ShapeShifter: 3D Variations Using Multiscale and Sparse Point-Voxel Diffusion

Mathieu Desbrun; Matthew Fisher; Nissim Maruani; Pierre Alliez; Wang Yifan

REVIEW 3 major objections 6 minor 129 references

ShapeShifter: 3D Variations Using Multiscale and Sparse Point-Voxel Diffusion

T0 review · 3 major / 6 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a single 3D exemplar, encoded as colored oriented points on a sparse voxel grid and learned by a multiscale diffusion hierarchy, can generate detailed shape variations after about 12 minutes of training and roughly…

desk verdict Solid single-exemplar 3D diffusion adaptation; worth refereeing, but the Sin3DM comparison needs a same-reference rerun before the quality claim is taken at face value. read the letter →

arxiv 2502.02187 v2 pith:LBEUIZTJ submitted 2025-02-04 cs.CV cs.AI

Nissim Maruani , Wang Yifan , Matthew Fisher , Pierre Alliez , Mathieu Desbrun This is my paper · ORCID

classification cs.CVcs.AI

keywords single-exemplar3Dgenerationdiffusionmodelssparsevoxelgridspointandnormalfeaturesmultiscalegenerativemodelingshapevariationgeometrydetailpreservationinteractiveediting

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

ShapeShifter sets out to prove that a single 3D mesh is enough to train a generative model that produces new variations of that mesh, provided the geometry is represented as colored, oriented points stored in a sparse voxel grid and processed by a multiscale diffusion network. The paper claims this combination preserves sharp geometric details that signed-distance-field and volumetric single-exemplar methods smooth away, while cutting training from hours to about 12 minutes and generation to about 10 seconds on a consumer GPU. The appeal is practical: artists and designers could turn one high-quality asset into a family of editable variants without curating a dataset, and the point-based output supports direct editing and interactive use.

What carries the argument

The load-bearing mechanism is a multiscale diffusion model running on a sparse voxel grid whose active cells carry a 10-dimensional explicit feature: point offset $(p_x,p_y,p_z)$, normal $(n_x,n_y,n_z)$, color $(c_r,c_g,c_b)$, and a surface mask $m$. This representation is what allows the model to learn from one exemplar: small receptive fields over local crops supply the shared internal statistics that make single-example training work, while sparse storage and convolutions keep memory and time low, and the explicit oriented points preserve high-frequency geometry that volumetric losses blur. A learned per-level upsampler replaces the bilinear upsampler of prior single-image work, and the forward process mixes the clean upsampled grid with the ground-truth level through a mixing schedule $\gamma(t)$, so each level both denoises and deblurs; inactive voxels in the ground truth are flooded with neighboring feature values so sparse-grid operations remain well-defined.

What would settle it

Generate twenty outputs from a single exemplar that contains a clear non-local structure, such as a house with a chimney, and measure the pairwise Chamfer distances among outputs and to the input; if for every seed the maximum distance between any two outputs stays below a few percent of the exemplar's bounding-box diagonal, the model is reproducing near-copies and the claim that it creates novel variations fails.

Watch

Extended reading notes

Core claim

The central claim is that explicit surface geometry, one point sample with its normal, color, and a surface mask per active voxel, together with sparse-voxel convolutions and a coarse-to-fine diffusion hierarchy, is enough to learn a single-exemplar 3D generator that beats prior methods on detail fidelity while being far cheaper. Each of five levels, from a $16^3$ to a $256^3$ grid, trains an independent diffusion model with a small receptive field over random crops of that level, so the network learns the exemplar's internal patch statistics rather than its global layout; a learned upsampler carries the coarse solution to the next level, and a pruning step removes voxels whose predicted mask is negative. The authors argue that this is why generated variants keep sharp corners, thin structures, and texture-aligned geometry that methods based on signed distance fields, triplanes, or plenoxels lose, and they report quality scores roughly four times better than the closest baseline while matching its diversity. They also show the same pipeline handles open surfaces by swapping the final meshing step, and that the explicit point output makes copy-paste editing and level-of-detail control straightforward.

Load-bearing premise

The load-bearing premise is that the statistical regularities inside one 3D exemplar, seen through small local crops, are enough to generate genuinely new but plausible variations; if single-exemplar learning can only reproduce near-copies, the variation claim fails, and the paper itself acknowledges the variation range is limited.

Editorial extensions

If this is right

A single textured mesh becomes a training set: with roughly 12 minutes of training on one consumer GPU, any one asset can seed a family of generated variants, making per-asset generators practical in production pipelines.
Details that previous methods rounded off, such as sharp corners, thin walls, and small protrusions, survive in the output because the supervision is explicit point positions and normals rather than a smoothed field.
Output at every level is already a colored oriented point set, so a designer can visualize, edit, or copy-paste parts of the shape at coarse resolution and get a new sample in about a second, without meshing or marching cubes.
Open surfaces are handled by the same trained model; only the final meshing operator changes, which removes a limitation of signed-distance-field-based single-exemplar generators.
Because levels are trained in parallel, training wall-clock time is set by the slowest level, and inference time can be traded against detail by stopping early at a chosen level.

Reading between the lines

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

If the representation scales as the paper's final remarks suggest, the same colored-point sparse-voxel encoding could replace signed-distance or triplane latents in large-dataset 3D diffusion, reducing memory cost and improving sharp-feature fidelity; this is an extension the paper does not demonstrate.
The small receptive field limits outputs to local rearrangements, so augmenting the hierarchy with random rotations, reflections, or non-uniform crop scaling could widen the variation range without any new data; a controlled study of such augmentations is a natural next experiment.
Because the reported quality metric voxelizes outputs at $256^3$, it may systematically undervalue the sub-voxel sharpness that is the method's main qualitative advantage; a point-based or high-resolution mesh distance would be a fairer test of the paper's central claim.
The sub-second generation at the coarse level suggests an interactive design loop in which an artist samples many rough variants and refines only the promising ones; testing this as a user interface, rather than as batch generation, is a plausible practical extension.

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.

Desk Editor's Note

read the letter

Colleague,

Read ShapeShifter. It's a competent and useful adaptation of SinDDM to single-exemplar 3D generation, swapping 2D patches for sparse voxel grids with explicit point, normal, and color features. The new pieces—the learned level-specific upsampler, QEM-aware coarse-to-fine feature extraction, and the sparse fVDB backbone—are well motivated, and the speed numbers (12 min training, ~10 s inference on a 3080) are believable given the architecture. The ablation against trilinear upsampling and SDF features supports the design choices. The open-surface handling via oriented points is a genuine plus. For its niche, this is a solid engineering contribution.

The soft spots are mostly in the evaluation, and the paper is upfront about several of them. Only 7-8 exemplars, point estimates without error bars, SSFID acknowledged to miss high-frequency detail, and the diversity metric acknowledged to reward artifacts. Those are limitations, not fatal flaws.

The bigger issue is the Sin3DM comparison. The stress-test note is correct: the paper compares Sin3DM's outputs against the unprocessed reference mesh, while Sin3DM was trained on watertight-preprocessed geometry that inflates thin details. The supplementary explains this choice—they wanted to measure against the original exemplar because preprocessing degrades it—but that means the reported SSFID gap conflates Sin3DM's preprocessing degradation with its generative quality. A fair comparison would either retrain Sin3DM on the unprocessed mesh or report both references. Without that, the quantitative claim of superiority is not apples-to-apples. The qualitative figures are suggestive but not decisive.

None of this undermines the core method. The central mechanism—small-receptive-field diffusion over local crops of a sparse explicit-feature grid—is a straightforward and sensible adaptation, and the paper does not oversell it. The limitation on variation range is stated honestly. The citation pattern looks fine; SinDDM, fVDB, and the single-exemplar 3D baselines are all properly credited.

