Pith. sign in

REVIEW 4 major objections 6 minor 54 references

A compact set-based 3D VAE can match voxel-level reconstruction by densifying anchored latents and decoding them locally across scales.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-31 14:56 UTC pith:7PYCRJD5

load-bearing objection Solid hierarchical densification of anchored VecSets that really does lift set-based VAE fidelity; the equal-token wins over unanchored baselines are partly inflated by free occupancy anchors, but the Lattice comparison and ablations still hold. the 4 major comments →

arxiv 2607.24436 v1 pith:7PYCRJD5 submitted 2026-07-27 cs.CV

MSVS-VAE: Multi-Scale Anchored VecSet for High-Fidelity 3D Reconstruction

Dehao Hao , Kaiyi Zhang , Tanghui Jia , Xiangjun Gao , Dongyu Yan , Weikai Chen , Zeyu Hu , Lingting Zhu
show 5 more authors
Yingda Yin Runze Zhang Li Yuan Xin Wang Long Quan
This is my paper
classification cs.CV
keywords 3D VAEVecSetlatent diffusionpoint-shuffle upsamplinglocal aggregationmulti-scale decodingSDF reconstructioncompact 3D representation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

High-fidelity 3D generation is bottlenecked by the VAE that turns shapes into latents for diffusion. Voxel VAEs reconstruct fine geometry well but need huge latent budgets; set-based VAEs stay compact and continuous but blur details because the latent set is sparse and decoding is overly global. This paper claims that gap is closable without giving up compactness: start from an anchored VecSet, progressively densify it with hierarchical point-shuffle upsampling, decode with a local geometry-aware operator instead of global attention, and fuse coarse-to-fine queries so long-range structure and local detail both survive. On Objaverse, ABO, and in-the-wild tests the method beats prior set and voxel VAEs while reporting roughly 10× faster decoding than earlier set methods and roughly 10× more compact latents than voxel baselines. A sympathetic reader cares because latent diffusion only scales if the VAE is both faithful and cheap to sample from.

Core claim

MSVS-VAE shows that hierarchical densification of anchored VecSet latents, paired with local AVS-Conv aggregation and multi-scale query fusion, lets a compact continuous set representation reach or surpass voxel-based reconstruction fidelity while remaining far smaller and decoding much faster than prior set-based VAEs.

What carries the argument

Hierarchical Point-Shuffle Upsampling of anchored VecSets, decoded by AVS-Conv (KNN-local geometry-aware aggregation) and Multi-Scale Query Decoding that fuses coarse global context with fine local residuals.

Load-bearing premise

The method depends on external active-voxel anchors from an off-the-shelf model; if those anchors miss structure or are unavailable, the claimed fidelity-compactness tradeoff may not hold in a pure set pipeline.

What would settle it

Train and evaluate the same architecture with anchors sampled only from the input surface (no off-the-shelf voxel generator) on the same Objaverse/ABO/in-the-wild splits; if Mesh Distance and F1@0.001 collapse relative to the reported 20k–40k token numbers, the central claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Set-based latents become a practical backbone for latent diffusion without sacrificing fine geometry.
  • Token budgets around 20k can already beat voxel baselines that use 50k–210k tokens on the reported benchmarks.
  • Local neighborhood decoding (AVS-Conv) makes dense surface querying scale with latent size far better than global cross-attention.
  • Test-time increases in latent count can be used as a quality dial without changing the trained weights.
  • Watertight continuous surfaces remain available while approaching voxel-level detail, reducing topology brittleness from discrete sparse grids.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If anchors must come from another generator, end-to-end pure-set pipelines may still need a learned anchor proposal stage before this densify-and-decode recipe fully replaces voxel pipelines.
  • The same multi-scale local query pattern could transfer to appearance or texture latents, which the paper flags as future work.
  • Sharp-feature metrics improve but the continuous implicit still struggles on knife-edge geometry; dual-contouring-style cues might be the next bottleneck to close.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes MSVS-VAE, a set-based 3D VAE intended to close the reconstruction-fidelity gap between compact VecSet-style latents and sparse voxel-based representations. Three components are introduced: (1) hierarchical point-shuffle upsampling (PSU), which progressively densifies anchored VecSet latents (each latent split into r children with predicted geometric offsets and halved channels) to raise spatial capacity; (2) AVS-Conv, a PointConv-style KNN-based local aggregation operator that replaces global cross-attention in both encoder and decoder, reducing query cost; and (3) multi-scale query decoding (MSQ), which fuses AVS-Conv features queried independently at each latent scale to balance global context against local detail. On Objaverse, ABO, in-the-wild shapes, and Dora-Bench sharp-edge subsets, the method reports large gains over set-based baselines (Dora, HY3D2.1, Lattice) at equal token budgets, matches or exceeds voxel-based baselines (SparC3D, SparseFlex-1024, Direct3D-S2) at ~10–40% of their token counts, and achieves 5.5–14.4× lower query latency than Lattice. Ablations identify PSU as the critical component and MSQ as a consistent smaller gain; test-time scaling with token budget is demonstrated up to 80K tokens.

