Pith. sign in

REVIEW 4 major objections 6 minor 8 cited by

Ultra3D: Efficient and High-Fidelity 3D Generation with Part Attention

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Part-local attention speeds 3D generation 6.7x without quality loss.

desk verdict A clean efficiency win for sparse-voxel 3D generation, with a quality-equivalence claim that outruns its evidence. read the letter →

arxiv 2507.17745 v3 pith:2LDLUVGD submitted 2025-07-23 cs.CV cs.AI

Yiwen Chen , Zhihao Li , Yikai Wang , Hu Zhang , Qin Li , Chi Zhang , Guosheng Lin This is my paper · ORCID
classification cs.CVcs.AI
keywords 3DgenerationsparsevoxelspartattentiondiffusiontransformerVecSetimage-to-3Dsegmentationefficient
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

Ultra3D argues that the two heavy stages of sparse-voxel 3D generation can be made much cheaper without hurting output quality. In the first stage, instead of running a diffusion transformer over a dense low-resolution grid to predict voxel coordinates, it generates a coarse mesh with the compact VecSet representation and voxelizes that mesh. In the second stage, it replaces most global attention with Part Attention, which lets each voxel attend only to voxels in the same semantic part. The paper reports up to 6.7x faster latent generation, a 3.3x faster overall pipeline, and state-of-the-art visual fidelity at 1024 resolution. If correct, this makes high-resolution 3D asset generation practical at much lower compute.

What carries the argument

Part Attention is the load-bearing mechanism: each active voxel carries a part index from an external segmentation model, and self-attention is masked so tokens only interact within their part group, while cross-attention is restricted to image pixels whose projected part set contains the voxel's part. The paper pairs this with a VecSet-based first stage that generates a coarse mesh, voxelizes it to obtain coordinates and part labels, and a part annotation pipeline that runs PartField features through agglomerative clustering into a fixed 8 groups, with filtering, to supply labels at scale. A block structure of one full-attention block followed by three Part Attention blocks gives cross-part communication at low cost.

What would settle it

Take a set of objects whose automatic part segmentation is known to be poor, such as meshes with neighborhood inconsistency above the paper's 25% threshold, generate each with Part Attention and with full attention, and compare surface fidelity and user preference; if quality drops measurably when labels are wrong, the no-quality-loss claim fails. A simpler probe: replace the learned part labels with random 8-group assignments at inference and measure the resulting degradation.

Watch

Extended reading notes

Core claim

The central discovery is that global attention over all sparse-voxel tokens is largely redundant once the coarse object structure is already known. Ultra3D shows that the first stage only needs to fix the layout, so a coarse mesh from VecSet suffices, and the second stage only needs to refine local surfaces, so attention can be masked to within part groups. Self-attention is restricted with the mask in Eq. 1, and cross-attention to image tokens is restricted to pixels onto which the same part projects, via Eq. 2. A small number of full-attention blocks at downsampled resolution keep parts stylistically coherent. The result is a 6.7x speed-up in latent generation and 3.3x overall pipeline speed-up with no measured drop in quality, enabling 1024-resolution generation.

Load-bearing premise

Part Attention only preserves quality if the part labels given to it at inference match the object's real semantic boundaries; the paper's own filtering discards a fraction of meshes where labels are imbalanced or inconsistent, and a wrong label split would cut off attention where it is needed.

Editorial extensions

If this is right

  • The 6.7x cut in second-stage attention cost makes training and inference at sparse-voxel resolutions of 128 rather than downsampled 64 affordable, which the paper shows is what high-fidelity 1024 meshes need.
  • Because only the coarse layout comes from VecSet, the less expressive VecSet geometry does not cap final surface detail; fidelity is set by the sparse-voxel refinement stage.
  • Part Attention turns part segmentation into a first-class input of generation, so progress in automatic part segmentation directly translates into faster and better 3D generation.
  • The fixed 8-part training grouping does not lock inference: the model accepts other part counts, so users can trade speed against quality per object.
  • Replacing 3D window attention with semantic part grouping avoids the style fragmentation that fixed spatial windows cause.