Who gets value: researchers working on exemplar-based 3D generation or on sparse 3D diffusion will want this. It deserves a serious referee; the evaluation issues are fixable with a same-reference comparison and more shapes/error bars, and the method itself is reproducible in principle (code link is mentioned on the project page, though not in the manuscript text).

Recommendation: send to peer review, but make the authors rerun the Sin3DM comparison under matched conditions and tighten the metrics before acceptance.

Referee Report

3 major / 6 minor

Summary. ShapeShifter proposes a single-exemplar 3D generative model. The input mesh is converted into multiscale sparse voxel grids whose active voxels carry 10-dimensional features: point offset, normal, color, and a surface mask. A hierarchy of diffusion models with small receptive fields is trained level-by-level, with learned upsamplers between levels; training can be parallelized, and inference is coarse-to-fine with pruning. Outputs are colored oriented point clouds that can be meshed via Poisson reconstruction or APSS. The paper claims better preservation of fine geometric details than SDF- and volume-based single-exemplar baselines (Sin3DM, Sin3DGen, SSG), training in about 12 minutes and inference in about 10.7 seconds on one RTX 3080, support for open surfaces, and interactive editing. Quantitative evaluation uses SSFID and pairwise IoU on eight textured exemplars, supplemented by qualitative comparisons and ablations.

Significance. If substantiated, the paper is a practically useful contribution: it significantly reduces the training cost of single-exemplar 3D generation, provides a compact explicit representation that supports editing and open surfaces, and the authors release code. The integration of sparse voxel convolutions with explicit point/normal/color features is a sensible and well-motivated design. However, the significance is tempered by the narrow evaluation: the headline quantitative advantage rests on a baseline comparison that is not controlled for preprocessing, and the metrics are acknowledged to be blind to the very high-frequency details the paper emphasizes. The method's generation range is also explicitly limited in Sec. 6. These issues are fixable, but the current evidence does not yet fully support the strength of the claims.

major comments (3)

[Sec. 5.1, Table 1; Supplementary Sec. 10] The quantitative comparison with Sin3DM is not apples-to-apples. The supplementary states that Sin3DM preprocesses input meshes to make them watertight, which inflates thin details and degrades the exemplar, and that the authors did not retrain Sin3DM on the unprocessed exemplar, instead comparing its outputs to the unprocessed reference. Consequently, the Sin3DM SSFID scores in Table 1 conflate preprocessing degradation with generative quality, and the table does not by itself establish that ShapeShifter preserves fine details better. This is load-bearing because the paper's central claim is geometric fidelity relative to prior work. Please provide a controlled comparison: retrain or fine-tune the baselines on the same unprocessed exemplars, or evaluate all methods against a common reference that each method actually observed, and quantify the preprocessing gap separately.
[Sec. 5.1] The metric used to support the detail-preservation claim is acknowledged to be blind to high-frequency detail. The paper notes that SSFID overlooks geometric details and that the 256^3 voxelization removes sharp features, yet the headline "clear advantage in quality" in Table 1 is based on this metric. The qualitative evidence in Fig. 2 and the supplementary figures is suggestive but covers only a handful of exemplars. Please add detail-sensitive quantitative evaluations, for example high-resolution point-to-surface distances, normal or curvature statistics, or edge-preservation measures, and report per-exemplar results with variance or multiple seeds, so that the central claim is actually tested by the evidence.
[Sec. 5.1, Table 1; Supplementary Sec. 10] The diversity evaluation is not a reliable measure of useful variation. The paper itself states that pairwise IoU rewards artifacts and that random binary grids score 0.66, which means the diversity columns cannot support the conclusion that ShapeShifter "performs similar to Sin3DM" in diversity. Since the task is generation of variations, please report an artifact-aware diversity measure or a small human assessment, and clearly separate diversity from noise in the interpretation.

minor comments (6)

[Sec. 5.1, Table 2] The sentence "trains each level of the hierarchy of diffusion models in 12 minutes" is inconsistent with Table 2 and the abstract, which report a total of about 12 minutes for encoding and training; please clarify.
[Sec. 3.1] "a 23 average pooling" should read "a 2×2×2 average pooling" for clarity.
[Sec. 3.2] "monotocally decreasing" is a typo for "monotonically decreasing"; please also specify the functional form of the ᾱ(t) and γ(t) schedules used in the experiments.
[Eq. (1); Sec. 3.2] The mask m is described with range [−1, 1]; please state explicitly that it encodes inactive voxels as −1 and active voxels as +1, rather than a continuous mask.
[References] Reference [3] lists "and others" as an author; please complete the citation with the full author list.
[Sec. 5.2, Fig. 4] The text refers to resizing the initial grid G0_T, but Algorithm 2 uses G1_T for the coarsest noisy grid; please unify the notation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the single-exemplar training objective, learned upsampler, and evaluation all use the exemplar as the intended task input rather than as a hidden fit.

full rationale

The paper's derivation chain is self-contained for its stated task: single-exemplar 3D generation. The multiscale diffusion models are trained with a standard reconstruction loss on corrupted crops of the single exemplar (Eqs. 2-3), which is the explicit definition of the task rather than a hidden circular step. The learned upsampler (Eq. 4) is a trained component whose purpose is disclosed, and it does not rename a fitted parameter as a prediction of an independent quantity. The evaluation compares generated point clouds against the exemplar using SSFID and IoU; the paper explicitly acknowledges the metric blindspots and discloses that Sin3DM scores were computed against unprocessed references while Sin3DM used watertight preprocessing (Supplementary Sec. 10). That is an evaluation-fairness caveat, not a circular reduction of the central claim to its inputs. Self-citations to PoNQ for QEM averaging are non-load-bearing because QEM averaging is a standard prior technique and the paper does not rely on an imported uniqueness theorem or on an unverified self-citation to justify its core generative architecture. The central contributions - sparse point-voxel features, parallel multiscale training, and learned upsampling - are implemented and compared rather than derived from the claims themselves. No circular step could be quoted from the paper, so the circularity score is 0.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The method introduces no new physical entities. Its free parameters are standard architectural and training hyperparameters, plus the empirical feature scaling. The main unstated assumptions are the single-exemplar generalization premise (explicitly acknowledged as limiting in Sec. 6) and the adequacy of reconstructing a mesh from the generated oriented point cloud.

free parameters (5)

Feature value ranges = position [-0.5,0.5], normal [-1,1], color [-2,2], mask [-1,1]
Sec. 3.1: ranges 'empirically chosen since feature scale can be important in diffusion models'. These scaling choices affect the diffusion loss and training dynamics.
Number of levels L = 5 levels (16 to 256 resolution)
Sec. 4: default choice, not systematically varied; affects quality, training time, and memory.
Receptive field sizes = 5^3 for M1, 9^3 for rest
Sec. 4: chosen to prevent overfitting to global structure; a key inductive bias for single-exemplar learning that is set by hand.
Sampling timesteps T[l] = T[1]=1000, T[l>1]=300
Sec. 4: chosen to prevent destroying previous-level predictions; affects diversity and inference speed.
Training iterations and dropout = M1: 20k iters, others: 40k iters; upsampler 10k iters; dropout 1% and 5%
Sec. 4: no ablation varying these; they constrain the training budget and the model's capacity.

assumptions (4)

domain assumption The internal statistics of a single exemplar are sufficient to train a generative model that produces plausible variations.
Core assumption inherited from SinGAN/SinDDM literature; the paper cites this as the basis for the method (Sec. 3) and later admits the variation range is limited (Sec. 6).
domain assumption The diffusion reconstruction loss on sparse voxel features (point offsets, normals, colors, mask) learns a meaningful distribution over surface geometry.
The paper assumes that L2 reconstruction of these 10-dimensional features, flooded into inactive voxels, is a good proxy for shape distribution learning (Sec. 3.2).
standard math The fVDB sparse convolution framework correctly and efficiently implements the required operations on sparse grids.
The method relies on fVDB for all sparse convolutions and operations (Sec. 3.1); the paper does not verify the library's correctness beyond reporting results.
domain assumption Poisson reconstruction (or APSS for open surfaces) faithfully converts the generated oriented point cloud into a usable mesh.
Sec. 3.3 assumes the final point cloud can be meshed by Poisson, which is known to smooth or distort some details; the paper uses it without discussion of its limitations.

