Pith. sign in

REVIEW 4 major objections 5 minor 54 references

LoomNet: Enhancing Multi-View Image Generation via Latent Space Weaving

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read LoomNet claims that multi-view consistency is best achieved by having parallel diffusion models build one shared latent triplane from per-view hypotheses, and that this representation yields state-of-the-art image quality and 3D…

desk verdict A credible new architecture for multi-view consistency — the SOTA numbers are plausible but unverified, so treat them as provisional pending code and error bars. read the letter →

arxiv 2507.05499 v1 pith:4YU5HTP6 submitted 2025-07-07 cs.CV

Giulio Federico , Fabio Carrara , Claudio Gennaro , Giuseppe Amato , Marco Di Benedetto This is my paper
classification cs.CV
keywords multi-viewgenerationdiffusionmodelslatenttriplaneviewconsistencysingle-image3Dsplattingcross-attentionfusionreconstruction
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

LoomNet claims that multi-view image generation from a single photo can be made consistent by having several parallel diffusion models collaboratively build one shared latent scene representation, rather than generating each view independently or synchronizing only through epipolar geometry. The paper proposes a communication module that splats each view's decoder features onto three orthogonal planes, fuses the planes across views with attention, and weaves the result into a continuous triplane from which all views are rendered. On the Google Scanned Objects benchmark this yields state-of-the-art image quality and 3D reconstruction, with PSNR 21.11 and Volume IoU 0.5366 in the variable-elevation setting, and 16 views generated in about 15 seconds. The practical promise is faster, higher-quality single-image 3D content creation for augmented reality, robotics, and games.

What carries the argument

The shared latent triplane built by the four-stage communication module: per-view splatting (each feature map projected along rays onto $XY$, $YZ$, and $XZ$ planes), pixel-wise cross-attention fusion into learnable planes, weaving (AdaLayerNorm, self-attention, and MLP refinement repeated at each decoder block), and latent rendering that samples the triplane along rays and corrects each pixel feature through an EG3D-style MLP decoder. This object carries the consistency claim because all views are decoded from one scene representation instead of being generated independently.

What would settle it

Compare LoomNet against a control in which the fusion and weaving stages are replaced by simple averaging of the splatted planes with identical training; if image quality and Chamfer distance stay close, the attention-fusion and weaving mechanism is not the source of consistency. A second check would be to render depth from the shared triplane and measure multi-view depth agreement on Google Scanned Objects, since inconsistent depths would indicate the representation is not truly 3D.

Watch

Extended reading notes

Core claim

The central discovery is that a frozen Zero-1-to-3 diffusion UNet, run in parallel for $N$ viewpoints, can be made view-consistent by treating its intermediate decoder features as per-view scene hypotheses and weaving them into a shared latent triplane. Each view's feature map is ray-splatted onto three orthogonal planes; learnable fusion planes attend over the corresponding pixels of all views; a weaving stage with self-attention and MLP refinement interpolates missing regions; and the final triplane renders every view through an EG3D-style decoder. Because every view reads from the same latent scene, the generated images inherit consistency, and the reconstructed meshes are more accurate than those produced by methods that synchronize via a global feature volume or epipolar constraints.

Load-bearing premise

The load-bearing assumption is that the frozen Zero-1-to-3 UNet's intermediate decoder features already carry enough 3D structure that splatting them into planes and fusing the planes produces a coherent scene; if those features are only 2D appearance codes, the shared triplane averages them and the consistency gain vanishes.

Editorial extensions

If this is right

  • Frozen 2D diffusion backbones can be adapted for multi-view consistency by training only a lightweight communication module, avoiding expensive full fine-tuning of the generator.
  • The shared triplane is usable as a unified 3D latent representation for downstream tasks, not just view synthesis, since it encodes scene structure that transfers to improved meshes.
  • Performance degrades only slightly under variable elevation and irregular camera distributions, suggesting the consistency mechanism generalizes beyond fixed camera rigs.
  • Generating 16 views in about 15 seconds makes single-image 3D generation practical for near-interactive applications.

