P.; Kretzschmar, H.; Dotiwalla, X.; Chouard, A.; Patnaik, V.; Tsui, P.; Guo, J.; Zhou, Y.; Chai, Y.; Caine, B.; et al. 2020. Scalability in perception for autonomous driving: Waymo open dataset. In Proceedings of the IEEE/CVF conference on computer vision and pattern reco...

    2020

  26. [34]

    Wang, Y.; Li, Z.; Guo, L.; Yang, W.; Kot, A.; and Wen, B. 2024. Contextgs: Compact 3d gaussian splatting with anchor level context model. Advances in neural information processing systems, 37: 51532--51551

    2024

  27. [35]

    C.; Sheikh, H

    Wang, Z.; Bovik, A. C.; Sheikh, H. R.; and Simoncelli, E. P. 2004. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4): 600--612

    2004

  28. [36]

    Wang, Z.; and Xu, D. 2024. Pygs: Large-scale scene representation with pyramidal 3d gaussian splatting. arXiv preprint arXiv:2405.16829

    2024 arXiv

  29. [37]

    Watson, D.; Chan, W.; Martin-Brualla, R.; Ho, J.; Tagliasacchi, A.; and Norouzi, M. 2022. Novel view synthesis with diffusion models. arXiv preprint arXiv:2210.04628

    2022 arXiv

  30. [38]

    Wizadwongsa, S.; Phongthawee, P.; Yenphraphai, J.; and Suwajanakorn, S. 2021. Nex: Real-time view synthesis with neural basis expansion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 8534--8543

    2021

  31. [39]

    P.; Szeliski, R.; Barron, J

    Yariv, L.; Hedman, P.; Reiser, C.; Verbin, D.; Srinivasan, P. P.; Szeliski, R.; Barron, J. T.; and Mildenhall, B. 2023. Bakedsdf: Meshing neural sdfs for real-time view synthesis. In ACM SIGGRAPH 2023 conference proceedings, 1--9

    2023

  32. [40]

    A.; Shechtman, E.; and Wang, O

    Zhang, R.; Isola, P.; Efros, A. A.; Shechtman, E.; and Wang, O. 2018. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recognition, 586--595

    2018

  33. [41]

    Zhou, T.; Tucker, R.; Flynn, J.; Fyffe, G.; and Snavely, N. 2018. Stereo magnification: Learning view synthesis using multiplane images. arXiv preprint arXiv:1805.09817

    2018 arXiv

Pith tools

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