how reviews work

0 comments

Cite this review

Pith. "Pith review of ShapeShifter: 3D Variations Using Multiscale and Sparse Point-Voxel Diffusion." pith.science (2026). https://pith.science/paper/LBEUIZTJ

@misc{pith2026250202187,
  author       = {Pith},
  title        = {Pith review of: ShapeShifter: 3D Variations Using Multiscale and Sparse Point-Voxel Diffusion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LBEUIZTJ}},
  note         = {Machine review of arXiv:2502.02187}
}

read the original abstract

This paper proposes ShapeShifter, a new 3D generative model that learns to synthesize shape variations based on a single reference model. While generative methods for 3D objects have recently attracted much attention, current techniques often lack geometric details and/or require long training times and large resources. Our approach remedies these issues by combining sparse voxel grids and point, normal, and color sampling within a multiscale neural architecture that can be trained efficiently and in parallel. We show that our resulting variations better capture the fine details of their original input and can handle more general types of surfaces than previous SDF-based methods. Moreover, we offer interactive generation of 3D shape variants, allowing more human control in the design loop if needed.

Figures

Figures reproduced from arXiv: 2502.02187 by the authors.

**Figure 1.** ShapeShifter. Given a 3D exemplar, we propose to train a hierarchical diffusion model to create variations preserving the geometric details and styles of the exemplar. By combining compact yet explicit 3D features (colored, oriented points) with a sparse voxel grid, we shorten training times from hours to minutes, while yielding significantly better geometric quality than prior work. The hierarchical point represe… view at source ↗

**Figure 2.** Geometric details. Our generation captures significantly more geometric details present in the exemplar mesh (leftmost). Prior work, Sin3DGen [50] and Sin3DM [107], operates with plenoxels and neural radiance fields encoded in single-resolution triplane features, respectively, which lack the capability to sufficiently represent and supervise high-resolution geometric details. In contrast, our method employs a colore… view at source ↗

**Figure 3.** Multiscale diffusion on sparse voxel grid. We start from noise ϵ∼ C (0, I) at the coarsest level l= 1, and obtain the 3D feature grid G l through reverse diffusion. Each subsequent level uses the output of the previous level. Inactive voxels are first pruned, then upsampled with a level-specific upsampler U l . The upsampled grid G˜ l is subsequently noised and passed through the diffusion model to obtain a clean… view at source ↗

Figures from the paper (10 more)

**Figure 5.** Figure 5: demonstrates that a generated output can be further edit by copy-and-pasting parts of the output within one of the levels of the multiscale description of the shape, here to remove windows or adding a bay window. While existing works can offer similar capabilities, th…

**Figure 6.** Figure 6: Open surfaces. Our oriented points representation also handles open surfaces by simply using APSS [26] to mesh the generated point set, while it is challenging for existing methods. 5.4. Ablation studies Learned Upsampler. We demonstrate the benefit of our learned ups…

**Figure 7.** Figure 7: Ablation test of upsampling. Comparing trilinear interpolation (left) with learned upsampler (right), the interpolation causes artifacts (see circled areas), whereas the learned upsampler provides a more detailed and structurally coherent output. for the reduced featu…

**Figure 8.** Figure 8: Ablation test of features. Comparing SDF (left) with our proposed point and normal features (right) at the same resolution (1283 ) demonstrates that our proposed features produce richer geometric details. The mesh color encodes the normal direction to reflect the diff…

**Figure 9.** Figure 9: Texturing. As explained Sec. 3.3, ultra-high resolution texture can be obtained by applying state-of-the-art AI imageenhancing tools on the texture maps created by our method from the colored point set outputs. 5.5. Texture Augmentation Finally, we show that one can …

**Figure 10.** Figure 10: ShapeShifter. Given a 3D exemplar (left), we train a hierarchical diffusion model to create novel variations that preserve the geometric details and styles of the exemplar (center), whereas a large generative model such as Rodin [119] tends to lose the geometric deta…

**Figure 11.** Figure 11: Visual inspection of SSG [106] results. While SSG can generate 3D outputs with very short inference time, results are typically blobby or overly smooth, with spurious artifacts often visible due to its voxel-based generation process. In contrast, our method generates …

**Figure 12.** Figure 12: QEM-averaging ablation. While QEM-averaging (proposed in [65]) keeps sharp features (like corners or spikes) in place helping our generative approach maintain these local details, a usual averaging would move the “corner” points inwards, increasing the probability of…

**Figure 13.** Figure 13: Samples of our results I. This figure shows a variety of input models and some of the generated variants (both shown without and with texture to facilitate visual inspection) ShapeShifter outputs [PITH_FULL_IMAGE:figures/full_fig_p016_13.png]

**Figure 14.** Figure 14: Samples of our results II. This figure shows input models that were not used in the main paper, and some of the generated variants (both shown without and with texture to facilitate visual inspection) ShapeShifter outputs. Note that the last two examples (vase and pig…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

129 extracted references · 55 canonical work pages

[1]

Photo-realistic floating wood

All about Blender-3D. Photo-realistic floating wood. http s://www.cgtrader.com/free-3d-models/pla nt/other/photo-realistic-floating-wood ,
[2]

Learning representations and generative models for 3d point clouds

Panos Achlioptas, Olga Diamanti, Ioannis Mitliagkas, and Leonidas Guibas. Learning representations and generative models for 3d point clouds. In International Conference on Machine Learning, pages 40–49. PMLR, 2018. 2

2018
[3]

State of the art on diffusion models for visual computing

and others. State of the art on diffusion models for visual computing. In Computer Graphics Forum, page e15063. Wiley Online Library, 2024. 2

2024
[4]

Small town

Pedram Ashoori. Small town. https://www.cgtrad er.com/free-3d-models/exterior/cityscap e/small-town-87b127c8-c991-4063-aa69- e58800686299, 2020. License: Royalty Free. 6

2020
[5]

Patchmatch: A randomized correspon- dence algorithm for structural image editing

Connelly Barnes, Eli Shechtman, Adam Finkelstein, and Dan B Goldman. Patchmatch: A randomized correspon- dence algorithm for structural image editing. ACM Trans. Graph., 28(3):24, 2009. 3

2009
[6]

Learn- ing long-term dependencies with gradient descent is diffi- cult

Yoshua Bengio, Patrice Simard, and Paolo Frasconi. Learn- ing long-term dependencies with gradient descent is diffi- cult. IEEE Transactions on Neural Networks , 5(2):157– 166, 1994. 2

1994
[7]

Blender - a 3D modelling and rendering package

Blender Online Community. Blender - a 3D modelling and rendering package. Blender Foundation, Stichting Blender Foundation, Amsterdam, 2018. 6

2018
[8]

Texfusion: Synthesizing 3d textures with text-guided image diffusion models

Tianshi Cao, Karsten Kreis, Sanja Fidler, Nicholas Sharp, and Kangxue Yin. Texfusion: Synthesizing 3d textures with text-guided image diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 4169–4181, 2023. 4

2023

Show all 129 references

[9]

Industrial building

Lukas Carnota. Industrial building. https://www.cg trader.com/free-3d-models/exterior/offi ce/indusrtial-building , 2015. License: Royalty Free. 6

2015
[10]

Efficient geometry-aware 3d generative adversarial networks

Eric R Chan, Connor Z Lin, Matthew A Chan, Koki Nagano, Boxiao Pan, Shalini De Mello, Orazio Gallo, Leonidas J Guibas, Jonathan Tremblay, Sameh Khamis, et al. Efficient geometry-aware 3d generative adversarial networks. In Proceedings of the IEEE/CVF Conference on Computer Vis...