Reading between the lines

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

  • Beyond the paper: because the communication module is view-agnostic and the backbone is frozen, the same module could plausibly be reused with other frozen diffusion backbones or with more than 16 views, though the paper does not test this.
  • Beyond the paper: the ablation shows that removing positional encoding causes the largest quality drop, so directly probing whether the splatted features encode true 3D geometry, for example by rendering depth from the shared triplane and checking multi-view depth consistency, would clarify the source of the consistency gain.
  • Beyond the paper: the two-stage generate-then-reconstruct pipeline is a stated limitation, so training the shared latent representation end-to-end with a reconstruction loss is a natural next step that could further tighten geometry.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes LoomNet, a multi-view diffusion architecture for generating consistent novel views from a single image. The method runs N parallel instances of a shared Zero-1-to-3 UNet, each denoising a separate view, and inserts a communication module at each decoder level. The module performs per-view splatting of UNet decoder feature maps onto three orthogonal planes, cross-attention fusion of all views' planes into learnable fusion planes, a weaving stage of self-attention and MLP refinement to produce a continuous shared triplane, and a final latent-rendering stage that samples the triplane along rays to correct each view's features. The UNet backbone is frozen; only the communication module is trained. Experiments on the GSO dataset report state-of-the-art PSNR, SSIM, LPIPS, Chamfer Distance, and Volume IoU against EpiDiff, SyncDreamer, Zero-1-to-3, and other baselines, with 16 views generated in 15 seconds. An ablation study examines the effect of latent rendering at all layers, mean fusion, sample count, and positional encoding.

Significance. If the reported results are reproducible, LoomNet makes a meaningful contribution to multi-view generation: it offers a fast, communication-efficient alternative to global volume approaches like SyncDreamer while outperforming the epipolar-constrained EpiDiff on both image quality and 3D reconstruction metrics. The architecture is clearly and completely specified, the design story is coherent, and the ablations (except for one interpretation error discussed below) align with the proposed mechanism. The paper does not release code or provide statistical error bars, which limits verification of the central claim. Nevertheless, the core idea of building a shared triplane latent space through per-view splatting and attention-based fusion is a plausible and original way to enforce cross-view consistency.

major comments (4)
  1. [§4.2, Tables 1 and 2] The central claim of state-of-the-art performance rests on single-run measurements. Tables 1 and 2 report only point estimates without standard deviations, confidence intervals, or the number of random seeds. Given that diffusion sampling is stochastic and the gap to the next-best method is sometimes large (e.g., PSNR 21.11 vs. 18.83 in the variable-elevation scenario), the reported improvement could be due to sampling luck. Please report mean ± std over at least three seeds, or provide per-object error bars, and state the number of test objects used.
  2. [§4.5, Table 3] The ablation discussion is internally inconsistent. The text states that 'Removing PE causes the largest drop,' but Table 3 shows that applying latent rendering at all decoder layers reduces PSNR from 21.60 to 19.43 (a loss of 2.17), while removing PE reduces PSNR to 20.66 (a loss of 0.94). The table also labels the variant 'LT at all decoder layers,' which should presumably read 'LR' (latent rendering). Please correct the typo and revise the interpretation so that the text matches the reported numbers.
  3. [§4.2] The evaluation protocol is under-specified. It is not stated how many GSO objects were evaluated, how many generated views were used for the reconstruction step, whether the baselines were run using official weights or a reimplementation, or on what hardware the runtimes were measured. Because the paper's headline includes a runtime claim ('16 views in 15 seconds'), please provide the exact GPU model, inference batch size, number of diffusion steps, and confirm that baseline runtimes were measured under comparable conditions.
  4. [§3.1 and §4.1] The core assumption that frozen Zero-1-to-3 decoder features constitute per-view scene hypotheses is not directly examined. The paper reports no visualization of the learned fusion planes, no analysis of whether features from different views are aligned at corresponding 3D points, and no probe of the geometric consistency of the splatted features. The strong downstream reconstruction results are encouraging, but a direct analysis—for example, visualizing the planes, measuring feature agreement across views for known 3D correspondences, or ablating the splatting geometry—would substantiate the claim that the shared latent space encodes geometry rather than averaging 2D appearance.
minor comments (5)
  1. [§3.4] The symbol D is used both for the LDM decoder and for the MLP decoder in the latent rendering stage; please use different notation (e.g., D_MLP) to avoid ambiguity.
  2. [Table 1] The baseline is labeled 'Zero123' but the reference list uses 'Zero-1-to-3'; please make the label consistent.
  3. [§5.1] The sentence 'LoomNet is less effective than methods like SyncDreamer at propagating errors consistently across views' is confusing; it presumably means that SyncDreamer produces consistently wrong errors, which is not a desirable property. Please rephrase to clarify.
  4. [§4.5] The text says 'Further details can be found in the supplementary materials,' but no supplementary document is mentioned in the submission; please include it or remove the reference.
  5. [Figure 2] The caption contains the typo 'thesplatting'; it should read 'the splatting'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: LoomNet's claims are empirical benchmark results, not derivations from their own outputs.

