Pith. sign in

REVIEW 3 major objections 5 minor 60 references

HIVE-3D: hierarchical voxel refinement turns a single scene image into a high-resolution 3D scene while preserving the coarse layout.

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

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

T0 review · deepseek-v4-flash

2026-08-02 05:04 UTC pith:PUPDRBKP

load-bearing objection A plausible engineering contribution undermined by overclaiming and an unvalidated load-bearing lifting step. the 3 major comments →

arxiv 2607.13468 v1 pith:PUPDRBKP submitted 2026-07-15 cs.CV cs.LG

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

Bin Zang , Wenting Zheng , Xiaoliang Luo , Zhiyuan Fang , Shi Li , Lvchun Wang , Wei Yu , Yi Zhao
show 3 more authors
Tian Xie Yuchi Huo Rengan Xie
This is my paper
classification cs.CV cs.LG
keywords 3D scene generationsingle-image 3Dvoxel super-resolutionhierarchical generation2D-to-3D liftingcross-attentioncoarse-to-fine3D Gaussian splatting
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper tries to show that single-image 3D scene generation need not choose between a globally coherent layout and high-fidelity components. It proposes HIVE-3D, which starts from a coarse holistic scene, segments the 2D image, lifts those segments onto the 3D voxels using attention scores, organizes them into a hierarchical component tree, and then refines each component with a voxel super-resolution model conditioned on both the coarse voxel and an upscaled image. Recursing top-down and re-registering each refined piece yields a scene that is both coherent and detailed. The paper reports consistently better geometry and visual metrics than the compared methods and shows that quality increases with recursion depth.

Core claim

The central claim is that generative resolution can be upgraded at the component level without breaking the scene. The pipeline first builds a coarse scene with a pretrained structured-latent voxel generator, then exploits the fact that the generator's cross-attention maps already encode which image tokens correspond to which active voxels. Thresholding the aggregated attention over a 2D instance mask selects the voxels for that component, turning 2D segmentation into 3D segmentation. Each component is then regenerated by a lightweight adapter that injects the coarse voxel as an additional condition into the frozen generator, so the refined voxel inherits the coarse structure while adding de

What carries the argument

The hierarchical component tree and the voxel super-resolution adapter. The tree maps image regions to voxel subsets through averaged voxel-to-image cross-attention scores, softmax-normalized over image tokens, aggregated over masked tokens, and thresholded to select voxels. The adapter freezes the original sparse-structure generator, adds a trainable projection module and a dedicated cross-attention layer per block, and encodes coarse voxels with the pretrained sparse-structure VAE so the new condition lives in the same latent space as the diffusion process. Coarse-to-fine recursion avoids large resolution jumps that destabilize registration. Scale is estimated from mean point-to-centroid d

Load-bearing premise

The load-bearing premise is that thresholding the generator's attention scores cleanly separates each object's voxels from the rest of the scene; the paper reports no threshold value or accuracy for this segmentation, and its own failure analysis shows that errors in the initial scene or its parsing propagate into poor registration.

What would settle it

On a dataset with ground-truth 3D instance masks, run only the 2D-to-3D lifting stage and compare the thresholded voxel set for each image segment against the true instance voxels; if per-instance IoU is not high (say below 0.5 on average), the hierarchical refinement is built on misassigned components and the central claim fails.

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

If this is right

  • Scene generation from one image can reach per-object detail comparable to single-object generators while keeping the global layout.
  • Refinement depth becomes a tunable knob: more recursion levels produce finer geometry at a predictable runtime cost.
  • Because refinement is component-wise, the pipeline can in principle expand to larger scenes by deepening the tree.
  • The attention-based 2D-to-3D lifting means instance segmentation in 3D comes almost for free from the base generator's cross-attention, without extra 3D supervision.
  • Outlier-robust registration, rather than closest-point matching, is the right choice when coarse and fine components differ greatly in point density.