Significance. If the numbers hold, this is a meaningful step for latent-diffusion 3D generation: set-based VAEs have trailed voxel-based ones on fine geometry, and closing that gap at 20K tokens with 5.5–14.4× decoding speedups over the direct predecessor would matter to the many pipelines (CLAY, TripoSG, Hunyuan3D) built on VecSet-style latents. The paper ships the right kind of evidence for an empirical systems paper: multi-benchmark quantitative tables including sharp-edge metrics on Dora-Bench, component ablations that isolate PSU as the critical piece and MSQ as a consistent refinement, a latency table, and test-time scaling curves. The wins over Lattice — the only like-for-like anchored baseline — are large and consistent across all four token budgets and three datasets, which is the strongest and fairest evidence in the paper. No code or machine-checked artifacts are provided, and two of the baselines (Lattice, SparC3D) are the authors' own re-implementations, which tempers reproducibility credit.

major comments (4)
  1. [§4.2 (anchor sampling)] The sentence "we sample latent anchor points C ... from the centers of active voxels intersected by the mesh, obtained from an off-the-shelf generative model (e.g., HY3D2.1)" is ambiguous between (a) voxelizing the input mesh itself (which is always available at VAE training/eval time and is presumably what Lattice does) and (b) importing the sparse structure predicted by an external generator. The distinction is load-bearing for the paper's framing as a self-contained set-based VAE: under reading (b), both reconstruction fidelity and the downstream generative pipeline depend on the external model's active-voxel structure being correct and complete, and no ablation isolates anchor-source quality. Please (i) state unambiguously where anchors come from at training time, at reconstruction-eval time, and in a full generation pipeline; (ii) add a robustness ablation with degraded anchors — e.
  2. [Abstract and Table 1 (compactness accounting)] The abstract advertises "~10x higher compactness than voxel-based baselines" and Tab. 1 compares at equal token count (e.g., the 4k row: MSVS-VAE MD 5.644 vs Dora 17.352 vs HY3D2.1 11.485). But each MSVS-VAE token carries an explicit anchor position (3 coordinates) derived from the shape's occupancy structure — a coarse occupancy map that unanchored baselines (Dora, HY3D2.1) do not receive at the same token budget, and whose 3M additional scalars (plus, under reading (b) above, a dependence on an external structure model) are not counted in the compactness ledger. Anchored-vs-unanchored is a legitimate design axis, and the comparison against Lattice — which shares the anchored design and is beaten soundly at every budget — is fair and is in fact the strongest evidence in the paper. But the token-only accounting should be corrected: state the total latent dimensionality (positions + featu
  3. [Table 2 vs Abstract ("consistently outperforms")] The abstract claims MSVS-VAE "consistently outperforms prior set-based and voxel-based VAEs," but Table 2 does not support this at the 20K budget: on Level-3 S-MD, SparseFlex scores 3.21 vs MSVS-VAE-20K 3.22, and on Level-4 S-MD SparseFlex scores 3.08 vs 3.58 — i.e., a voxel-based baseline is better on the sharp-edge metric the table is designed to highlight, on both difficulty levels. The claim holds at 40K tokens. Please either soften the abstract wording (e.g., "matches or outperforms ... at comparable or smaller budgets") or make explicit in the text that the sharp-metric advantage over SparseFlex requires the 40K configuration.
  4. [§5.2 (re-implemented baselines, † in Tables 1–2)] The two most important comparators in Tab. 1 — Lattice (the design ancestor and fair anchored baseline) and SparC3D (the strongest voxel baseline on sharp metrics) — are the authors' own re-implementations (marked †). Their numbers presumably differ from the published ones, and no detail is given on training data, steps, or whether the re-implementations reproduce the published results on a common benchmark. Since the headline claim is a large margin over Lattice at every token budget, the fairness of that re-implementation is load-bearing. Please report (i) the re-implementation protocol, (ii) a sanity comparison of re-implemented vs published numbers where available, and (iii) whether re-implemented baselines were given the same 400k-mesh training set and comparable compute.
minor comments (6)
  1. [§4.3 / §5.1] Key hyperparameters are not reported in the main text: number of upsampling stages L, split factor r, KNN size K, latent channel dimensions C_l, and the resulting M_l at each scale. These are needed to reproduce the architecture and to interpret the token-budget rows in Tab. 1. Please add a configuration table.
  2. [Eq. (5) (indexing)] The hierarchy is defined as Z^{(l)}, l = 0,...,L with Z^{(0)} = Z, but the fusion in Eq. (5) concatenates h_i^{(1)} through h_i^{(L)}. Either the coarsest scale is excluded from fusion (in which case say so and explain) or the indexing is off by one.
  3. [Tables 1–2 (MD scaling)] Unit conventions are inconsistent: Table 1 states MD is scaled by ×10^4, Table 2 states "MD and S-MD are scaled by 10^{-4}" — the latter is presumably meant to read ×10^4 as well. Also, no absolute length scale is given (are meshes normalized to a unit cube/sphere?), which is needed to interpret MD and the F1 thresholds.
  4. [Table 4 (latency)] No hardware is specified for the latency benchmark, and the comparison is only against Lattice; a voxel-based decoder latency row (even if unfavorable, as the Limitations section concedes) would make the efficiency story more complete. Also state whether cuBQL KNN build time is included in the reported latency.
  5. [§5.1 (progressive crop training)] The progressive crop fine-tuning stage supervises only near-surface queries within a crop, but evaluation metrics in Tab. 1 are computed on full shapes. Please clarify how full-shape latents are produced at test time (single global forward pass with M tokens, or chunked encoding with merging?), since this affects whether the reported numbers reflect the global or the locally fine-tuned regime.
  6. [Figures 5–7 / presentation] Fig. 5's ablation differences are described as visible only on zoom; consider adding error heatmaps or insets. Typo: "MSVS-V AE" appears with a stray space in several places (title, §4.1 heading, Table 2). Fig. 7 is referenced before Fig. 6 in the text. No code/model release statement is included; given that two baselines required re-implementation, releasing at least evaluation code would substantially strengthen the paper.