full rationale

The paper makes an architectural proposal and supports it with measurements on the external GSO benchmark under EpiDiff's protocol. The central claim—that the shared triplane latent space built from per-view splatting, fusion, and weaving improves multi-view consistency—is not derived from the reported metrics by construction. The splatting equation (Eq. 3), cross-attention fusion (Eqs. 4-6), weaving update (Eqs. 7-8), latent rendering (Eq. 9), and total loss (Eq. 10) do not contain PSNR, SSIM, LPIPS, Chamfer Distance, or Volume IoU as inputs. The model is trained on LVIS/Objaverse and evaluated on Google Scanned Objects, so the reported improvements are externally falsifiable comparisons against independent baselines (Zero123, SyncDreamer, EpiDiff, Point-E, Shap-E, One-2-3-45). The frozen Zero-1-to-3 backbone is an external pretrained model, and while the authors assume its features are geometrically informative, that assumption is a correctness risk about feature grounding, not a circular reduction: the paper never defines the shared latent space in terms of the evaluation quantities, nor does it fit a parameter to the test set and call the result a prediction. The ablation study compares variants of the same method, which is standard practice and not circular. There are no load-bearing self-citations: the reference list does not cite the present authors' prior work as justification for the architecture's central mechanism. The stated limitation that SyncDreamer can have higher inter-view consistency is a concession about a competing method and does not reduce LoomNet's claim to its own inputs. In short, no equation, fitted constant, or self-citation chain makes the derivation equivalent to the output. The paper's contribution is empirical and benchmark-tested, so the appropriate circularity score is 0.

Assumptions & free parameters 5 free parameters · 5 assumptions · 2 invented entities

The empirical method rests on a set of hand-chosen hyperparameters (cube scale, ray sample count, TV weight, fusion iterations, triplane resolution) and architectural assumptions about the geometric meaning of frozen UNet features and the adequacy of attention-based fusion and weaving. These are internal design choices validated only through aggregated benchmark metrics, with no direct measurement of the intermediate latent representation.