Where Pith is reading between the lines

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

  • The adapter design is not tied to one specific base generator; the same frozen-generator-plus-conditioning trick could stack multiple resolution upgrades or extend to other structured latents, making the hierarchy itself the reusable contribution.
  • The attention-threshold segmentation is the pivotal step; a direct validation study with ground-truth instance labels would either confirm or refute the whole pipeline, and a learned voxel-grouping head could replace the threshold if needed.
  • The component tree suggests an editing interface: refine, replace, or re-pose individual objects in a scene without regenerating the whole scene.
  • If the method transfers to multi-view or video input, the same tree structure could aggregate appearance from several viewpoints and enforce cross-view consistency during the diffusion steps, an extension the paper itself names as future work.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. HIVE-3D proposes a hierarchical, coarse-to-fine pipeline for single-image 3D scene generation. It first generates a coarse scene with TRELLIS, decomposes the scene by lifting 2D segmentations (Florence-2 + SAM2) into 3D voxel components using cross-attention from TRELLIS's GL transformer, and organizes these components into a scene tree. A voxel super-resolution model, trained with IP-Adapter-style adapters, refines each component conditioned on the coarse voxels and a super-resolved image. Refined components are rescaled via a centroid-distance proxy and registered back into the scene with RANSAC. Experiments on 3D-FRONT and real images report improved CD and F-Score over TRELLIS, but also show degraded IoU, SSIM, and LPIPS in some settings.

Significance. If the claims are substantiated, HIVE-3D would be a useful and well-engineered contribution to high-resolution 3D scene generation, showing how to reuse strong 2D segmentation and foundation 3D models in a modular pipeline. The paper is clearly written and includes detailed implementation and ablation sections, which supports reproducibility. The main strengths are the hierarchical voxel SR idea, the explicit system design, and the honest reporting of runtime and failure cases. However, the headline claim of state-of-the-art performance is not consistently supported by the paper's own quantitative results, and the load-bearing component-lifting step is not validated. The significance of the work depends on addressing these issues.

major comments (3)
  1. [Abstract and Sec. 5.2 / Tables 1 and 6] The abstract and Sec. 5.2 claim that HIVE-3D 'significantly outperforms previous approaches' and shows 'consistently strong performance across all metrics.' These claims are contradicted by the paper's own numbers. Table 1 reports IoU 0.7449 vs TRELLIS 0.8603, and Table 6 reports SSIM 0.79 vs 0.80 and LPIPS 0.33 vs 0.31 (both worse than TRELLIS). Only CD and F-Score improve. Please qualify the SOTA claim, report statistical significance, or provide a reasoned argument for why the IoU/SSIM/LPIPS regressions are acceptable. As written, the central claim is not supported.
  2. [Sec. 4.1 / A.1] The 2D-to-3D attention-based component lifting is a load-bearing step: every downstream refinement, scale estimate, and RANSAC registration operates on the voxels selected by thresholding aggregated GL cross-attention scores. The paper does not report the threshold value, does not validate that the selected voxels correspond to the intended image segment, and does not analyze sensitivity to the threshold. This step is borrowed from Fuse3D but applied here to full scenes, where attention maps may be noisier. A quantitative validation (e.g., voxel-to-mask IoU against ground-truth segments on a few 3D-FRONT scenes, with a threshold sweep) is necessary to support the pipeline. Appendix D concedes that failures in initial parsing propagate to registration.
  3. [Sec. 4.3 / A.2] Scale estimation uses the mean distance of surface points to the centroid as a 'robust proxy' for object scale, motivated by pose invariance. This is an unvalidated geometric assumption. The ablation in Table 4 shows the full model outperforms the variant without scale estimation, but it does not validate the proxy against ground-truth scale ratios or alternative estimators (e.g., bounding-box-based). Because an incorrect scale factor directly breaks the subsequent RANSAC registration, and the paper itself reports a severe registration drop when scale estimation is ablated, this assumption needs explicit validation for the method to be reliably applied.
minor comments (5)
  1. [Table 2] LPIPS worsens from depth 1 to depth 2/3 (0.38 to 0.41), while the text says quality 'progressively improves.' Please clarify or correct the interpretation.
  2. [Table 6 / Sec. 5.2] Table 6 is titled 'Quantitative Comparisons on 3D-FRONT and real data,' but the evaluation protocol for real images is not specified. How are reference images for SSIM/LPIPS/PSNR obtained on real data? Please clarify.
  3. [Sec. 4.1 / Figure 2] The abbreviation 'HYPIR' in Figure 2 is used without introduction; define it in the caption or the main text.
  4. [A.4] Minor typos: 'sparse structure V AE encoder' and 'V oxel-SR' contain inconsistent spacing; the title also has 'V oxel' with a space. These should be cleaned up for camera-ready.
  5. [Sec. C.5] The comparison with VIAFormer and ULTRA3D is qualitative and high-level; consider adding a brief quantitative comparison if the respective methods and evaluation settings are compatible, or at least state why quantitative comparison is omitted.