2022
[11]

Single-stage dif- fusion nerf: A unified approach to 3d generation and recon- struction

Hansheng Chen, Jiatao Gu, Anpei Chen, Wei Tian, Zhuowen Tu, Lingjie Liu, and Hao Su. Single-stage dif- fusion nerf: A unified approach to 3d generation and recon- struction. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 2416–2425, 2023. 3

2023
[12]

Fan- tasia3d: Disentangling geometry and appearance for high- quality text-to-3d content creation

Rui Chen, Yongwei Chen, Ningxin Jiao, and Kui Jia. Fan- tasia3d: Disentangling geometry and appearance for high- quality text-to-3d content creation. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 22246–22256, 2023. 3

2023
[13]

On the Importance of Noise Scheduling for Diffusion Models, 2023

Ting Chen. On the Importance of Noise Scheduling for Diffusion Models, 2023. arXiv:2301.10972. 4

2023 arXiv
[14]

Meshanything: Artist-created mesh gener- ation with autoregressive transformers

Yiwen Chen, Tong He, Di Huang, Weicai Ye, Sijin Chen, Jiaxiang Tang, Xin Chen, Zhongang Cai, Lei Yang, Gang Yu, et al. Meshanything: Artist-created mesh gener- ation with autoregressive transformers. arXiv preprint arXiv:2406.10163, 2024. 2

2024 arXiv
[15]

Learning implicit fields for generative shape modeling

Zhiqin Chen and Hao Zhang. Learning implicit fields for generative shape modeling. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019. 3

2019
[16]

Decor- gan: 3d shape detailization by conditional refinement

Zhiqin Chen, Vladimir G Kim, Matthew Fisher, Noam Aigerman, Hao Zhang, and Siddhartha Chaudhuri. Decor- gan: 3d shape detailization by conditional refinement. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition , pages 15740–15749, 2021. 6

2021
[17]

Sdfusion: Multimodal 3d shape completion, reconstruction, and generation

Yen-Chi Cheng, Hsin-Ying Lee, Sergey Tulyakov, Alexan- der G Schwing, and Liang-Yan Gui. Sdfusion: Multimodal 3d shape completion, reconstruction, and generation. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 4456–4465, 2023. 3

2023
[18]

Objaverse-xl: A universe of 10m+ 3d objects

Matt Deitke, Ruoshi Liu, Matthew Wallingford, Huong Ngo, Oscar Michel, Aditya Kusupati, Alan Fan, Chris- tian Laforte, Vikram V oleti, Samir Yitzhak Gadre, et al. Objaverse-xl: A universe of 10m+ 3d objects. Advances in Neural Information Processing Systems, 36, 2024. 2

2024
[19]

DJMaesen. Cliff. https://sketchfab.com/3d- models/cliff-082da1166a814c6e9c9e6c1b38 159e4e, 2021. License: CC Attribution. 6

2021
[20]

Generating natural im- ages with direct patch distributions matching

Ariel Elnekave and Yair Weiss. Generating natural im- ages with direct patch distributions matching. In European Conference on Computer Vision, pages 544–560. Springer,
[21]

Plenoxels: Radiance fields without neural networks

Sara Fridovich-Keil, Alex Yu, Matthew Tancik, Qinhong Chen, Benjamin Recht, and Angjoo Kanazawa. Plenoxels: Radiance fields without neural networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5501–5510, 2022. 3

2022
[22]

Get3d: A generative model of high quality 3d tex- tured shapes learned from images

Jun Gao, Tianchang Shen, Zian Wang, Wenzheng Chen, Kangxue Yin, Daiqing Li, Or Litany, Zan Gojcic, and Sanja Fidler. Get3d: A generative model of high quality 3d tex- tured shapes learned from images. Advances In Neural In- formation Processing Systems, 35:31841–31854, 2022. 2

2022
[23]

Generative adversarial networks.Com- munications of the ACM, 63(11):139–144, 2020

Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial networks.Com- munications of the ACM, 63(11):139–144, 2020. 2

2020
[24]

Drop the gan: In defense of patches nearest neighbors as single image generative models

Niv Granot, Ben Feinstein, Assaf Shocher, Shai Bagon, and Michal Irani. Drop the gan: In defense of patches nearest neighbors as single image generative models. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13460–13469, 2022. 3

2022
[25]

Generating sequences with recurrent neural networks

Alex Graves. Generating sequences with recurrent neural networks. arXiv preprint arXiv:1308.0850, 2013. 2

2013 arXiv
[26]

Algebraic point set surfaces

Ga ¨el Guennebaud and Markus Gross. Algebraic point set surfaces. In ACM SIGGRAPH 2007 papers , page 23, San Diego California, 2007. ACM. 5, 7

2007
[27]

3dgen: Triplane latent diffusion for textured mesh generation

Anchit Gupta, Wenhan Xiong, Yixin Nie, Ian Jones, and Barlas O ˘guz. 3dgen: Triplane latent diffusion for textured mesh generation. arXiv preprint arXiv:2303.05371, 2023. 3

2023 arXiv
[28]

Diverse generation from a single video made possible

Niv Haim, Ben Feinstein, Niv Granot, Assaf Shocher, Shai Bagon, Tali Dekel, and Michal Irani. Diverse generation from a single video made possible. InEuropean Conference on Computer Vision, pages 491–509. Springer, 2022. 3

2022
[29]

Multiscale texture synthesis

Charles Han, Eric Risser, Ravi Ramamoorthi, and Eitan Grinspun. Multiscale texture synthesis. In ACM SIG- GRAPH 2008 papers, pages 1–8. 2008. 3

2008
[30]

Meshcnn: a network with an edge

Rana Hanocka, Amir Hertz, Noa Fish, Raja Giryes, Shachar Fleishman, and Daniel Cohen-Or. Meshcnn: a network with an edge. ACM Transactions on Graphics (ToG), 38(4):1– 12, 2019. 3

2019
[31]

Deep geometric texture synthesis

Amir Hertz, Rana Hanocka, Raja Giryes, and Daniel Cohen-Or. Deep geometric texture synthesis. ACM Trans. Graph., 39(4), 2020. 3

2020
[32]

Improved techniques for training single-image gans

Tobias Hinz, Matthew Fisher, Oliver Wang, and Stefan Wermter. Improved techniques for training single-image gans. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages 1300–1309,
[33]

Denoising dif- fusion probabilistic models

Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. Advances in Neural Informa- tion Processing systems, 33:6840–6851, 2020. 2, 5

2020
[34]

LRM: Large reconstruction model for single image to 3d

Yicong Hong, Kai Zhang, Jiuxiang Gu, Sai Bi, Yang Zhou, Difan Liu, Feng Liu, Kalyan Sunkavalli, Trung Bui, and Hao Tan. LRM: Large reconstruction model for single image to 3d. In The Twelfth International Conference on Learning Representations, 2024. 3

2024
[35]

2d gaussian splatting for geometrically ac- curate radiance fields

Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao. 2d gaussian splatting for geometrically ac- curate radiance fields. In ACM SIGGRAPH Conference Pa- pers, 2024. 8

2024
[36]

Texgen: Text-guided 3d texture generation with multi-view sampling and resampling

Dong Huo, Zixin Guo, Xinxin Zuo, Zhihao Shi, Juwei Lu, Peng Dai, Songcen Xu, Li Cheng, and Yee-Hong Yang. Texgen: Text-guided 3d texture generation with multi-view sampling and resampling. In European Conference on Computer Vision, pages 352–368. Springer, 2025. 3

2025
[37]