Reading between the lines

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

  • The paper's robustness to part-count variation hints that the exact semantic identity of groups matters less than the locality constraint itself; if so, nearly any consistent grouping would yield the speed-up, and the mechanism could be trained on cheaper proxy groupings.
  • A testable extension is to make the number of parts adaptive per object at inference, using the same clustering pipeline, and measure the quality-speed frontier; the paper's fixed 8-group compromise leaves this headroom open.
  • The dependence on an external segmenter at inference could be removed by using recent part-aware VecSet generators to emit part labels jointly, an integration the paper itself notes as possible.
  • The cross-attention masking in Eq. 2 relies on camera estimates from the VecSet decoder; when those estimates are wrong, part-to-pixel correspondence degrades, so pose-error sensitivity is a natural stress test the paper does not quantify.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes Ultra3D, an image-to-3D generation framework built on sparse voxel representations. The first stage replaces the dense-grid DiT used by prior work (e.g., Trellis) with a VecSet-based generator that produces a coarse mesh, which is then voxelized into sparse voxel coordinates. The second stage refines per-voxel latent features with a diffusion transformer that uses Part Attention, a masked attention mechanism that restricts self-attention to voxels sharing the same part label and restricts cross-attention to image regions onto which the same part projects. To obtain part labels, the authors build a pipeline based on PartField followed by agglomerative clustering and two filtering criteria, with a fixed number of eight part groups. Experiments report up to 6.7x speedup in part self-attention, 4.1x in part cross-attention, and 3.3x end-to-end pipeline speedup, while claiming state-of-the-art visual fidelity at 1024 resolution based on qualitative comparisons and user studies.

Significance. If the central claims hold, Ultra3D would be a practical contribution: it addresses a real computational bottleneck in high-resolution sparse-voxel 3D generation, and the idea of exploiting semantic part structure for localized attention is well motivated by the geometry of 3D objects. The speedup measurements appear genuine and are reported under a unified FlashAttention-2 implementation, which is a strength. The use of external building blocks (VecSet, PartField, Trellis, SparC3D) is properly acknowledged, and the annotation pipeline is a useful engineering contribution in itself. However, the claim that Part Attention preserves quality relative to full attention rests on a very small user study with high no-preference rates and no significance testing, and the inference-time label distribution differs from the filtered training distribution. These are load-bearing issues for the paper's main claim, not mere presentation concerns.

major comments (4)
  1. [Sec. 4.2, Sec. 4.3, Eq. (1), Eq. (2)]
  2. [Table 1(b), Sec. 5.3]
  3. [Sec. 5.2, Sec. 4.2]
  4. [Sec. 4.2, Sec. 5.1]
minor comments (6)
  1. [Throughout]
  2. [Table 2 and Sec. 5.3]
  3. [Sec. 4.3]
  4. [Fig. 5 and Sec. 4.2]
  5. [Sec. 5.1 and Sec. 6]
  6. [Sec. 4.1, Eq. (1)-(2)]

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: speedup and quality claims are measured against self-run baselines, part labels come from an external model, and self-citations are contextual rather than load-bearing.

full rationale

The paper's central claims are empirical rather than derived from its own definitions. The 6.7x Part Self Attention speedup and 3.3x pipeline speedup are reported as measured results using FlashAttention-2 on self-implemented baselines (Table 2), not as consequences of an equation that presupposes the conclusion. The 'without compromising quality' claim is supported by a user study (Table 1b) comparing Part Attention against a full-attention variant, and the full-attention and window-attention baselines are explicitly described as finetuned variants of the same checkpoint, which is a legitimate experimental comparison rather than a circular construction. The part labels used by Part Attention come from PartField (Liu et al., 2025), an external segmentation model, and the coarse mesh is produced by VecSet (Zhang et al., 2023), also an external representation. The paper self-cites its own prior work (MeshAnything, MeshAnything V2, DeepMesh, MeshGen, SparC3D), but these citations appear in related-work context or as adopted building blocks (e.g., the SparConv-VAE from Li et al. 2025b); none is invoked to justify the paper's central claim that Part Attention preserves quality or that VecSet-based coarse layout generation loses nothing. The most plausible concern is that inference-time part labels are computed on the coarse generated mesh without the filtering applied during training, and inference cameras are estimated rather than ground-truth, so the model may face an out-of-distribution label/camera regime. That is a robustness and evaluation-gap concern, not a circularity: the paper never equates the inference-time distribution with the training distribution by definition, and it does not fit a parameter and then rename that fit as a prediction. No quoted equation or construction reduces a claimed result to its own input, so no circular step can be exhibited under the stated rules.