Circularity Check

0 steps flagged

No significant circularity: standard empirical VAE systems paper with external reconstruction metrics.

full rationale

MSVS-VAE proposes architectural modules (hierarchical point-shuffle upsampling, AVS-Conv, multi-scale query decoding), trains a VAE on filtered Objaverse, and reports held-out surface metrics (MD, F1, sharp-edge scores) plus wall-clock query latency against external baselines. None of the load-bearing claims reduce by construction to fitted free parameters or to a self-citation uniqueness chain. Anchors taken from active voxel centers (Sec. 4.2) and progressive densification are design inputs that may affect fairness of equal-token comparisons, but Mesh Distance and F1 are measured against independent ground-truth geometry, not algebraic restatements of those anchors. Ablations (Tab. 3) remove components and show metric degradation rather than tautological identity. Citations to Lattice, VecSet, PointConv, and point-shuffle literature supply prior techniques from other groups; they are not author-owned uniqueness theorems that force the result. This is the normal non-circular empirical pattern; score 0.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 3 invented entities

Load-bearing content is architectural and empirical, not theorematic. The claim rests on standard 3D-VAE domain assumptions (SDF/TSDF supervision, KL-regularized latents, iso-surface extraction), on Lattice-style anchored tokens, and on several hand-chosen training/architecture knobs (token budgets, truncation, upsampling hierarchy, KNN width, external anchors). Invented entities are engineering modules, not new physical objects; their only evidence is ablation and benchmark gains inside this paper.