Circularity Check

0 steps flagged

No significant circularity: HIVE-3D's result is an empirical pipeline validated against external baselines; the Fuse3D attention-lifting dependency is a borrowed, falsifiable premise, not a self-referential derivation.

full rationale

The paper's claimed contribution is an empirical pipeline, not a mathematical derivation. The coarse scene is produced by TRELLIS; the hierarchical tree is built with external segmentation models (Florence-2, SAM2) and the Fuse3D attention-lifting heuristic; the voxel super-resolution model is a separately trained conditional generator whose objective (Eq. 2) is the standard flow-matching loss, not a quantity defined by the evaluation metrics; scale and pose are estimated from the two geometries being aligned and are not presented as independent predictions. Quantitative comparisons (Tables 1 and 6) are made against external baselines on 3D-FRONT and real images, and the ablations (Tables 2-4, 7) test components rather than fitting them to the test set. The only circularity-adjacent issue is that the 2D-to-3D attention correspondence is attributed to Fuse3D (Jin et al., 2025), whose authors overlap with this paper (Xie, Zheng, Huo). However, this is a borrowed, externally falsifiable premise about TRELLIS cross-attention behavior, not a claim whose output is defined as its input, and no equation in the paper reduces a prediction to a fitted value or to the evaluation target. The paper explicitly acknowledges limitations from 2D segmentation errors (Sec. 6) and TRELLIS-bound failures (Appendix D); these are robustness/validity gaps, not equivalence-by-construction. Under the strict standard requiring a quoted reduction, no circular step is present.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The central pipeline rests on pretrained TRELLIS and segmentation priors purchased upstream; the paper's own additions are mostly hand-chosen thresholds/hyperparameters for lifting and registration. The most consequential unexamined parameter is the attention threshold for voxel selection.

free parameters (4)
  • Attention threshold for voxel selection
    Section 4.1/A.1: 'By applying a threshold to these aggregated scores, we select a set of voxel indices...' No threshold value or sensitivity analysis is given; it determines which voxels belong to each component and is load-bearing.
  • Cross-attention heads used for lifting = 0, 4, 12
    A.1: averaging attention maps from heads 0, 4, and 12; chosen by hand, no justification.
  • Voxel SR training hyperparameters = lr 1e-4, wd 1e-2, batch 24, 100k iters, EMA 0.9999, dropout 10%
    A.4: standard but hand-chosen; affects quality of learned adapter.
  • RANSAC registration hyperparameters = voxel size 0.0156, FPFH radius 0.0780, max corr 0.0234, 100k iterations, conf 0.999
    A.3: hand-chosen; registration success is a stated bottleneck; no sensitivity analysis.
axioms (5)
  • domain assumption TRELLIS's GL flow transformer learns cross-attention correspondences between image tokens and 3D voxels that can be used for segmentation.
    Section 4.1, borrowed from Fuse3D (Jin et al. 2025), authored in part by this paper's authors; neither reproduced nor formally justified here.
  • domain assumption Off-the-shelf Florence-2 and SAM2 produce correct hierarchical 2D masks.
    Section 4.1: detection and segmentation errors are acknowledged as a limitation in the conclusion.
  • domain assumption A frozen TRELLIS GS with new cross-attention layers can learn to condition on coarse voxel latents without catastrophic forgetting.
    Section 4.2: no analysis of distribution shift; relies on IP-Adapter analogy.
  • ad hoc to paper Mean distance-to-centroid is a reliable scale proxy invariant to pose and detail changes.
    A.2: 'this distance serves as a robust proxy for overall size'; used to compute scale factor s = d/d'.
  • domain assumption Objaverse-XL 10,000-asset subset is representative for scene components.
    Section 5.1: curation criteria ('excluding excessive mesh complexity') may bias toward certain object types.

pith-pipeline@v1.3.0-alltime-deepseek · 16065 in / 12667 out tokens · 113741 ms · 2026-08-02T05:04:50.688065+00:00 · methodology

0 comments
read the original abstract

Recently, a line of works can generate impressive 3D objects from a single image, but they are limited by restricted representation resolution, making them unsuitable for 3D scene generation. In this work, we introduce HIVE-3D, a novel method for high-quality 3D scene generation based on hierarchical voxel enhancement framework. Specifically, given a single scene image as input, we first produce a coarse initial scene, then introduce image segmentation and attention-based retrieval to align 2D image components with 3D scene components. Subsequently, we organize these scene relations into a hierarchical component tree, where nodes closer to the leaves denote finer-grained components. Finally, we propose a voxel super-resolution model that generates refined voxels for the target instance while maintaining strong consistency with the coarse voxels. Equipped with this model, we perform coarse-to-fine hierarchical super-resolution on images and voxels for each component, producing a high-resolution and high-quality 3D scene. Extensive experiments demonstrate that our method significantly outperforms previous approaches, achieving state-of-the-art performance.