Fighting pillar

ImpJive. Fighting pillar. https://sketchfab.co m/3d- models/fighting- pillar- 14e73d2d 9e8a4981b49d0e6c56d30af5 , 2021. License: CC Attribution-ShareAlike. 6

2021
[38]

libigl, 2023

Alec Jacobson and Daniele Panozzo. libigl, 2023. 5

2023
[39]

Shap-e: Generat- ing conditional 3d implicit functions

Heewoo Jun and Alex Nichol. Shap-e: Generat- ing conditional 3d implicit functions. arXiv preprint arXiv:2305.02463, 2023. 3

2023 arXiv
[40]

Scal- ing up gans for text-to-image synthesis

Minguk Kang, Jun-Yan Zhu, Richard Zhang, Jaesik Park, Eli Shechtman, Sylvain Paris, and Taesung Park. Scal- ing up gans for text-to-image synthesis. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10124–10134, 2023. 4

2023
[41]

3ingan: Learning a 3d generative model from images of a self-similar scene

Animesh Karnewar, Oliver Wang, Tobias Ritschel, and Niloy J Mitra. 3ingan: Learning a 3d generative model from images of a self-similar scene. In International Con- ference on 3D Vision (3DV) , pages 342–352. IEEE, 2022. 3

2022
[42]

Neural 3d mesh renderer

Hiroharu Kato, Yoshitaka Ushiku, and Tatsuya Harada. Neural 3d mesh renderer. In Proceedings of the IEEE Con- ference on Computer Vision and Pattern Recognition, pages 3907–3916, 2018. 3

2018
[43]

Screened poisson surface reconstruction

Michael Kazhdan and Hugues Hoppe. Screened poisson surface reconstruction. ACM Trans. Graph. , 32(3):1–13,
[44]

3d gaussian splatting for real-time radiance field rendering

Bernhard Kerbl, Georgios Kopanas, Thomas Leimk ¨uhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph., 42(4):139–1,
[45]

An introduction to variational autoencoders

Diederik P Kingma, Max Welling, et al. An introduction to variational autoencoders. Foundations and Trends in Ma- chine Learning, 12(4):307–392, 2019. 2

2019
[46]

Sinddm: A single image denoising diffusion model

Vladimir Kulikov, Shahar Yadin, Matan Kleiner, and Tomer Michaeli. Sinddm: A single image denoising diffusion model. In International Conference on Machine Learning, pages 17920–17930. PMLR, 2023. 3, 4

2023
[47]

Akropolis

Choly Kurd. Akropolis. https://www.turbosquid .com/3d-models/acropolis-3ds-free/61088 5, 2021. License: Educational Uses. 6

2021
[48]

Srdiff: Single image super-resolution with diffusion probabilistic models

Haoying Li, Yifan Yang, Meng Chang, Shiqi Chen, Huajun Feng, Zhihai Xu, Qi Li, and Yueting Chen. Srdiff: Single image super-resolution with diffusion probabilistic models. Neurocomputing, 479:47–59, 2022. 4

2022
[49]

Instant3d: Fast text-to-3d with sparse-view generation and large reconstruction model

Jiahao Li, Hao Tan, Kai Zhang, Zexiang Xu, Fujun Luan, Yinghao Xu, Yicong Hong, Kalyan Sunkavalli, Greg Shakhnarovich, and Sai Bi. Instant3d: Fast text-to-3d with sparse-view generation and large reconstruction model. In International Conference on Learning Representations ,
[50]

Patch-based 3d natural scene generation from a single ex- ample

Weiyu Li, Xuelin Chen, Jue Wang, and Baoquan Chen. Patch-based 3d natural scene generation from a single ex- ample. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 16762– 16772, 2023. 1, 2, 3, 6, 7

2023
[51]

Sweet- dreamer: Aligning geometric priors in 2d diffusion for con- sistent text-to-3d

Weiyu Li, Rui Chen, Xuelin Chen, and Ping Tan. Sweet- dreamer: Aligning geometric priors in 2d diffusion for con- sistent text-to-3d. In International Conference on Learning Representations, 2024. 3

2024
[52]

Luciddreamer: Towards high- fidelity text-to-3d generation via interval score matching

Yixun Liang, Xin Yang, Jiantao Lin, Haodong Li, Xiaogang Xu, and Yingcong Chen. Luciddreamer: Towards high- fidelity text-to-3d generation via interval score matching. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6517–6526, 2024. 3

2024
[53]

Magic3d: High- resolution text-to-3d content creation

Chen-Hsuan Lin, Jun Gao, Luming Tang, Towaki Takikawa, Xiaohui Zeng, Xun Huang, Karsten Kreis, Sanja Fidler, Ming-Yu Liu, and Tsung-Yi Lin. Magic3d: High- resolution text-to-3d content creation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogniti...

2023
[54]

One-2-3-45: Any sin- gle image to 3d mesh in 45 seconds without per-shape opti- mization

Minghua Liu, Chao Xu, Haian Jin, Linghao Chen, Mukund Varma T, Zexiang Xu, and Hao Su. One-2-3-45: Any sin- gle image to 3d mesh in 45 seconds without per-shape opti- mization. Advances in Neural Information Processing Sys- tems, 36, 2024. 3

2024
[55]

Zero-1-to-3: Zero-shot one image to 3d object

Ruoshi Liu, Rundi Wu, Basile Van Hoorick, Pavel Tok- makov, Sergey Zakharov, and Carl V ondrick. Zero-1-to-3: Zero-shot one image to 3d object. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 9298–9309, 2023. 3

2023
[56]

Soft rasterizer: A differentiable renderer for image-based 3d reasoning

Shichen Liu, Tianye Li, Weikai Chen, and Hao Li. Soft rasterizer: A differentiable renderer for image-based 3d reasoning. In Proceedings of the IEEE/CVF International Conference on Computer Vision , pages 7708–7717, 2019. 3

2019
[57]

Syncdreamer: Generating multiview-consistent images from a single-view image

Yuan Liu, Cheng Lin, Zijiao Zeng, Xiaoxiao Long, Lingjie Liu, Taku Komura, and Wenping Wang. Syncdreamer: Generating multiview-consistent images from a single-view image. In International Conference on Learning Represen- tations, 2024. 3

2024
[58]

Neural volumes: learning dynamic renderable volumes from im- ages

Stephen Lombardi, Tomas Simon, Jason Saragih, Gabriel Schwartz, Andreas Lehrmann, and Yaser Sheikh. Neural volumes: learning dynamic renderable volumes from im- ages. ACM Transactions on Graphics (TOG), 38(4):1–14,
[59]

Wonder3d: Single image to 3d using cross-domain diffusion

Xiaoxiao Long, Yuan-Chen Guo, Cheng Lin, Yuan Liu, Zhiyang Dou, Lingjie Liu, Yuexin Ma, Song-Hai Zhang, Marc Habermann, Christian Theobalt, et al. Wonder3d: Single image to 3d using cross-domain diffusion. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Patt...