free parameters (5)
  • Cube side length S = 1.5
    Defines the fixed 3D volume into which all per-view features are splatted and from which rays sample global features. Objects that do not fit this scale would produce empty or clipped planes; it is chosen by hand and not learned or adapted per object.
  • Number of ray samples M = 16
    Samples per ray during splatting and latent rendering. Ablation shows M=8 gives a small drop (21.60 to 21.39 PSNR), so the value trades quality against speed.
  • Total variation weight lambda = 0.001
    Strength of the smoothness regularizer on the learnable fusion planes. Chosen by hand and not ablated; it shapes the spatial coherence of the planes.
  • Fusion/weaving iterations per decoder block = 3, 4, 6, 8 (bottleneck, decoder levels 1-3)
    Number of communication passes per level; chosen by hand and not ablated, directly affects the amount of inter-view information exchange.
  • Triplane resolution = 32x32x512
    Resolution and channel count of the shared latent planes; a capacity versus quality tradeoff chosen by hand.
assumptions (5)
  • domain assumption A fixed cube of side 1.5 centered at the origin contains the objects in both training and evaluation domains.
    Required for Eq. 3 splatting and latent rendering to have non-empty feature coverage; if the object scale mismatches, planes under- or over-sample.
  • domain assumption Pixel features from the frozen Zero-1-to-3 UNet are informative 3D scene hypotheses when augmented with Plucker and depth encodings.
    Section 3.1; the whole splatting-fusion pipeline operates on these features without fine-tuning the backbone, yet no analysis of feature geometry is provided.
  • ad hoc to paper Pixel-wise cross-attention can fuse N homologous planes without destructive blending.
    Section 3.2; this is a proposed mechanism rather than a standard theorem; its adequacy is established only by the final metrics.
  • domain assumption The Zero-1-to-3 diffusion loss plus total variation regularizer is a sufficient objective to train the communication module while keeping the backbone frozen.
    Section 3.5 and 4.1; no comparison against alternative losses or unfrozen backbones is reported.
  • ad hoc to paper Self-attention across the three planes interpolates missing regions and improves temporal consistency.
    Section 3.3; the weaving stage is a proposed operation, and its effect is only measured jointly with other components; the ablation removes PE but does not isolate weaving.
invented entities (2)
  • Learnable fusion planes (XY, YZ, XZ)
    purpose: Query-based aggregation of per-view splatted planes into a shared scene consensus
    These are newly introduced learnable parameters; their behavior is only indirectly validated through final image and reconstruction metrics, with no separate verification of the fused representation itself.
  • Shared triplane latent space produced after weaving
    purpose: A unified scene embedding from which all views are rendered
    It is an internal representation; the paper provides no direct measurement (such as a feature-consistency probe) that it is a true geometric scene rather than a convenient averaging buffer.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LoomNet: Enhancing Multi-View Image Generation via Latent Space Weaving." pith.science (2026). https://pith.science/paper/4YU5HTP6

@misc{pith2026250705499,
  author       = {Pith},
  title        = {Pith review of: LoomNet: Enhancing Multi-View Image Generation via Latent Space Weaving},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4YU5HTP6}},
  note         = {Machine review of arXiv:2507.05499}
}
read the original abstract

Generating consistent multi-view images from a single image remains challenging. Lack of spatial consistency often degrades 3D mesh quality in surface reconstruction. To address this, we propose LoomNet, a novel multi-view diffusion architecture that produces coherent images by applying the same diffusion model multiple times in parallel to collaboratively build and leverage a shared latent space for view consistency. Each viewpoint-specific inference generates an encoding representing its own hypothesis of the novel view from a given camera pose, which is projected onto three orthogonal planes. For each plane, encodings from all views are fused into a single aggregated plane. These aggregated planes are then processed to propagate information and interpolate missing regions, combining the hypotheses into a unified, coherent interpretation. The final latent space is then used to render consistent multi-view images. LoomNet generates 16 high-quality and coherent views in just 15 seconds. In our experiments, LoomNet outperforms state-of-the-art methods on both image quality and reconstruction metrics, also showing creativity by producing diverse, plausible novel views from the same input.

Figures

Figures reproduced from arXiv: 2507.05499 by the authors.