Figures

Figures reproduced from arXiv: 2607.13468 by Bin Zang, Lvchun Wang, Rengan Xie, Shi Li, Tian Xie, Wei Yu, Wenting Zheng, Xiaoliang Luo, Yi Zhao, Yuchi Huo, Zhiyuan Fang.

Figure 1
Figure 1. Figure 1: HIVE-3D generates high-resolution, high-fidelity 3D scenes from a single image using a hierarchical voxel enhancement framework within a coarse-to-fine scheme. Abstract Recently, a line of works can generate impressive 3D objects from a single image, but they are lim￾ited by restricted representation resolution, mak￾ing them unsuitable for 3D scene generation. In this work, we introduce HIVE-3D, a novel me… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of HIVE-3D. Our method first produces a coarse 3D scene. We then construct a hierarchical scene tree by decomposing the scene and aligning 2D image parts with 3D voxel components using segmentation and attention-based retrieval. To enhance detail, we propose a voxel super-resolution model that refines each component while maintaining consistency with its coarser representation. Finally, guided by … view at source ↗
Figure 3
Figure 3. Figure 3: The network structures for voxel super-resolution model. different granularities along with the image description. Yet the current tree contains only initial scene voxels, which are relatively coarse. Enhancing their resolution while preserv￾ing global consistency remains challenging. 4.2 Voxel Super-resolution Model In this section, we introduce a method to enhance the resolu￾tion and quality of coarse sc… view at source ↗
Figure 4
Figure 4. Figure 4: Generation quality comparison with previous methods(Zoom in for details). rotation R and translation t. A critical challenge is the severe imbalance in the number of Gaussians between the finer-grained child components and its low-resolution coun￾terpart, which induces substantial mismatches and outliers. To address this, we adopt a registration strategy that em￾phasizes robustness to outliers rather than … view at source ↗
Figure 5
Figure 5. Figure 5: Scene generation results under different recursion depths(Zoom in for details). The qualitative results are presented in [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Impact of coarse voxels on the generation of instance voxels [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The effect of the scale factor on scene [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: More results on synthetic data. repeatedly samples such minimal correspondence sets, generates candidate rigid transformation hypotheses, and selects the transformation with the largest set of geometrically consistent inlier correspondences. The output of this process is a rigid transformation matrix T =  R t 0 ⊤ 1  ∈ SE(3), where R ∈ SO(3) is the estimated rotation matrix and t ∈ R 3 is the translation … view at source ↗
Figure 9
Figure 9. Figure 9: Comparison between the direct two-layer partitioning and our multi-level recursive refinement. C.3 Qualitative Comparison with PartPacker We conduct qualitative comparisons between our method and PartPacker (Tang et al., 2026). It should be noted that PartPacker primarily focuses on geometric synthesis and generates only untextured meshes. As illustrated in [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative comparison with PartPacker. C.4 Qualitative Comparison with SAM3D We conduct qualitative comparisons between SAM3D (Chen et al., 2025a) and our method in terms of 3D scene generation quality. Since SAM3D performs holistic scene generation, its results are constrained by the overall scene resolution. As shown in the first row of [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Qualitative comparison with SAM3D. C.5 Comparison with VIAFormer and ULTRA3D Although VIAFormer (Fang et al., 2026) and ULTRA3D (Chen et al., 2025b) are also related to structured 3D generation, our method differs substantially from these approaches in both formulation and design objectives. VIAFormer primarily focuses on denoising and refining existing occupancy representations under a fixed and stable s… view at source ↗
Figure 12
Figure 12. Figure 12: Representative failure case of our pipeline [PITH_FULL_IMAGE:figures/full_fig_p019_12.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

60 extracted references · 1 canonical work pages

  1. [1]