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

The method rests on external models (PartField, VecSet, Trellis, Sparc3D) and hand-set engineering choices: the fixed 8-part clustering, the 25% filtering thresholds, the 1:3 full-to-part block ratio, and inference constants. Part Attention itself introduces no new physical or representational entity; its correctness depends on the quality of external part labels and cameras.

free parameters (6)
  • Fixed number of part clusters K = 8
    Set by hand for all meshes during training (Sec 4.3: 'empirically set the number to 8'), balancing over-segmentation vs computational savings. The model is claimed to be robust to other K at inference (Fig 5).
  • Part filtering threshold: sum of squared voxel ratios = 25%
    Samples with a higher score are removed during preprocessing (Sec 5.1); threshold chosen from the distribution in Fig 6.
  • Part filtering threshold: neighborhood inconsistency = 25%
    Samples with higher inconsistency are removed during preprocessing (Sec 5.1); threshold chosen from the distribution in Fig 6.
  • Full-attention to part-attention block ratio = 1 full : 3 part blocks
    Design choice for the DiT (Sec 4.2: 'stack one full-attention block followed by three Part Attention blocks'); not ablated in the paper.
  • Classifier-free guidance scale = 3.5
    Inference hyperparameter (Sec 5.1); chosen by the authors.
  • Number of inference steps = 25
    Inference hyperparameter (Sec 5.1); chosen by the authors.
assumptions (6)
  • domain assumption PartField provides accurate part semantics across arbitrary meshes
    Sec 4.1 (Eq 1, Eq 2) and Sec 4.3 rely on part labels from PartField; the filtering step concedes that some meshes are not well segmented.
  • domain assumption Cameras are known accurately enough for part-image projection
    Part Cross Attention (Sec 4.1, Eq 2) projects parts onto the image using ground-truth cameras at training and estimated cameras at inference; an erroneous camera invalidates the masking constraint.
  • domain assumption The second stage only needs local refinement, not global modeling
    Sec 4.1 motivates removing global attention by claiming sparse voxels already capture global shape; this is a modeling assumption not proven by measurement.
  • domain assumption Trellis two-stage pipeline and Sparc3D VAE are valid foundations
    Sec 3.1 and Sec 4.2 build on these external systems without re-deriving them.
  • domain assumption Fixed 8-part training transfers to variable part counts
    Sec 4.2 claims robustness to varying K (Fig 5), but this is only demonstrated qualitatively.
  • domain assumption Agglomerative clustering with fixed K yields consistent semantic parts
    Sec 4.3 clusters per-voxel features from PartField into exactly 8 groups; the paper acknowledges this is a compromise.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Ultra3D: Efficient and High-Fidelity 3D Generation with Part Attention." pith.science (2026). https://pith.science/paper/2LDLUVGD

@misc{pith2026250717745,
  author       = {Pith},
  title        = {Pith review of: Ultra3D: Efficient and High-Fidelity 3D Generation with Part Attention},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2LDLUVGD}},
  note         = {Machine review of arXiv:2507.17745}
}
read the original abstract

Recent advances in sparse voxel representations have significantly improved the quality of 3D content generation, enabling high-resolution modeling with fine-grained geometry. However, existing frameworks suffer from severe computational inefficiencies due to the quadratic complexity of attention mechanisms in their two-stage diffusion pipelines. In this work, we propose Ultra3D, an efficient 3D generation framework that significantly accelerates sparse voxel modeling without compromising quality. Our method leverages the compact VecSet representation to efficiently generate a coarse object layout in the first stage, reducing token count and accelerating voxel coordinate prediction. To refine per-voxel latent features in the second stage, we introduce Part Attention, a geometry-aware localized attention mechanism that restricts attention computation within semantically consistent part regions. This design preserves structural continuity while avoiding unnecessary global attention, achieving up to 6.7x speed-up in latent generation. To support this mechanism, we construct a scalable part annotation pipeline that converts raw meshes into part-labeled sparse voxels. Extensive experiments demonstrate that Ultra3D supports high-resolution 3D generation at 1024 resolution and achieves state-of-the-art performance in both visual fidelity and user preference.