Figure 1
Figure 1. Given a set of desired novel perspectives of an input image, [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Extraction of rays from each model’s feature map for [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 2
Figure 2. Architecture overview. Given an input image and target viewpoints, each model independently encodes its view via a shared [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figures from the paper (7 more)
Figure 5
Figure 5. Figure 5: Each sampled point along the ray is enriched with the [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 4
Figure 4. Figure 4: Sampling along rays and splatting of points onto three [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 7
Figure 7. Figure 7: From raw learnable fusion planes to attentive fusion [PITH_FULL_IMAGE:figures/full_fig_p005_7.png]
Figure 6
Figure 6. Figure 6: Pixel-wise cross-attention between the N splatting planes (top) and the learnable fusion plane (center), producing the final fused plane (bottom) that encodes a consistent visual and ge￾ometric synthesis across views. 3.3. Weaving Stage The three learnable fusion plane…
Figure 8
Figure 8. Figure 8: Features sampled along each ray define a global context [PITH_FULL_IMAGE:figures/full_fig_p006_8.png]
Figure 9
Figure 9. Figure 9: Qualitative comparison in generating novel views from a single image. [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: Qualitative comparison in surface reconstruction from single image. [PITH_FULL_IMAGE:figures/full_fig_p008_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

54 extracted references · 28 canonical work pages

  1. [1]

    Ren- derdiffusion: Image diffusion for 3d reconstruction, inpaint- ing and generation

    Titas Anciukevi ˇcius, Zexiang Xu, Matthew Fisher, Paul Hen- derson, Hakan Bilen, Niloy J Mitra, and Paul Guerrero. Ren- derdiffusion: Image diffusion for 3d reconstruction, inpaint- ing and generation. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 12608–12618, 2023. 2

    work page 2023

  2. [2]

    Large-vocabulary 3d diffusion model with transformer

    Ziang Cao, Fangzhou Hong, Tong Wu, Liang Pan, and Ziwei Liu. Large-vocabulary 3d diffusion model with transformer. arXiv preprint arXiv:2309.07920, 2023. 2

    arXiv 2023

  3. [3]

    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 Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 16123–16133, 2022. 6

    work page 2022

  4. [4]

    Single-stage diffusion nerf: A unified approach to 3d generation and reconstruction

    Hansheng Chen, Jiatao Gu, Anpei Chen, Wei Tian, Zhuowen Tu, Lingjie Liu, and Hao Su. Single-stage diffusion nerf: A unified approach to 3d generation and reconstruction. In Proceedings of the IEEE/CVF international conference on computer vision, pages 2416–2425, 2023. 2

    work page 2023

  5. [5]

    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. 2

    2023

  6. [6]

    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. InPro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4456–4465, 2023. 2

    work page 2023

  7. [7]

    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. Advances in Neural Informa- tion Processing Systems, 36:35799–35813, 2023. 2, 6

    work page 2023

  8. [8]

    Google scanned objects: A high- quality dataset of 3d scanned household items

    Laura Downs, Anthony Francis, Nate Koenig, Brandon Kin- man, Ryan Hickman, Krista Reymann, Thomas B McHugh, and Vincent Vanhoucke. Google scanned objects: A high- quality dataset of 3d scanned household items. In 2022 In- ternational Conference on Robotics and Automation (ICRA), pages 2553–2560. IEEE, 2022. 6

    work page 2022

Show all 54 references
  1. [9]

    Hyperdiffusion: Generating implicit neu- ral fields with weight-space diffusion

    Ziya Erkoc ¸, Fangchang Ma, Qi Shan, Matthias Nießner, and Angela Dai. Hyperdiffusion: Generating implicit neu- ral fields with weight-space diffusion. In Proceedings of the IEEE/CVF international conference on computer vision, pages 14300–14310, 2023. 2

    2023

  2. [10]

    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

  3. [11]

    Control3diff: Learning con- trollable 3d diffusion models from single-view images

    Jiatao Gu, Qingzhe Gao, Shuangfei Zhai, Baoquan Chen, Lingjie Liu, and Josh Susskind. Control3diff: Learning con- trollable 3d diffusion models from single-view images. In 2024 International Conference on 3D Vision (3DV) , pages 685–696. IEEE, 2024. 2

    2024

  4. [12]