    CoRR , volume =

    Alec Radford and Jong Wook Kim and Chris Hallacy and Aditya Ramesh and Gabriel Goh and Sandhini Agarwal and Girish Sastry and Amanda Askell and Pamela Mishkin and Jack Clark and Gretchen Krueger and Ilya Sutskever , title =. CoRR , volume =. 2021 , url =. 2103.00020 , timestamp =

    Pith/arXiv arXiv 2021

  2. [2]

    CVPR , year=

    The Unreasonable Effectiveness of Deep Features as a Perceptual Metric , author=. CVPR , year=

  3. [3]

    FirstName LastName , title =

  4. [4]

    FirstName Alpher , title =

  5. [5]

    Journal of Foo , volume = 13, number = 1, pages =

    FirstName Alpher and FirstName Fotheringham-Smythe , title =. Journal of Foo , volume = 13, number = 1, pages =

  6. [6]

    Journal of Foo , volume = 14, number = 1, pages =

    FirstName Alpher and FirstName Fotheringham-Smythe and FirstName Gamow , title =. Journal of Foo , volume = 14, number = 1, pages =

  7. [7]

    FirstName Alpher and FirstName Gamow , title =

  8. [8]

    ACM Transactions on Graphics (TOG) , volume=

    Diffcad: Weakly-supervised probabilistic cad model retrieval and alignment from an rgb image , author=. ACM Transactions on Graphics (TOG) , volume=. 2024 , publisher=

    2024

  9. [9]

    2024 International Conference on 3D Vision (3DV) , pages=

    Single-view 3d scene reconstruction with high-fidelity shape and texture , author=. 2024 International Conference on 3D Vision (3DV) , pages=. 2024 , organization=

    2024

  10. [10]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Buol: A bottom-up framework with occupancy-aware lifting for panoptic 3d scene reconstruction from a single image , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  11. [11]

    Advances in Neural Information Processing Systems , volume=

    Panoptic 3d scene reconstruction from a single rgb image , author=. Advances in Neural Information Processing Systems , volume=

  12. [12]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Learning 3d object shape and layout without 3d supervision , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  13. [13]

    European Conference on Computer Vision , pages=

    Towards high-fidelity single-view holistic reconstruction of indoor scenes , author=. European Conference on Computer Vision , pages=. 2022 , organization=

    2022

  14. [14]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Total3dunderstanding: Joint layout, object pose and mesh reconstruction for indoor scenes from a single image , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  15. [15]

    Advances in neural information processing systems , volume=

    Atiss: Autoregressive transformers for indoor scene synthesis , author=. Advances in neural information processing systems , volume=

  16. [16]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Holistic 3d scene understanding from a single image with implicit representation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  17. [17]

    Uni-3D: A Universal Model for Panoptic 3D Scene Reconstruction , year=

    Zhang, Xiang and Chen, Zeyuan and Wei, Fangyin and Tu, Zhuowen , booktitle=. Uni-3D: A Universal Model for Panoptic 3D Scene Reconstruction , year=

  18. [18]

    ROCA: Robust CAD Model Retrieval and Alignment from a Single Image , url=

    Gumeli, Can and Dai, Angela and Niebner, Matthias , year=. ROCA: Robust CAD Model Retrieval and Alignment from a Single Image , url=. doi:10.1109/cvpr52688.2022.00399 , booktitle=

    arXiv 2022

  19. [19]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Im2cad , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  20. [20]

    European Conference on Computer Vision , pages=

    Mask2cad: 3d shape prediction by learning to segment and retrieve , author=. European Conference on Computer Vision , pages=. 2020 , organization=

    2020

  21. [21]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Patch2cad: Patchwise embedding learning for in-the-wild shape retrieval from a single image , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  22. [22]

    arXiv preprint arXiv:2210.01044 , year=

    Sparc: Sparse render-and-compare for cad model alignment in a single rgb image , author=. arXiv preprint arXiv:2210.01044 , year=

    Pith/arXiv arXiv

  23. [23]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Structured 3d latents for scalable and versatile 3d generation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  24. [24]

    ACM Transactions on Graphics (TOG) , volume=

    Cast: Component-aligned 3d scene reconstruction from an rgb image , author=. ACM Transactions on Graphics (TOG) , volume=. 2025 , publisher=

    2025

  25. [25]

    European Conference on Computer Vision , pages=

    Comboverse: Compositional 3d assets creation using spatially-aware diffusion guidance , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  26. [26]