free parameters (6)
  • TSDF truncation τ = 1/128
    Ground-truth SDF is clamp(s/τ,-1,1) with τ set empirically to 1/128; this directly shapes the reconstruction loss near the surface.
  • Latent token budget M and test-time scales = train 4k; eval up to 40k+
    Training uses M=4096 then evaluates 4k/10k/20k/40k (and higher) token budgets; reported SOTA comparisons depend on chosen operating points.
  • Upsampling hierarchy (L stages, split factor r, channel halving)
    Point-shuffle expands Ml to r×Ml with Cl+1=Cl/2; these discrete design choices control densification capacity and are not derived.
  • KNN neighborhood size K
    AVS-Conv aggregates K nearest supports; locality–quality tradeoff is a free architectural hyperparameter.
  • Loss weights λ_recon and λ_KL
    Total loss mixes reconstruction and KL terms with coefficients that must be set by the authors; values not fully pinned in text.
  • Progressive crop training schedule = 300k + 100k steps
    300k global steps then 100k localized chunk fine-tuning with near-surface-only SDF supervision strongly affects high-frequency recovery.
axioms (5)
  • domain assumption A continuous SDF (or truncated SDF) plus iso-surface extraction is an adequate target for high-fidelity shape reconstruction.
    Sec. 3–4 and training objectives define success via TSDF regression and mesh metrics; open surfaces and sharp features are later admitted as hard under this formulation.
  • domain assumption Anchored VecSet tokens with explicit active-voxel centers supply useful spatial locality for set-based VAEs (Lattice-style).
    Encoder builds on Lattice anchors (Sec. 4.2); without this prior representation choice the hierarchical densification story does not start.
  • ad hoc to paper Local Euclidean KNN neighborhoods in 3D are a sufficient receptive field for encoding and decoding geometry when multi-scale fusion is added.
    AVS-Conv replaces global cross-attention entirely (Sec. 4.2–4.3); multi-scale fusion is introduced specifically to patch overly-local artifacts.
  • ad hoc to paper External off-the-shelf active-voxel anchors (e.g., HY3D2.1) are acceptable and representative inputs to the encoder.
    Sec. 4.2 samples C from an external generative model’s active voxels rather than learning anchors purely inside the VAE.
  • domain assumption Standard VAE ELBO-style training with KL toward N(0,I) yields latents suitable for downstream latent diffusion (motivating context).
    Introduction frames VAE quality as the bottleneck for latent diffusion; paper itself evaluates reconstruction, not full generative FID-style 3D metrics.
invented entities (3)
  • AVS-Conv no independent evidence
    purpose: Geometry-aware local query-to-support aggregation with dynamic MLP weights on relative offsets, residual mean/identity path, and FFN refinement.
    Core operator replacing global cross-attention in encoder, upsampling, and decoding; evidence is internal ablations/latency tables, not independent theory.
  • Hierarchical Point-Shuffle Upsampling (PSU) for anchored VecSets no independent evidence
    purpose: Progressively densify compact anchored latents by feature shuffle + predicted geometric offsets + local refinement.
    Primary mechanism claimed to close the set-vs-voxel fidelity gap; ablation shows largest drop when removed.
  • Multi-Scale Query Decoding (MSQ) no independent evidence
    purpose: Fuse AVS-Conv features from coarse-to-fine latent scales before SDF MLP head.
    Introduced to supply long-range context lost under purely local finest-scale decoding; supported by milder ablation gains.

pith-pipeline@v1.2.0-grok45-kimik3 · 19280 in / 4184 out tokens · 75956 ms · 2026-07-31T14:56:38.626013+00:00 · methodology

0 comments
read the original abstract

High-fidelity 3D generative modeling increasingly relies on the latent diffusion paradigm, where the reconstruction quality of the underlying 3D VAE becomes a primary bottleneck. Existing approaches largely follow two paradigms: sparse voxel-based representations achieve strong reconstruction quality but incur significant memory and computational overhead, while set-based representations are compact and continuous yet typically lag in fidelity due to latent sparsity and excessive global smoothness. We propose MSVS-VAE, a hierarchical set-based VAE that closes this fidelity gap without sacrificing compactness. Our key idea is to progressively densify anchored VecSet latents via hierarchical point-shuffle upsampling, increasing spatial capacity for fine-grained geometry modeling. To efficiently decode from the densified hierarchy, we replace global cross-attention with AVS-Conv, a geometry-aware local aggregation operator operating within local neighborhoods rather than the exhaustive latent set. We further introduce multi-scale query decoding to fuse coarse-to-fine latent features, where coarse scales provide stable global context, and fine scales refine localized geometry, reducing artifacts from overly local receptive fields. Extensive experiments on Objaverse, ABO, and in-the-wild benchmarks demonstrate that MSVS-VAE consistently outperforms prior set-based and voxel-based VAEs, delivering approximately 10x faster decoding than prior set-based methods and approximately 10x higher compactness than voxel-based baselines.

Figures

Figures reproduced from arXiv: 2607.24436 by Dehao Hao, Dongyu Yan, Kaiyi Zhang, Lingting Zhu, Li Yuan, Long Quan, Runze Zhang, Tanghui Jia, Weikai Chen, Xiangjun Gao, Xin Wang, Yingda Yin, Zeyu Hu.