Figures

Figures reproduced from arXiv: 2507.17745 by the authors.

Figure 1
Figure 1. Image-to-3D Generation Results of ULTRA3D. ULTRA3D delivers high-quality 3D meshes with fine-grained geometric details while maintaining efficient generation. Please zoom in to view detailed geometry. ABSTRACT Recent advances in sparse voxel representations have significantly improved the quality of 3D content generation, enabling high-resolution modeling with fine￾grained geometry. However, existing frameworks suff… view at source ↗
Figure 2
Figure 2. Expeiments on different attention mechanisms. Each color denotes an attention group, within which attention is computed independently. All other settings remain unchanged, with only the attention mechanism being replaced. 3D Window Attention partitions the object space into 8 fixed regions by splitting at the center along each axis. This fixed partitioning often misaligns with semantic boundaries, leading to degrade… view at source ↗
Figure 3
Figure 3. Pipeline Overview. We introduce ULTRA3D, an efficient and high-quality 3D generation framework that first generates sparse voxel layout via VecSet and then refines it by generating per￾voxel latent. The core of ULTRA3D is Part Attention, an efficient localized attention mechanism that performs attention computation independently within each part group. Besides, when the input condition is an image, each part group p… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Impact of Resolution on Generation Quality. We compare results under different configurations, where “512 64” denotes a mesh resolution of 512 and a sparse voxel resolution of 64. In previous works, to reduce computational cost in the second stage, the sparse voxels ar…
Figure 5
Figure 5. Figure 5: Robustness of Part Annotation. Although our method is trained using data with exactly 8 part groups, we find it to be robust to variations in part annotation. Varying the number of part groups has little impact on generation quality, suggesting that increasing the numb…
Figure 6
Figure 6. Figure 6: Percentile for Filtering Metrics of Part Annotation. We apply two metrics to filter poorly segmented samples: (1) the sum of squared voxel ratios, which identifies imbalanced part distributions, and (2) neighborhood inconsistency, which measures the proportion of voxel…
Figure 7
Figure 7. Figure 7: Comparison with Prior Methods. Our method produces higher fidelity and richer sur￾face details. As highlighted in the red boxes, our results align more closely with the input image compared to other methods. During inference, we first pass the input image through our s…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 8 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Hunyuan3D-Buffalo 1.0: A Unified Multimodal Model for Scalable 3D Generation, Understanding, and Editing

    cs.CV 2026-08 conditional novelty 6.0 of 10

    A unified 3D multimodal model combines understanding, text-to-3D generation, instruction-guided editing, and part generation in one architecture, trained on an 87M-sample corpus, with claimed state-of-the-art results.

  2. PartMat: Material-Aware 3D Part Decomposition with a Single Global Latent

    cs.CV 2026-08 conditional novelty 6.0 of 10

    PartMat decomposes and generates 3D objects into material-defined parts using a single global latent, keeping decoding cost constant regardless of part count.

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

    cs.CV 2026-07 conditional novelty 6.0 of 10

    Hierarchical point-shuffle densification plus local AVS-Conv multi-scale decoding lets compact VecSet VAEs approach voxel-level 3D reconstruction fidelity at much lower token and query cost.

  4. HIVE-3D: Hierarchical Voxel Enhancement for High-Quality 3D Scene Generation

    cs.CV 2026-07 conditional novelty 6.0 of 10

    A hierarchical coarse-to-fine pipeline refines single-image 3D scenes component-by-component using a learned voxel super-resolution model conditioned on coarse voxels.

  5. EditVerse3D: High-Quality 3D Object Editing with Region-Aware Learning

    cs.CV 2026-07 conditional novelty 6.0 of 10

    An end-to-end 3D editing framework achieves high-fidelity local edits from coarse bounding boxes and 2D image prompts using region-aware loss reweighting and a large-scale parts-derived training dataset.

  6. InSpace: Structure-Aware 3D Indoor Scene Generation from a Single 360{\deg} Image

    cs.CV 2026-07 conditional novelty 6.0 of 10

    InSpace generates complete structure-aware 3D indoor scenes (layout plus textured assets) from a single equirectangular 360° image via three-stage flow matching with view- and asset-selective attention.