2024
[60]

Marching cubes: A high resolution 3d surface construction algorithm

William E Lorensen and Harvey E Cline. Marching cubes: A high resolution 3d surface construction algorithm. In Seminal graphics: pioneering efforts that shaped the field , pages 347–353. 1998. 8

1998
[61]

Score distillation via reparametrized ddim

Artem Lukoianov, Haitz S ´aez de Oc ´ariz Borde, Kristjan Greenewald, Vitor Campagnolo Guizilini, Timur Bagaut- dinov, Vincent Sitzmann, and Justin Solomon. Score distillation via reparametrized ddim. arXiv preprint arXiv:2405.15891, 2024. 3

2024 arXiv
[62]

Diffusion probabilistic mod- els for 3d point cloud generation

Shitong Luo and Wei Hu. Diffusion probabilistic mod- els for 3d point cloud generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2837–2845, 2021. 3

2021
[63]

Magnific ai: Accelerating scientific research

Magnific AI. Magnific ai: Accelerating scientific research. https://magnific.ai/. Accessed: 2024-11-13. 8

2024
[64]

V oroMesh: Learning Wa- tertight Surface Meshes with V oronoi Diagrams

Nissim Maruani, Roman Klokov, Maks Ovsjanikov, Pierre Alliez, and Mathieu Desbrun. V oroMesh: Learning Wa- tertight Surface Meshes with V oronoi Diagrams. In 2023 IEEE/CVF International Conference on Computer Vision (ICCV), pages 14519–14528, Paris, France, 2023. IEEE. 3

2023
[65]

PoNQ: A Neural QEM-Based Mesh Representation

Nissim Maruani, Maks Ovsjanikov, Pierre Alliez, and Mathieu Desbrun. PoNQ: A Neural QEM-Based Mesh Representation. In 2024 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 3647– 3657, Seattle, W A, USA, 2024. IEEE. 3, 4, 15

2024
[66]

Occupancy net- works: Learning 3d reconstruction in function space

Lars Mescheder, Michael Oechsle, Michael Niemeyer, Se- bastian Nowozin, and Andreas Geiger. Occupancy net- works: Learning 3d reconstruction in function space. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 4460–4470, 2019. 3

2019
[67]

Latent-nerf for shape-guided gen- eration of 3d shapes and textures

Gal Metzer, Elad Richardson, Or Patashnik, Raja Giryes, and Daniel Cohen-Or. Latent-nerf for shape-guided gen- eration of 3d shapes and textures. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12663–12673, 2023. 3

2023
[68]

Nerf: Representing scenes as neural radiance fields for view syn- thesis

Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view syn- thesis. Communications of the ACM, 65(1):99–106, 2021. 3

2021
[69]

Mitchel, Carlos Esteves, and Ameesh Makadia

Thomas W. Mitchel, Carlos Esteves, and Ameesh Makadia. Single mesh diffusion models with field latents for texture generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages 7953–7963, 2024. 3

2024
[70]

Polygen: An autoregressive generative model of 3d meshes

Charlie Nash, Yaroslav Ganin, SM Ali Eslami, and Peter Battaglia. Polygen: An autoregressive generative model of 3d meshes. In International Conference on Machine Learn- ing, pages 7220–7229. PMLR, 2020. 2

2020
[71]

Point-e: A system for generat- ing 3d point clouds from complex prompts

Alex Nichol, Heewoo Jun, Prafulla Dhariwal, Pamela Mishkin, and Mark Chen. Point-e: A system for generat- ing 3d point clouds from complex prompts. arXiv preprint arXiv:2212.08751, 2022. 3

2022 arXiv
[72]

Sinfusion: training diffusion models on a single image or video

Yaniv Nikankin, Niv Haim, and Michal Irani. Sinfusion: training diffusion models on a single image or video. In Proceedings of the 40th International Conference on Ma- chine Learning, pages 26199–26214, 2023. 3, 4

2023
[73]

Self-organising textures

Eyvind Niklasson, Alexander Mordvintsev, Ettore Ran- dazzo, and Michael Levin. Self-organising textures. Distill, 6(2):e00027–003, 2021. 3

2021
[74]

Deepsdf: Learning continuous signed distance functions for shape represen- tation

Jeong Joon Park, Peter Florence, Julian Straub, Richard Newcombe, and Steven Lovegrove. Deepsdf: Learning continuous signed distance functions for shape represen- tation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 165–174,
[75]

PyTorch: An Im- perative Style, High-Performance Deep Learning Library,

Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas K¨opf, Edward Yang, Zach DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu Fa...
[76]

Barron, and Ben Milden- hall

Ben Poole, Ajay Jain, Jonathan T. Barron, and Ben Milden- hall. Dreamfusion: Text-to-3d using 2d diffusion. In Inter- national Conference on Learning Representations, 2023. 3

2023
[77]

Dynamic Point Fields

Sergey Prokudin, Qianli Ma, Maxime Raafat, Julien Valentin, and Siyu Tang. Dynamic Point Fields. In 2023 IEEE/CVF International Conference on Computer Vision (ICCV), pages 7964–7976, 2023. 2

2023
[78]

Pointnet: Deep learning on point sets for 3d classification and segmentation

Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. Pointnet: Deep learning on point sets for 3d classification and segmentation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 652– 660, 2017. 3

2017
[79]

Richdreamer: A gen- eralizable normal-depth diffusion model for detail richness in text-to-3d

Lingteng Qiu, Guanying Chen, Xiaodong Gu, Qi Zuo, Mutian Xu, Yushuang Wu, Weihao Yuan, Zilong Dong, Liefeng Bo, and Xiaoguang Han. Richdreamer: A gen- eralizable normal-depth diffusion model for detail richness in text-to-3d. In Proceedings of the IEEE/CVF Conference on Comput...

2024
[80]

Accelerating 3d deep learning with pytorch3d

Nikhila Ravi, Jeremy Reizenstein, David Novotny, Tay- lor Gordon, Wan-Yen Lo, Justin Johnson, and Georgia Gkioxari. Accelerating 3d deep learning with pytorch3d. arXiv:2007.08501, 2020. 15

2007 arXiv
[81]

Xcube: Large-scale 3d generative modeling using sparse voxel hierarchies

Xuanchi Ren, Jiahui Huang, Xiaohui Zeng, Ken Museth, Sanja Fidler, and Francis Williams. Xcube: Large-scale 3d generative modeling using sparse voxel hierarchies. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4209–4219, 2024. 3

2024
[82]

Scube: Instant large- scale scene reconstruction using voxsplats

Xuanchi Ren, Yifan Lu, Hanxue Liang, Jay Zhangjie Wu, Huan Ling, Mike Chen, Francis Fidler, Sanja annd Williams, and Jiahui Huang. Scube: Instant large- scale scene reconstruction using voxsplats. In Conference on Neural Information Processing Systems, 2024. 3

2024
[83]

Variational infer- ence with normalizing flows

Danilo Rezende and Shakir Mohamed. Variational infer- ence with normalizing flows. In International Conference on Machine Learning, pages 1530–1538. PMLR, 2015. 2

2015
[84]

Texture: Text-guided texturing of 3d shapes

Elad Richardson, Gal Metzer, Yuval Alaluf, Raja Giryes, and Daniel Cohen-Or. Texture: Text-guided texturing of 3d shapes. In ACM SIGGRAPH Conference Proceedings, pages 1–11, 2023. 2, 4

2023
[85]

Seamlessgan: Self-supervised synthesis of tileable texture maps

Carlos Rodriguez-Pardo and Elena Garces. Seamlessgan: Self-supervised synthesis of tileable texture maps. IEEE Transactions on Visualization and Computer Graphics, 29 (6):2914–2925, 2022. 3

2022
[86]

Graf: Generative radiance fields for 3d-aware im- age synthesis

Katja Schwarz, Yiyi Liao, Michael Niemeyer, and Andreas Geiger. Graf: Generative radiance fields for 3d-aware im- age synthesis. Advances in Neural Information Processing Systems, 33:20154–20166, 2020. 3

2020
[87]

Sin- gan: Learning a generative model from a single natural im- age

Tamar Rott Shaham, Tali Dekel, and Tomer Michaeli. Sin- gan: Learning a generative model from a single natural im- age. In Proceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 4570–4580, 2019. 3, 4, 6, 14

2019
[88]

MVDream: Multi-view diffusion for 3d generation

Yichun Shi, Peng Wang, Jianglong Ye, Long Mai, Kejie Li, and Xiao Yang. MVDream: Multi-view diffusion for 3d generation. In International Conference on Learning Representations, 2024. 3

2024
[89]

Ingan: Capturing and retargeting the” dna” of a natural im- age

Assaf Shocher, Shai Bagon, Phillip Isola, and Michal Irani. Ingan: Capturing and retargeting the” dna” of a natural im- age. In Proceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 4492–4501, 2019. 3

2019
[90]

Meshgpt: Generating triangle meshes with decoder-only transformers

Yawar Siddiqui, Antonio Alliegro, Alexey Artemov, Ta- tiana Tommasi, Daniele Sirigatti, Vladislav Rosov, Angela Dai, and Matthias Nießner. Meshgpt: Generating triangle meshes with decoder-only transformers. In Proceedings of the IEEE/CVF Conference on Computer Vision and Patte...