Figure 1
Figure 1. Figure 1: MSVS-VAE can achieve high-fidelity reconstruction using a compact vecset￾based representation from a set of surface points. the latent diffusion paradigm [8,35], the performance of the 3D Variational Au￾toencoder (VAE) [15, 49] has become the primary bottleneck. A desirable 3D VAE must satisfy two conflicting requirements: it must be compact enough to allow efficient diffusion generative modeling, yet expr… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed MSVS-VAE. The encoder extracts compact la￾tents from surface points via AVS-Conv and self-attention blocks, while the decoder progressively upsamples the anchored latents and predicts SDF using a multi-scale query decoding module. 4.2 MSVS-VAE Encoder Given an input point cloud X = {(xn, an)} N n=1, where xn ∈ R 3 is the 3D position and an ∈ R 3 is the surface normal, we first extr… view at source ↗
Figure 3
Figure 3. Figure 3: Illustrations of the three core modules in MSVS-VAE. (a) AVS-Conv: ag￾gregates local support features for each query via dynamic KNN-weighted convolution and a feed-forward residual block. (b) Point-Shuffle Upsample: expands each latent into r child latents by splitting both its geometry and feature, followed by AVS-Conv refinement. (c) Multi-Scale Query Decoding: fuses features independently queried from … view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison of VAE reconstruction. Best viewed with zoom-in [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative ablation results of the proposed components. Removing MSQ re￾sults in noisier surfaces (Please zoom in for more details), while removing PSU signif￾icantly degrades geometric reconstruction. The full model achieves smoother surfaces and more faithful reconstruction. 5.4 Ablation Studies Multi-scale Query Decoding (MSQ). As shown in Tab. 3, multi-scale query decoding consistently improves over t… view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative test-time scaling with respect to token budget. Increasing the num￾ber of tokens at inference time progressively improves reconstruction quality, enabling more faithful recovery of fine-grained geometric details. Test-Time Scaling (TTS). We also evaluate the test-time scaling ability of our MSVS-VAE by varying the latent set size at inference time. We report reconstruction metrics with 4K, 10K,… view at source ↗
Figure 7
Figure 7. Figure 7: Reconstruction quality improves as the token budget increases. Our method consistently outperforms Lattice across token scales and achieves strong performance with far fewer tokens than voxel-based methods. (Evaluated on in-the-wild benchmark) state-of-the-art reconstruction performance across multiple benchmarks, while remaining significantly more compact and efficient than voxel-based alterna￾tives. Thes… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

54 extracted references · 12 linked inside Pith

  1. [1]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Barron, J.T., Mildenhall, B., Tancik, M., Hedman, P., Martin-Brualla, R., Srini- vasan, P.P.: Mip-nerf: A multiscale representation for anti-aliasing neural radiance fields. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 5855–5864 (2021)

    2021

  2. [2]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Chen, R., Zhang, J., Liang, Y., Luo, G., Li, W., Liu, J., Li, X., Long, X., Feng, J., Tan, P.: Dora: Sampling and benchmarking for 3d shape variational auto-encoders. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 16251–16261 (2025)

    2025

  3. [3]

    arXiv preprint arXiv:2507.17745 (2025)

    Chen, Y., Li, Z., Wang, Y., Zhang, H., Li, Q., Zhang, C., Lin, G.: Ultra3d: Efficient and high-fidelity 3d generation with part attention. arXiv preprint arXiv:2507.17745 (2025)

    Pith/arXiv arXiv 2025

  4. [4]

    ACM Transactions on Graphics (TOG)41(4), 1–13 (2022)

    Chen, Z., Tagliasacchi, A., Funkhouser, T., Zhang, H.: Neural dual contouring. ACM Transactions on Graphics (TOG)41(4), 1–13 (2022)

    2022

  5. [5]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Collins, J., Goel, S., Deng, K., Luthra, A., Xu, L., Gundogdu, E., Zhang, X., Vicente, T.F.Y., Dideriksen, T., Arora, H., et al.: Abo: Dataset and benchmarks for real-world 3d object understanding. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 21126–21136 (2022) 16 D. Hao et al

    2022

  6. [6]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Deitke, M., Schwenk, D., Salvador, J., Weihs, L., Michel, O., VanderBilt, E., Schmidt, L., Ehsani, K., Kembhavi, A., Farhadi, A.: Objaverse: A universe of annotated 3d objects. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 13142–13153 (2023)

    2023

  7. [7]

    In: Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision

    Deng, K., Liu, H.T.D., Zhu, Y., Sun, X., Shang, C., Bhat, K.S., Ramanan, D., Zhu, J.Y., Agrawala, M., Zhou, T.: Efficient autoregressive shape generation via octree-based adaptive tokenization. In: Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision. pp. 11685–11696 (2025)

    2025

  8. [8]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Esser, P., Rombach, R., Ommer, B.: Taming transformers for high-resolution image synthesis. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 12873–12883 (2021)

    2021

  9. [9]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    He, X., Zou, Z.X., Chen, C.H., Guo, Y.C., Liang, D., Yuan, C., Ouyang, W., Cao, Y.P., Li, Y.: Sparseflex: High-resolution and arbitrary-topology 3d shape modeling. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 14822–14833 (2025)

    2025

  10. [10]