    Instant neural surface reconstruction, 2022

    Yuan-Chen Guo. Instant neural surface reconstruction, 2022. https://github.com/bennyguo/instant-nsr-pl. 6

    2022

  5. [13]

    3dgen: Triplane latent diffusion for textured mesh generation

    Anchit Gupta, Wenhan Xiong, Yixin Nie, Ian Jones, and Bar- las O˘guz. 3dgen: Triplane latent diffusion for textured mesh generation. arXiv preprint arXiv:2303.05371, 2023. 2

    2023 arXiv

  6. [14]

    Denoising dif- fusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. Advances in neural information processing systems, 33:6840–6851, 2020. 2

    2020

  7. [15]

    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, 2023. 6

    2023 arXiv

  8. [16]

    Epidiff: Enhancing multi-view synthesis via localized epipolar-constrained diffusion

    Zehuan Huang, Hao Wen, Junting Dong, Yaohui Wang, Yangguang Li, Xinyuan Chen, Yan-Pei Cao, Ding Liang, Yu Qiao, Bo Dai, et al. Epidiff: Enhancing multi-view synthesis via localized epipolar-constrained diffusion. In Proceedings of the IEEE/CVF Conference on Computer Vision and...

    2024

  9. [17]

    Zero-shot text-guided object genera- tion with dream fields

    Ajay Jain, Ben Mildenhall, Jonathan T Barron, Pieter Abbeel, and Ben Poole. Zero-shot text-guided object genera- tion with dream fields. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition , pages 867–876, 2022. 2

    2022

  10. [18]

    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. 2, 6, 7

    2023 arXiv

  11. [19]

    Neuralfield-ldm: Scene genera- tion with hierarchical latent diffusion models

    Seung Wook Kim, Bradley Brown, Kangxue Yin, Karsten Kreis, Katja Schwarz, Daiqing Li, Robin Rombach, Antonio Torralba, and Sanja Fidler. Neuralfield-ldm: Scene genera- tion with hierarchical latent diffusion models. In Proceed- ings of the IEEE/CVF conference on computer visio...

    2023

  12. [20]

    Diffusion- sdf: Text-to-shape via voxelized diffusion

    Muheng Li, Yueqi Duan, Jie Zhou, and Jiwen Lu. Diffusion- sdf: Text-to-shape via voxelized diffusion. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12642–12651, 2023. 2

    2023

  13. [21]

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

    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 optimiza- tion. Advances in Neural Information Processing Systems , 36:22226–22246, 2023. 3, 6, 7

    2023

  14. [22]

    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. 2, 3, 6, 7

    2023

  15. [23]

    Syncdreamer: Gen- erating multiview-consistent images from a single-view im- age

    Yuan Liu, Cheng Lin, Zijiao Zeng, Xiaoxiao Long, Lingjie Liu, Taku Komura, and Wenping Wang. Syncdreamer: Gen- erating multiview-consistent images from a single-view im- age. arXiv preprint arXiv:2309.03453, 2023. 2, 3, 6, 7

    2023 arXiv

  16. [24]

    Meshdif- fusion: Score-based generative 3d mesh modeling

    Zhen Liu, Yao Feng, Michael J Black, Derek Nowrouzezahrai, Liam Paull, and Weiyang Liu. Meshdif- fusion: Score-based generative 3d mesh modeling. arXiv preprint arXiv:2303.08133, 2023. 2

    2023 arXiv

  17. [25]

    Sparseneus: Fast generalizable neural sur- face reconstruction from sparse views

    Xiaoxiao Long, Cheng Lin, Peng Wang, Taku Komura, and Wenping Wang. Sparseneus: Fast generalizable neural sur- face reconstruction from sparse views. In European Confer- ence on Computer Vision , pages 210–227. Springer, 2022. 3

    2022

  18. [26]

    Diffusion probabilistic models for 3d point cloud generation

    Shitong Luo and Wei Hu. Diffusion probabilistic models for 3d point cloud generation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 2837–2845, 2021. 2