  7. Muses: Designing, Composing, Generating Nonexistent Fantasy 3D Creatures without Training

    cs.CV 2026-01 conditional novelty 6.0 of 10

    Muses creates new fantasy 3D animals by designing a combined skeleton, fusing voxel parts from separate 3D models along that skeleton, then restyling textures via image editing — with no training.

  8. PoseMaster: A Unified 3D Native Framework for Stylized Pose Generation

    cs.CV 2025-06 conditional novelty 6.0 of 10

    PoseMaster produces a 3D character mesh from one image and a target 3D skeleton, preserving identity and pose in a single unified model, and it outperforms two-stage 2D-to-3D baselines on the VRoid pose canonicalizati...

Reference graph

Works this paper leans on

29 extracted references · 1 canonical work pages · cited by 8 Pith papers

  1. [1]

    Longformer: The long-document transformer

    Iz Beltagy, Matthew E Peters, and Arman Cohan. Longformer: The long-document transformer. arXiv preprint arXiv:2004.05150,

    arXiv 2004

  2. [5]

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

    Matt Deitke, Ruoshi Liu, Matthew Wallingford, Huong Ngo, Oscar Michel, Aditya Kusupati, Alan Fan, Christian Laforte, Vikram V oleti, Samir Yitzhak Gadre, et al. Objaverse-xl: A universe of 10m+ 3d objects. arXiv preprint arXiv:2307.05663, 2023a. Matt Deitke, Dustin Schwenk, Jordi Salvador, Luca Weihs, Oscar Michel, Eli VanderBilt, Ludwig Schmidt, Kiana Eh...

    arXiv

  3. [8]

    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. arXiv preprint arXiv:2311.04400,

    arXiv

  4. [9]

    Diederik P Kingma and Jimmy Ba

    URL https://arxiv.org/abs/2506.15442. Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980,

    arXiv

  5. [10]

    Yushi Lan, Fangzhou Hong, Shuai Yang, Shangchen Zhou, Xuyi Meng, Bo Dai, Xingang Pan, and Chen Change Loy

    URL https://arxiv.org/abs/2506.16504. Yushi Lan, Fangzhou Hong, Shuai Yang, Shangchen Zhou, Xuyi Meng, Bo Dai, Xingang Pan, and Chen Change Loy. Ln3diff: Scalable latent neural fields diffusion for speedy 3d generation. In ECCV,

    arXiv

  6. [11]

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

    13 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 gen- eration and large reconstruction model. In ICLR, 2024a. Peng Li, Yuan Liu, Xiaoxiao Long, Feihu Zhang, Cheng Lin, Mengfei Li, Xingqun Qi, Shang- hang Zhang, Wenhan Luo, Ping ...

    arXiv

  7. [12]

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le

    URL https://arxiv.org/abs/2506.05573. Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling. In International Conference on Learning Representations (ICLR),

    arXiv

  8. [13]

    One- 2-3-45: Any single image to 3d mesh in 45 seconds without per-shape optimization

    Minghua Liu, Chao Xu, Haian Jin, Linghao Chen, Mukund Varma T, Zexiang Xu, and Hao Su. One- 2-3-45: Any single image to 3d mesh in 45 seconds without per-shape optimization. In NeurIPS, 2024a. Minghua Liu, Mikaela Angelina Uy, Donglai Xiang, Hao Su, Sanja Fidler, Nicholas Sharp, and Jun Gao. Partfield: Learning 3d feature fields for part segmentation and ...

    arXiv

Show all 29 references
  1. [14]

    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 ICLR, 2024b. Xiaoxiao Long, Yuan-Chen Guo, Cheng Lin, Yuan Liu, Zhiyang Dou, Lingjie Liu, Yuexin Ma, So...

    arXiv

  2. [16]

    Barron, and Ben Mildenhall

    14 Ben Poole, Ajay Jain, Jonathan T. Barron, and Ben Mildenhall. Dreamfusion: Text-to-3d using 2d diffusion. In The Eleventh International Conference on Learning Representations, ICLR 2023, Kigali, Rwanda, May 1-5, 2023 . OpenReview.net,

    2023

  3. [17]