2024
[91]

Deep unsupervised learning using nonequilibrium thermodynamics

Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In International Con- ference on Machine Learning , pages 2256–2265. PMLR,
[92]

Singraf: Learning a 3d generative radiance field for a single scene

Minjung Son, Jeong Joon Park, Leonidas Guibas, and Gor- don Wetzstein. Singraf: Learning a 3d generative radiance field for a single scene. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 8507–8517, 2023. 1, 2, 3

2023
[93]

Denois- ing diffusion implicit models

Jiaming Song, Chenlin Meng, and Stefano Ermon. Denois- ing diffusion implicit models. In International Conference on Learning Representations, 2021. 5

2021
[94]

Lgm: Large multi- view gaussian model for high-resolution 3d content cre- ation

Jiaxiang Tang, Zhaoxi Chen, Xiaokang Chen, Tengfei Wang, Gang Zeng, and Ziwei Liu. Lgm: Large multi- view gaussian model for high-resolution 3d content cre- ation. In European Conference on Computer Vision, pages 1–18. Springer, 2025. 3

2025
[95]

Canyon landscape

Simon Ustal. Canyon landscape. https://sketchfa b.com/3d-models/canyon-landscape-c395e9 eb54ba4f40820ccfb98d3c2832, 2020. License: CC Attribution. 6

2020
[96]

Lion: Latent point diffu- sion models for 3d shape generation

Arash Vahdat, Francis Williams, Zan Gojcic, Or Litany, Sanja Fidler, Karsten Kreis, et al. Lion: Latent point diffu- sion models for 3d shape generation. Advances in Neural Information Processing Systems, 35:10021–10039, 2022. 3

2022
[97]

Pixel recurrent neural networks

A ¨aron Van Den Oord, Nal Kalchbrenner, and Koray Kavukcuoglu. Pixel recurrent neural networks. In Interna- tional Conference on Machine Learning, pages 1747–1756. PMLR, 2016. 2

2016
[98]

Score jacobian chaining: Lifting pretrained 2d diffusion models for 3d generation

Haochen Wang, Xiaodan Du, Jiahao Li, Raymond A Yeh, and Greg Shakhnarovich. Score jacobian chaining: Lifting pretrained 2d diffusion models for 3d generation. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12619–12629, 2023. 3

2023
[99]

Neus: Learning neural implicit surfaces by volume rendering for multi-view recon- struction

Peng Wang, Lingjie Liu, Yuan Liu, Christian Theobalt, Taku Komura, and Wenping Wang. Neus: Learning neural implicit surfaces by volume rendering for multi-view recon- struction. arXiv preprint arXiv:2106.10689, 2021. 6

2021 arXiv
[100]

PF-LRM: Pose-free large reconstruction model for joint pose and shape prediction

Peng Wang, Hao Tan, Sai Bi, Yinghao Xu, Fujun Luan, Kalyan Sunkavalli, Wenping Wang, Zexiang Xu, and Kai Zhang. PF-LRM: Pose-free large reconstruction model for joint pose and shape prediction. In International Confer- ence on Learning Representations, 2024. 3

2024
[101]

Taming mode collapse in score distillation for text-to-3d generation

Peihao Wang, Dejia Xu, Zhiwen Fan, Dilin Wang, Sreyas Mohan, Forrest Iandola, Rakesh Ranjan, Yilei Li, Qiang Liu, Zhangyang Wang, et al. Taming mode collapse in score distillation for text-to-3d generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- ...

2024
[102]

Prolificdreamer: High- fidelity and diverse text-to-3d generation with variational score distillation

Zhengyi Wang, Cheng Lu, Yikai Wang, Fan Bao, Chongx- uan Li, Hang Su, and Jun Zhu. Prolificdreamer: High- fidelity and diverse text-to-3d generation with variational score distillation. Advances in Neural Information Process- ing Systems, 36, 2024. 3

2024
[103]

Meshlrm: Large reconstruction model for high- quality mesh

Xinyue Wei, Kai Zhang, Sai Bi, Hao Tan, Fujun Luan, Valentin Deschaintre, Kalyan Sunkavalli, Hao Su, and Zex- iang Xu. Meshlrm: Large reconstruction model for high- quality mesh. arXiv preprint arXiv:2404.12385, 2024. 3

2024 arXiv
[104]

Synsin: End-to-end view synthesis from a single image

Olivia Wiles, Georgia Gkioxari, Richard Szeliski, and Justin Johnson. Synsin: End-to-end view synthesis from a single image. In Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition , pages 7467–7477, 2020. 3

2020
[105]

fvdb: A deep- learning framework for sparse, large scale, and high perfor- mance spatial intelligence

Francis Williams, Jiahui Huang, Jonathan Swartz, Gergely Klar, Vijay Thakkar, Matthew Cong, Xuanchi Ren, Ruilong Li, Clement Fuji-Tsang, Sanja Fidler, et al. fvdb: A deep- learning framework for sparse, large scale, and high perfor- mance spatial intelligence. ACM Transactions...

2024
[106]

Learning to generate 3d shapes from a single example

Rundi Wu and Changxi Zheng. Learning to generate 3d shapes from a single example. ACM Transactions on Graphics (TOG), 41(6):1–19, 2022. 1, 2, 3, 6, 14

2022
[107]

Sin3DM: Learning a diffusion model from a single 3d tex- tured shape

Rundi Wu, Ruoshi Liu, Carl V ondrick, and Changxi Zheng. Sin3DM: Learning a diffusion model from a single 3d tex- tured shape. In International Conference on Learning Rep- resentations, 2024. 1, 2, 3, 6, 15

2024
[108]

3d shapenets: A deep representation for volumetric shapes

Zhirong Wu, Shuran Song, Aditya Khosla, Fisher Yu, Linguang Zhang, Xiaoou Tang, and Jianxiong Xiao. 3d shapenets: A deep representation for volumetric shapes. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 1912–1920, 2015. 3

1912
[109]

Grm: Large gaussian reconstruction model for efficient 3d reconstruc- tion and generation

Yinghao Xu, Zifan Shi, Wang Yifan, Sida Peng, Ceyuan Yang, Yujun Shen, and Wetzstein Gordon. Grm: Large gaussian reconstruction model for efficient 3d reconstruc- tion and generation. arxiv: 2403.14621, 2024. 3

2024 arXiv
[110]

DMV3d: Denoising multi- view diffusion using 3d large reconstruction model

Yinghao Xu, Hao Tan, Fujun Luan, Sai Bi, Peng Wang, Ji- ahao Li, Zifan Shi, Kalyan Sunkavalli, Gordon Wetzstein, Zexiang Xu, and Kai Zhang. DMV3d: Denoising multi- view diffusion using 3d large reconstruction model. In In- ternational Conference on Learning Representations, 2024. 3

2024
[111]

Pointflow: 3d point cloud generation with continuous normalizing flows

Guandao Yang, Xun Huang, Zekun Hao, Ming-Yu Liu, Serge Belongie, and Bharath Hariharan. Pointflow: 3d point cloud generation with continuous normalizing flows. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 4541–4550, 2019. 2

2019
[112]

Dreammesh: Jointly manipulating and texturing trian- gle meshes for text-to-3d generation