    2021

  19. [27]

    Controllable mesh generation through sparse latent point diffusion models

    Zhaoyang Lyu, Jinyi Wang, Yuwei An, Ya Zhang, Dahua Lin, and Bo Dai. Controllable mesh generation through sparse latent point diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 271–280, 2023. 2

    2023

  20. [28]

    Realfusion: 360deg reconstruction of any object from a single image

    Luke Melas-Kyriazi, Iro Laina, Christian Rupprecht, and Andrea Vedaldi. Realfusion: 360deg reconstruction of any object from a single image. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 8446–8455, 2023. 2

    2023

  21. [29]

    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. 2

    2021

  22. [30]

    Dit-3d: Exploring plain diffusion transformers for 3d shape generation

    Shentong Mo, Enze Xie, Ruihang Chu, Lanqing Hong, Matthias Niessner, and Zhenguo Li. Dit-3d: Exploring plain diffusion transformers for 3d shape generation. Advances in neural information processing systems, 36:67960–67971,

  23. [31]

    Diffrf: Rendering-guided 3d radiance field diffusion

    Norman M ¨uller, Yawar Siddiqui, Lorenzo Porzi, Samuel Rota Bulo, Peter Kontschieder, and Matthias Nießner. Diffrf: Rendering-guided 3d radiance field diffusion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 4328–4338, 2023. 2

    2023

  24. [32]

    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. 2, 6, 7

    2022 arXiv

  25. [33]

    Au- todecoding latent 3d diffusion models

    Evangelos Ntavelis, Aliaksandr Siarohin, Kyle Olszewski, Chaoyang Wang, Luc V Gool, and Sergey Tulyakov. Au- todecoding latent 3d diffusion models. Advances in Neural Information Processing Systems, 36:67021–67047, 2023. 2

    2023

  26. [34]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. In Proceedings of the IEEE/CVF inter- national conference on computer vision , pages 4195–4205,

  27. [35]

    Film: Visual reasoning with a general conditioning layer

    Ethan Perez, Florian Strub, Harm De Vries, Vincent Du- moulin, and Aaron Courville. Film: Visual reasoning with a general conditioning layer. In Proceedings of the AAAI con- ference on artificial intelligence, 2018. 5

    2018

  28. [36]

    Dreamfusion: Text-to-3d using 2d diffusion

    Ben Poole, Ajay Jain, Jonathan T Barron, and Ben Milden- hall. Dreamfusion: Text-to-3d using 2d diffusion. arXiv preprint arXiv:2209.14988, 2022. 2

    2022 arXiv

  29. [37]

    Magic123: One image to high-quality 3d object generation using both 2d and 3d diffusion priors.arXiv preprint arXiv:2306.17843,

    Guocheng Qian, Jinjie Mai, Abdullah Hamdi, Jian Ren, Aliaksandr Siarohin, Bing Li, Hsin-Ying Lee, Ivan Sko- rokhodov, Peter Wonka, Sergey Tulyakov, et al. Magic123: One image to high-quality 3d object generation using both 2d and 3d diffusion priors.arXiv preprint arXiv:2306.17843,

    arXiv

  30. [38]

    Dream- booth3d: Subject-driven text-to-3d generation

    Amit Raj, Srinivas Kaza, Ben Poole, Michael Niemeyer, Nataniel Ruiz, Ben Mildenhall, Shiran Zada, Kfir Aber- man, Michael Rubinstein, Jonathan Barron, et al. Dream- booth3d: Subject-driven text-to-3d generation. In Proceed- ings of the IEEE/CVF international conference on comp...