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

    Jiaxiang Tang, Zhaoxi Chen, Xiaokang Chen, Tengfei Wang, Gang Zeng, and Ziwei Liu. Lgm: Large multi-view gaussian model for high-resolution 3d content creation. In ECCV, 2024a. Jiaxiang Tang, Zhaoshuo Li, Zekun Hao, Xian Liu, Gang Zeng, Ming-Yu Liu, and Qinsheng Zhang. Edgerun...

    arXiv

  4. [18]

    Dmitry Tochilkin, David Pankratz, Zexiang Liu, Zixuan Huang, Adam Letts, Yangguang Li, Ding Liang, Christian Laforte, Varun Jampani, and Yan-Pei Cao

    URL https://arxiv.org/abs/2506.09980. Dmitry Tochilkin, David Pankratz, Zexiang Liu, Zixuan Huang, Adam Letts, Yangguang Li, Ding Liang, Christian Laforte, Varun Jampani, and Yan-Pei Cao. Triposr: Fast 3d object reconstruction from a single image. arXiv preprint arXiv:2403.02151,

    arXiv

  5. [20]

    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, Zexi- ang Xu, and Kai Zhang. Pf-lrm: Pose-free large reconstruction model for joint pose and shape prediction. arXiv preprint arXiv:2311.12024, 2023b. Zhengyi Wang, Cheng Lu, Yikai Wang, Fan B...

    arXiv

  6. [21]

    Xinyue Wei, Kai Zhang, Sai Bi, Hao Tan, Fujun Luan, Valentin Deschaintre, Kalyan Sunkavalli, Hao Su, and Zexiang Xu

    URL https://arxiv.org/abs/2411.09595. Xinyue Wei, Kai Zhang, Sai Bi, Hao Tan, Fujun Luan, Valentin Deschaintre, Kalyan Sunkavalli, Hao Su, and Zexiang Xu. Meshlrm: Large reconstruction model for high-quality mesh. arXiv preprint arXiv:2404.12385,

    arXiv

  7. [22]

    URL https://arxiv.org/abs/2411. 07025. Kailu Wu, Fangfu Liu, Zhihan Cai, Runjie Yan, Hanyang Wang, Yating Hu, Yueqi Duan, and Kaisheng Ma. Unique3d: High-quality and efficient 3d mesh generation from a single image. CoRR, abs/2405.20343, 2024a. Shuang Wu, Youtian Lin, Feihu Zh...

    arXiv

  8. [23]

    Jianfeng Xiang, Zelong Lv, Sicheng Xu, Yu Deng, Ruicheng Wang, Bowen Zhang, Dong Chen, Xin Tong, and Jiaolong Yang

    URL https://arxiv.org/abs/2505.17412. Jianfeng Xiang, Zelong Lv, Sicheng Xu, Yu Deng, Ruicheng Wang, Bowen Zhang, Dong Chen, Xin Tong, and Jiaolong Yang. Structured 3d latents for scalable and versatile 3d generation,

    arXiv

  9. [24]

    Bojun Xiong, Si-Tong Wei, Xin-Yang Zheng, Yan-Pei Cao, Zhouhui Lian, and Peng-Shuai Wang

    URL https://arxiv.org/abs/2412.01506. Bojun Xiong, Si-Tong Wei, Xin-Yang Zheng, Yan-Pei Cao, Zhouhui Lian, and Peng-Shuai Wang. Octfusion: Octree-based diffusion models for 3d shape generation,

    arXiv

  10. [25]

    Jiale Xu, Weihao Cheng, Yiming Gao, Xintao Wang, Shenghua Gao, and Ying Shan

    URL https: //arxiv.org/abs/2408.14732. Jiale Xu, Weihao Cheng, Yiming Gao, Xintao Wang, Shenghua Gao, and Ying Shan. Instantmesh: Efficient 3d mesh generation from a single image with sparse-view large reconstruction models. arXiv preprint arXiv:2404.07191, 2024a. Yinghao Xu, ...

    arXiv

  11. [26]

    Chongjie Ye, Yushuang Wu, Ziteng Lu, Jiahao Chang, Xiaoyang Guo, Jiaqing Zhou, Hao Zhao, and Xiaoguang Han