    In: SIGGRAPH Asia 2022 conference papers

    Hui, K.H., Li, R., Hu, J., Fu, C.W.: Neural wavelet-domain diffusion for 3d shape generation. In: SIGGRAPH Asia 2022 conference papers. pp. 1–9 (2022)

    2022

  11. [11]

    arXiv preprint arXiv:2506.15442 (2025)

    Hunyuan3D, T., Yang, S., Yang, M., Feng, Y., Huang, X., Zhang, S., He, Z., Luo, D., Liu, H., Zhao, Y., et al.: Hunyuan3d 2.1: From images to high-fidelity 3d assets with production-ready pbr material. arXiv preprint arXiv:2506.15442 (2025)

    Pith/arXiv arXiv 2025

  12. [12]

    In: Proceedings of the 29th annual conference on Computer graphics and interactive techniques

    Ju, T., Losasso, F., Schaefer, S., Warren, J.: Dual contouring of hermite data. In: Proceedings of the 29th annual conference on Computer graphics and interactive techniques. pp. 339–346 (2002)

    2002

  13. [13]

    arXiv preprint arXiv:2305.02463 (2023)

    Jun, H., Nichol, A.: Shap-e: Generating conditional 3d implicit functions. arXiv preprint arXiv:2305.02463 (2023)

    Pith/arXiv arXiv 2023

  14. [14]

    ACM Trans

    Kerbl, B., Kopanas, G., Leimkühler, T., Drettakis, G., et al.: 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph.42(4), 139–1 (2023)

    2023

  15. [15]

    arXiv preprint arXiv:1312.6114 (2013)

    Kingma, D.P., Welling, M.: Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114 (2013)

    Pith/arXiv arXiv 2013

  16. [16]

    arXiv preprint arXiv:2506.16504 (2025)

    Lai, Z., Zhao, Y., Liu, H., Zhao, Z., Lin, Q., Shi, H., Yang, X., Yang, M., Yang, S., Feng, Y., et al.: Hunyuan3d 2.5: Towards high-fidelity 3d assets generation with ultimate details. arXiv preprint arXiv:2506.16504 (2025)

    Pith/arXiv arXiv 2025

  17. [17]

    arXiv preprint arXiv:2512.03052 (2025)

    Lai, Z., Zhao, Y., Zhao, Z., Liu, H., Lin, Q., Huang, J., Guo, C., Yue, X.: Lattice: Democratize high-fidelity 3d generation at scale. arXiv preprint arXiv:2512.03052 (2025)

    arXiv 2025

  18. [18]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Lai, Z., Zhao, Y., Zhao, Z., Liu, H., Wang, F., Shi, H., Yang, X., Lin, Q., Huang, J., Liu, Y., et al.: Unleashing vecset diffusion model for fast shape generation. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 2523–2533 (2025)

    2025

  19. [19]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Li, W., Liu, J., Yan, H., Chen, R., Liang, Y., Chen, X., Tan, P., Long, X.: Crafts- man3d: High-fidelity mesh generation with 3d native diffusion and interactive ge- ometry refiner. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 5307–5317 (2025)

    2025

  20. [20]

    arXiv preprint arXiv:2505.07747 (2025)

    Li, W., Zhang, X., Sun, Z., Qi, D., Li, H., Cheng, W., Cai, W., Wu, S., Liu, J., Wang, Z., et al.: Step1x-3d: Towards high-fidelity and controllable generation of textured 3d assets. arXiv preprint arXiv:2505.07747 (2025)

    Pith/arXiv arXiv 2025

  21. [21]

    IEEE Transactions on Pattern Analysis and Machine Intel- ligence (2025) MSVS-VAE 17

    Li, Y., Zou, Z.X., Liu, Z., Wang, D., Liang, Y., Yu, Z., Liu, X., Guo, Y.C., Liang, D., Ouyang, W., et al.: Triposg: High-fidelity 3d shape synthesis using large-scale rectified flow models. IEEE Transactions on Pattern Analysis and Machine Intel- ligence (2025) MSVS-VAE 17

    2025

  22. [22]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Li, Z., Müller, T., Evans, A., Taylor, R.H., Unberath, M., Liu, M.Y., Lin, C.H.: Neuralangelo: High-fidelity neural surface reconstruction. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 8456–8465 (2023)

    2023

  23. [23]

    arXiv preprint arXiv:2505.14521 (2025)

    Li, Z., Wang, Y., Zheng, H., Luo, Y., Wen, B.: Sparc3d: Sparse representa- tion and construction for high-resolution 3d shapes modeling. arXiv preprint arXiv:2505.14521 (2025)

    Pith/arXiv arXiv 2025

  24. [24]