    2023

  31. [39]

    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. In Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4209–4219, 2024. 2

    2024

  32. [40]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 3

    2022

  33. [41]

    3d neural field generation using triplane diffusion

    J Ryan Shue, Eric Ryan Chan, Ryan Po, Zachary Ankner, Jiajun Wu, and Gordon Wetzstein. 3d neural field generation using triplane diffusion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 20875–20886, 2023. 6

    2023

  34. [42]

    Dreamcraft3d: Hierarchi- cal 3d generation with bootstrapped diffusion prior

    Jingxiang Sun, Bo Zhang, Ruizhi Shao, Lizhen Wang, Wen Liu, Zhenda Xie, and Yebin Liu. Dreamcraft3d: Hierarchi- cal 3d generation with bootstrapped diffusion prior. arXiv preprint arXiv:2310.16818, 2023. 2

    2023 arXiv

  35. [43]

    Make-it-3d: High-fidelity 3d creation from a single image with diffusion prior

    Junshu Tang, Tengfei Wang, Bo Zhang, Ting Zhang, Ran Yi, Lizhuang Ma, and Dong Chen. Make-it-3d: High-fidelity 3d creation from a single image with diffusion prior. In Pro- ceedings of the IEEE/CVF international conference on com- puter vision, pages 22819–22829, 2023. 2

    2023

  36. [44]

    Lion: Latent point dif- fusion models for 3d shape generation

    Arash Vahdat, Francis Williams, Zan Gojcic, Or Litany, Sanja Fidler, Karsten Kreis, et al. Lion: Latent point dif- fusion models for 3d shape generation. Advances in Neural Information Processing Systems, 35:10021–10039, 2022. 2

    2022

  37. [45]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. Advances in neural information processing systems, 30, 2017. 5

    2017

  38. [46]

    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. 2

    2023

  39. [47]

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

    Peng Wang, Lingjie Liu, Yuan Liu, Christian Theobalt, Taku Komura, and Wenping Wang. Neus: Learning neural implicit surfaces by volume rendering for multi-view reconstruction. arXiv preprint arXiv:2106.10689, 2021. 2

    2021 arXiv

  40. [48]

    Rodin: A generative model for sculpting 3d digital avatars using diffusion

    Tengfei Wang, Bo Zhang, Ting Zhang, Shuyang Gu, Jianmin Bao, Tadas Baltrusaitis, Jingjing Shen, Dong Chen, Fang Wen, Qifeng Chen, et al. Rodin: A generative model for sculpting 3d digital avatars using diffusion. In Proceedings of the IEEE/CVF conference on computer vision and...

    2023

  41. [49]

    Pushing the limits of 3d shape gener- ation at scale

    Yu Wang, Xuelin Qian, Jingyang Huo, Tiejun Huang, Bo Zhao, and Yanwei Fu. Pushing the limits of 3d shape gener- ation at scale. arXiv preprint arXiv:2306.11510, 2023. 2

    2023 arXiv

  42. [50]

    Image quality assessment: from error visibility to structural similarity

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Si- moncelli. Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4):600–612, 2004. 6

    2004

  43. [51]

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

    Zhengyi Wang, Cheng Lu, Yikai Wang, Fan Bao, Chongxuan Li, Hang Su, and Jun Zhu. Prolificdreamer: High-fidelity and diverse text-to-3d generation with variational score distilla- tion. Advances in Neural Information Processing Systems , 36:8406–8441, 2023. 2

    2023

  44. [52]

    Gaussiandreamer: Fast generation from text to 3d gaussians by bridging 2d and 3d diffusion models

    Taoran Yi, Jiemin Fang, Junjie Wang, Guanjun Wu, Lingxi Xie, Xiaopeng Zhang, Wenyu Liu, Qi Tian, and Xinggang Wang. Gaussiandreamer: Fast generation from text to 3d gaussians by bridging 2d and 3d diffusion models. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision...

    2024

  45. [53]

    Text-to-3d with classifier score distillation

    Xin Yu, Yuan-Chen Guo, Yangguang Li, Ding Liang, Song- Hai Zhang, and Xiaojuan Qi. Text-to-3d with classifier score distillation. arXiv preprint arXiv:2310.19415, 2023. 2

    2023 arXiv

  46. [54]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shecht- man, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 586–595, 2018. 6

    2018

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.