    URL https: //arxiv.org/abs/2504.07943. Chongjie Ye, Yushuang Wu, Ziteng Lu, Jiahao Chang, Xiaoyang Guo, Jiaqing Zhou, Hao Zhao, and Xiaoguang Han. Hi3dgen: High-fidelity 3d geometry generation from images via normal bridging. arXiv preprint arXiv:2503.22236,

    arXiv

  12. [27]

    Biao Zhang, Jiapeng Tang, Matthias Niessner, and Peter Wonka

    URL https://arxiv.org/abs/2007.14062. Biao Zhang, Jiapeng Tang, Matthias Niessner, and Peter Wonka. 3dshape2vecset: A 3d shape representation for neural fields and generative diffusion models. ACM Transactions On Graphics (TOG), 42(4):1–16,

    2007 arXiv

  13. [28]

    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. In ECCV, 2024a. Longwen Zhang, Ziyu Wang, Qixuan Zhang, Qiwei Qiu, Anqi Pang, Haoran Jiang, Wei Yang, Lan Xu, and Jingyi Y...

    arXiv

  14. [29]

    Hunyuan3d 2.0: Scaling diffusion models for high resolution textured 3d assets generation, 2025b

    Zibo Zhao, Zeqiang Lai, Qingxiang Lin, Yunfei Zhao, Haolin Liu, Shuhui Yang, Yifei Feng, Mingxin Yang, Sheng Zhang, Xianghui Yang, Huiwen Shi, Sicong Liu, Junta Wu, Yihang Lian, Fan Yang, Ruining Tang, Zebin He, Xinzhou Wang, Jian Liu, Xuhui Zuo, Zhuo Chen, Biwen Lei, Haohan W...

    arXiv

  15. [2015]

    Meshxl: Neural coordinate field for generative 3d foundation models

    Sijin Chen, Xin Chen, Anqi Pang, Xianfang Zeng, Wei Cheng, Yijun Fu, Fukun Yin, Billzb Wang, Jingyi Yu, Gang Yu, et al. Meshxl: Neural coordinate field for generative 3d foundation models. Advances in Neural Information Processing Systems (NeurIPS) , 2024a. Yiwen Chen, Tong He...

    arXiv

  16. [2017]

    Yeh, and Greg Shakhnarovich

    Haochen Wang, Xiaodan Du, Jiahao Li, Raymond A. Yeh, and Greg Shakhnarovich. Score jacobian chaining: Lifting pretrained 2d diffusion models for 3d generation. In IEEE CVPR, pp. 12619– 12629, 2023a. Peng Wang and Yichun Shi. Imagedream: Image-prompt multi-view diffusion for 3d...

    arXiv

  17. [2020]

    Dinov2: Learning robust visual features without supervision

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193,

    arXiv

  18. [2021]

    3dtopia: Large text-to-3d generation model with hybrid diffusion priors

    Fangzhou Hong, Jiaxiang Tang, Ziang Cao, Min Shi, Tong Wu, Zhaoxi Chen, Tengfei Wang, Liang Pan, Dahua Lin, and Ziwei Liu. 3dtopia: Large text-to-3d generation model with hybrid diffusion priors. CoRR, abs/2403.02234,

    arXiv

  19. [2023]

    Shapenet: An information-rich 3d model repository

    Angel X Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio Savarese, Manolis Savva, Shuran Song, Hao Su, et al. Shapenet: An information-rich 3d model repository. arXiv preprint arXiv:1512.03012,

    arXiv

  20. [2024]

    Xianglong He, Zi-Xin Zou, Chia-Hao Chen, Yuan-Chen Guo, Ding Liang, Chun Yuan, Wanli Ouyang, Yan-Pei Cao, and Yangguang Li

    URL https://arxiv.org/abs/2412.09548. Xianglong He, Zi-Xin Zou, Chia-Hao Chen, Yuan-Chen Guo, Ding Liang, Chun Yuan, Wanli Ouyang, Yan-Pei Cao, and Yangguang Li. Sparseflex: High-resolution and arbitrary-topology 3d shape modeling, 2025a. URL https://arxiv.org/abs/2503.21732. ...

    2021 arXiv

  21. [2025]

    URL https://arxiv.org/abs/2505.04656. Tri Dao. Flashattention-2: Faster attention with better parallelism and work partitioning. arXiv preprint arXiv:2307.08691,

    arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.