    In: Seminal graphics: pioneering efforts that shaped the field, pp

    Lorensen, W.E., Cline, H.E.: Marching cubes: A high resolution 3d surface con- struction algorithm. In: Seminal graphics: pioneering efforts that shaped the field, pp. 347–353 (1998)

    1998

  25. [25]

    arXiv preprint arXiv:2511.04029 (2025)

    Luo, Y., He, X., Pan, C., Chen, Y., Wu, J., Li, Y., Ouyang, W., Hu, Y., Yang, G., Yap, C.: Faithful contouring: Near-lossless 3d voxel representation free from iso-surface. arXiv preprint arXiv:2511.04029 (2025)

    arXiv 2025

  26. [26]

    Commu- nications of the ACM65(1), 99–106 (2021)

    Mildenhall, B., Srinivasan, P.P., Tancik, M., Barron, J.T., Ramamoorthi, R., Ng, R.: Nerf: Representing scenes as neural radiance fields for view synthesis. Commu- nications of the ACM65(1), 99–106 (2021)

    2021

  27. [27]

    ACM transactions on graphics (TOG)41(4), 1–15 (2022)

    Müller,T.,Evans,A.,Schied,C.,Keller,A.:Instantneuralgraphicsprimitiveswith a multiresolution hash encoding. ACM transactions on graphics (TOG)41(4), 1–15 (2022)

    2022

  28. [28]

    Nichol, A., Jun, H., Dhariwal, P., Mishkin, P., Chen, M.: Point-e: A system for generating3dpointcloudsfromcomplexprompts.arXivpreprintarXiv:2212.08751 (2022)

    Pith/arXiv arXiv 2022

  29. [29]

    In: IEEE visualization 2004

    Nielson, G.M.: Dual marching cubes. In: IEEE visualization 2004. pp. 489–496. IEEE (2004)

    2004

  30. [30]

    Nvidia: cubql.https://github.com/NVIDIA/cuBQL(2025), accessed: 2026-06-30

    2025

  31. [31]

    In: Acm Siggraph 2005 Courses, pp

    Ohtake, Y., Belyaev, A., Alexa, M., Turk, G., Seidel, H.P.: Multi-level partition of unity implicits. In: Acm Siggraph 2005 Courses, pp. 173–es (2005)

    2005

  32. [32]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Qian, G., Abualshour, A., Li, G., Thabet, A., Ghanem, B.: Pu-gcn: Point cloud upsampling using graph convolutional networks. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 11683–11692 (2021)

    2021

  33. [33]

    In: International conference on machine learning

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al.: Learning transferable visual models from natural language supervision. In: International conference on machine learning. pp. 8748–8763. PmLR (2021)

    2021

  34. [34]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Ren, X., Huang, J., Zeng, X., Museth, K., Fidler, S., Williams, F.: Xcube: Large- scale 3d generative modeling using sparse voxel hierarchies. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 4209–4219 (2024)

    2024

  35. [35]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 10684–10695 (2022)

    2022

  36. [36]

    Seed,B.:Seed3d1.0:Fromimagestohigh-fidelitysimulation-ready3dassets(2025)

    2025

  37. [37]

    ACM Transactions on Graphics (ToG)42(4), 1–16 (2023)

    Shen, T., Munkberg, J., Hasselgren, J., Yin, K., Wang, Z., Chen, W., Gojcic, Z., Fidler, S., Sharp, N., Gao, J.: Flexible isosurface extraction for gradient-based mesh optimization. ACM Transactions on Graphics (ToG)42(4), 1–16 (2023)

    2023

  38. [38]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Shi, W., Caballero, J., Huszár, F., Totz, J., Aitken, A.P., Bishop, R., Rueckert, D., Wang, Z.: Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 1874–1883 (2016) 18 D. Hao et al

    2016

  39. [39]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Wang, T., Zhang, B., Zhang, T., Gu, S., Bao, J., Baltrusaitis, T., Shen, J., Chen, D., Wen, F., Chen, Q., et al.: Rodin: A generative model for sculpting 3d digital avatars using diffusion. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 4563–4573 (2023)

    2023

  40. [40]

    In: Proceed- ings of the IEEE/CVF conference on computer vision and pattern recognition

    Wu, L., Wang, D., Gong, C., Liu, X., Xiong, Y., Ranjan, R., Krishnamoorthi, R., Chandra, V., Liu, Q.: Fast point cloud generation with straight flows. In: Proceed- ings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 9445–9454 (2023)

    2023

  41. [41]

    Advances in Neural Information Processing Systems37, 121859–121881 (2024)

    Wu, S., Lin, Y., Zhang, F., Zeng, Y., Xu, J., Torr, P., Cao, X., Yao, Y.: Direct3d: Scalable image-to-3d generation via 3d latent diffusion transformer. Advances in Neural Information Processing Systems37, 121859–121881 (2024)