Haibo Yang, Yang Chen, Yingwei Pan, Ting Yao, Zhi- neng Chen, Zuxuan Wu, Yu-Gang Jiang, and Tao Mei. Dreammesh: Jointly manipulating and texturing trian- gle meshes for text-to-3d generation. arXiv preprint arXiv:2409.07454, 2024. 3

2024 arXiv
[113]

Differentiable surface splatting for point-based geometry processing.ACM Transactions on Graphics (TOG), 38(6):1–14, 2019

Wang Yifan, Felice Serena, Shihao Wu, Cengiz ¨Oztireli, and Olga Sorkine-Hornung. Differentiable surface splatting for point-based geometry processing.ACM Transactions on Graphics (TOG), 38(6):1–14, 2019. 3

2019
[114]

Shapegpt: 3d shape generation with a unified multi-modal language model

Fukun Yin, Xin Chen, Chi Zhang, Biao Jiang, Zibo Zhao, Jiayuan Fan, Gang Yu, Taihao Li, and Tao Chen. Shapegpt: 3d shape generation with a unified multi-modal language model. arXiv preprint arXiv:2311.17618, 2023. 2

2023 arXiv
[115]

Paint3d: Paint anything 3d with lighting-less texture diffu- sion models

Xianfang Zeng, Xin Chen, Zhongqi Qi, Wen Liu, Zibo Zhao, Zhibin Wang, Bin Fu, Yong Liu, and Gang Yu. Paint3d: Paint anything 3d with lighting-less texture diffu- sion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4252– 4262...

2024
[116]

3dilg: Ir- regular latent grids for 3d generative modeling.Advances in Neural Information Processing Systems, 35:21871–21885,

Biao Zhang, Matthias Nießner, and Peter Wonka. 3dilg: Ir- regular latent grids for 3d generative modeling.Advances in Neural Information Processing Systems, 35:21871–21885,
[117]

3dshape2vecset: A 3d shape representation for neural fields and generative diffusion models

Biao Zhang, Jiapeng Tang, Matthias Niessner, and Peter Wonka. 3dshape2vecset: A 3d shape representation for neural fields and generative diffusion models. ACM Trans- actions on Graphics (TOG), 42(4):1–16, 2023. 3

2023
[118]

Gs-lrm: Large reconstruction model for 3d gaussian splatting

Kai Zhang, Sai Bi, Hao Tan, Yuanbo Xiangli, Nanxuan Zhao, Kalyan Sunkavalli, and Zexiang Xu. Gs-lrm: Large reconstruction model for 3d gaussian splatting. InEuropean Conference on Computer Vision , pages 1–19. Springer,
[119]

Clay: A controllable large-scale generative model for creating high-quality 3d assets

Longwen Zhang, Ziyu Wang, Qixuan Zhang, Qiwei Qiu, Anqi Pang, Haoran Jiang, Wei Yang, Lan Xu, and Jingyi Yu. Clay: A controllable large-scale generative model for creating high-quality 3d assets. ACM Transactions on Graphics (TOG), 43(4):1–20, 2024. 3, 14

2024
[120]

Michelangelo: Conditional 3d shape generation based on shape-image-text aligned latent representation

Zibo Zhao, Wen Liu, Xin Chen, Xianfang Zeng, Rui Wang, Pei Cheng, Bin Fu, Tao Chen, Gang Yu, and Shenghua Gao. Michelangelo: Conditional 3d shape generation based on shape-image-text aligned latent representation. Advances in Neural Information Processing Systems, 36, 2024. 3

2024
[121]

Open3d: A modern library for 3d data processing

Qian-Yi Zhou, Jaesik Park, and Vladlen Koltun. Open3d: A modern library for 3d data processing. arXiv preprint arXiv:1801.09847, 2018. 5

2018 arXiv
[122]

Non-stationary texture synthe- sis by adversarial expansion

Yang Zhou, Zhen Zhu, Xiang Bai, Dani Lischinski, Daniel Cohen-Or, and Hui Huang. Non-stationary texture synthe- sis by adversarial expansion. ACM Transactions on Graph- ics (TOG), 37(4):1–13, 2018. 3

2018
[123]

Triplane meets gaussian splatting: Fast and generalizable single- view 3d reconstruction with transformers

Zi-Xin Zou, Zhipeng Yu, Yuan-Chen Guo, Yangguang Li, Ding Liang, Yan-Pei Cao, and Song-Hai Zhang. Triplane meets gaussian splatting: Fast and generalizable single- view 3d reconstruction with transformers. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- t...

2024
[126]

10, 13, and 14 to better il- lustrate the outputs of ShapeShifter on a variety of reference models

Additional results and renderings We provide more results in Fig. 10, 13, and 14 to better il- lustrate the outputs of ShapeShifter on a variety of reference models. Note that we also show that ShapeShifter can gen- erate purely geometric variants from untextured meshes, see l...
[127]

First, the validity of these two scores is debatable

Additional comments on metrics While we use the two commonly-used metrics (geomet- ric quality and diversity through SSFID and pairwise IoU scores) to evaluate our results and compare them to prior art, a few comments are in order. First, the validity of these two scores is de...
[128]

Here, we provide the inference timing for generating a sin- gle variant (i.e., using a batch size of 1 instead of 10) as shown in Tab

Inference timings In the main paper, the inference times for ShapeShifter and Sin3DM are reported for the generation of 10 variants. Here, we provide the inference timing for generating a sin- gle variant (i.e., using a batch size of 1 instead of 10) as shown in Tab. 4. Our me...
[129]

QEM averaging Finally, we demonstrate why our use of QEM averag- ing during our fine-to-coarse analysis of the input mod- els helps preserve sharp features of the ground truth. As Fig. 12 demonstrates, standard scale-by-scale averaging of the points and normals from the finest...
[2019]

arXiv:1912.01703 [cs, stat]. 5

1912 arXiv
[2020]

License: Royalty Free. 6

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.

[1] [1]

Photo-realistic floating wood

All about Blender-3D. Photo-realistic floating wood. http s://www.cgtrader.com/free-3d-models/pla nt/other/photo-realistic-floating-wood ,

[2] [2]

Learning representations and generative models for 3d point clouds

Panos Achlioptas, Olga Diamanti, Ioannis Mitliagkas, and Leonidas Guibas. Learning representations and generative models for 3d point clouds. In International Conference on Machine Learning, pages 40–49. PMLR, 2018. 2

2018

[3] [3]

State of the art on diffusion models for visual computing

and others. State of the art on diffusion models for visual computing. In Computer Graphics Forum, page e15063. Wiley Online Library, 2024. 2

2024

[4] [4]

Small town

Pedram Ashoori. Small town. https://www.cgtrad er.com/free-3d-models/exterior/cityscap e/small-town-87b127c8-c991-4063-aa69- e58800686299, 2020. License: Royalty Free. 6

2020

[5] [5]

Patchmatch: A randomized correspon- dence algorithm for structural image editing

Connelly Barnes, Eli Shechtman, Adam Finkelstein, and Dan B Goldman. Patchmatch: A randomized correspon- dence algorithm for structural image editing. ACM Trans. Graph., 28(3):24, 2009. 3

2009

[6] [6]

Learn- ing long-term dependencies with gradient descent is diffi- cult

Yoshua Bengio, Patrice Simard, and Paolo Frasconi. Learn- ing long-term dependencies with gradient descent is diffi- cult. IEEE Transactions on Neural Networks , 5(2):157– 166, 1994. 2

1994

[7] [7]

Blender - a 3D modelling and rendering package

Blender Online Community. Blender - a 3D modelling and rendering package. Blender Foundation, Stichting Blender Foundation, Amsterdam, 2018. 6

2018

[8] [8]

Texfusion: Synthesizing 3d textures with text-guided image diffusion models

Tianshi Cao, Karsten Kreis, Sanja Fidler, Nicholas Sharp, and Kangxue Yin. Texfusion: Synthesizing 3d textures with text-guided image diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 4169–4181, 2023. 4

2023