    2024

  42. [42]

    arXiv preprint arXiv:2505.17412 (2025)

    Wu, S., Lin, Y., Zhang, F., Zeng, Y., Yang, Y., Bao, Y., Qian, J., Zhu, S., Cao, X., Torr, P., et al.: Direct3d-s2: Gigascale 3d generation made easy with spatial sparse attention. arXiv preprint arXiv:2505.17412 (2025)

    Pith/arXiv arXiv 2025

  43. [43]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Wu, W., Qi, Z., Li, F.: Pointconv: Deep convolutional networks on 3d point clouds. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 9613–9622 (2019)

    2019

  44. [44]

    arXiv preprint arXiv:2512.14692 (2025)

    Xiang, J., Chen, X., Xu, S., Wang, R., Lv, Z., Deng, Y., Zhu, H., Dong, Y., Zhao, H., Yuan, N.J., et al.: Native and compact structured latents for 3d generation. arXiv preprint arXiv:2512.14692 (2025)

    Pith/arXiv arXiv 2025

  45. [45]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Xiang, J., Lv, Z., Xu, S., Deng, Y., Wang, R., Zhang, B., Chen, D., Tong, X., Yang, J.: Structured 3d latents for scalable and versatile 3d generation. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 21469–21480 (2025)

    2025

  46. [46]

    In: Computer Graphics Forum

    Xiong, B., Wei, S.T., Zheng, X.Y., Cao, Y.P., Lian, Z., Wang, P.S.: Octfusion: Octree-based diffusion models for 3d shape generation. In: Computer Graphics Forum. vol. 44, p. e70198. Wiley Online Library (2025)

    2025

  47. [47]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Yang, G., Huang, X., Hao, Z., Liu, M.Y., Belongie, S., Hariharan, B.: Pointflow: 3d point cloud generation with continuous normalizing flows. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 4541–4550 (2019)

    2019

  48. [48]

    Advances in Neural Information Processing Systems35, 21871–21885 (2022)

    Zhang, B., Nießner, M., Wonka, P.: 3dilg: Irregular latent grids for 3d generative modeling. Advances in Neural Information Processing Systems35, 21871–21885 (2022)

    2022

  49. [49]

    ACM Transactions On Graphics (TOG)42(4), 1–16 (2023)

    Zhang, B., Tang, J., Niessner, M., Wonka, P.: 3dshape2vecset: A 3d shape repre- sentation for neural fields and generative diffusion models. ACM Transactions On Graphics (TOG)42(4), 1–16 (2023)

    2023

  50. [50]

    arXiv preprint arXiv:2403.19655 (2024)

    Zhang, B., Cheng, Y., Yang, J., Wang, C., Zhao, F., Tang, Y., Chen, D., Guo, B.: Gaussiancube: A structured and explicit radiance representation for 3d generative modeling. arXiv preprint arXiv:2403.19655 (2024)

    Pith/arXiv arXiv 2024

  51. [51]

    ACM Transactions on Graphics (TOG)43(4), 1–20 (2024)

    Zhang, L., Wang, Z., Zhang, Q., Qiu, Q., Pang, A., Jiang, H., Yang, W., Xu, L., Yu, J.: Clay: A controllable large-scale generative model for creating high-quality 3d assets. ACM Transactions on Graphics (TOG)43(4), 1–20 (2024)

    2024

  52. [52]

    arXiv preprint arXiv:2501.12202 (2025)

    Zhao, Z., Lai, Z., Lin, Q., Zhao, Y., Liu, H., Yang, S., Feng, Y., Yang, M., Zhang, S., Yang, X., et al.: Hunyuan3d 2.0: Scaling diffusion models for high resolution textured 3d assets generation. arXiv preprint arXiv:2501.12202 (2025)

    Pith/arXiv arXiv 2025

  53. [53]

    Advances in neural information processing sys- tems36, 73969–73982 (2023) MSVS-VAE 19

    Zhao, Z., Liu, W., Chen, X., Zeng, X., Wang, R., Cheng, P., Fu, B., Chen, T., Yu, G., Gao, S.: Michelangelo: Conditional 3d shape generation based on shape-image- text aligned latent representation. Advances in neural information processing sys- tems36, 73969–73982 (2023) MSVS-VAE 19

    2023

  54. [54]

    ACM Transactions on Graphics (ToG)42(4), 1–13 (2023)

    Zheng, X.Y., Pan, H., Wang, P.S., Tong, X., Liu, Y., Shum, H.Y.: Locally atten- tional sdf diffusion for controllable 3d shape generation. ACM Transactions on Graphics (ToG)42(4), 1–13 (2023)

    2023