    2025 International Conference on 3D Vision (3DV) , pages=

    Gen3dsr: Generalizable 3d scene reconstruction via divide and conquer from a single view , author=. 2025 International Conference on 3D Vision (3DV) , pages=. 2025 , organization=

    2025

  27. [27]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Reparo: Compositional 3d assets generation with differentiable 3d layout alignment , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  28. [28]

    Advances in Neural Information Processing Systems , volume=

    Zero-shot scene reconstruction from single images with deep prior assembly , author=. Advances in Neural Information Processing Systems , volume=

  29. [29]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Midi: Multi-instance diffusion for single image to 3d scene generation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  30. [30]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    3d-front: 3d furnished rooms with layouts and semantics , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  31. [31]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Adding conditional control to text-to-image diffusion models , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  32. [32]

    , author=

    Lora: Low-rank adaptation of large language models. , author=. Iclr , volume=

  33. [33]

    Proceedings of the 33rd ACM International Conference on Multimedia , pages=

    Hiscene: creating hierarchical 3d scenes with isometric view generation , author=. Proceedings of the 33rd ACM International Conference on Multimedia , pages=

  34. [34]

    arXiv preprint arXiv:2508.15769 , year=

    Scenegen: Single-image 3d scene generation in one feedforward pass , author=. arXiv preprint arXiv:2508.15769 , year=

    arXiv

  35. [35]

    arXiv preprint arXiv:2308.06721 , year=

    Ip-adapter: Text compatible image prompt adapter for text-to-image diffusion models , author=. arXiv preprint arXiv:2308.06721 , year=

    Pith/arXiv arXiv

  36. [36]

    arXiv preprint arXiv:2405.14979 , year=

    Craftsman3d: High-fidelity mesh generation with 3d native generation and interactive geometry refiner , author=. arXiv preprint arXiv:2405.14979 , year=

    Pith/arXiv arXiv

  37. [37]

    Advances in Neural Information Processing Systems , volume=

    Direct3d: Scalable image-to-3d generation via 3d latent diffusion transformer , author=. Advances in Neural Information Processing Systems , volume=

  38. [38]

    ACM Transactions on Graphics (TOG) , volume=

    Clay: A controllable large-scale generative model for creating high-quality 3d assets , author=. ACM Transactions on Graphics (TOG) , volume=. 2024 , publisher=

    2024

  39. [39]

    Advances in neural information processing systems , volume=

    Michelangelo: Conditional 3d shape generation based on shape-image-text aligned latent representation , author=. Advances in neural information processing systems , volume=

  40. [40]

    Advances in neural information processing systems , volume=

    Denoising diffusion probabilistic models , author=. Advances in neural information processing systems , volume=

  41. [41]

    ACM Transactions on Graphics (TOG) , volume=

    Harnessing diffusion-yielded score priors for image restoration , author=. ACM Transactions on Graphics (TOG) , volume=. 2025 , publisher=

    2025

  42. [42]

    arXiv preprint arXiv:2304.07193 , year=

    Dinov2: Learning robust visual features without supervision , author=. arXiv preprint arXiv:2304.07193 , year=

    Pith/arXiv arXiv

  43. [43]

    Advances in Neural Information Processing Systems , volume=

    Objaverse-xl: A universe of 10m+ 3d objects , author=. Advances in Neural Information Processing Systems , volume=

  44. [44]

    Fischler and Robert C

    Martin A. Fischler and Robert C. Bolles , abstract =. Random Sample Consensus: A Paradigm for Model Fitting with Applications to Image Analysis and Automated Cartography , editor =. Readings in Computer Vision , publisher =. 1987 , isbn =. doi:https://doi.org/10.1016/B978-0-08-051581-6.50070-2 , url =

    work page doi:10.1016/b978-0-08-051581-6.50070-2 1987

  45. [45]

    IEEE Transactions on Pattern Analysis & Machine Intelligence , volume=

    A method for registration of 3-D shapes , author=. IEEE Transactions on Pattern Analysis & Machine Intelligence , volume=

  46. [46]

    Proceedings of the SIGGRAPH Asia 2025 Conference Papers , pages=

    Fuse3D: Generating 3D Assets Controlled by Multi-Image Fusion , author=. Proceedings of the SIGGRAPH Asia 2025 Conference Papers , pages=

    2025

  47. [47]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Florence-2: Advancing a unified representation for a variety of vision tasks , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  48. [48]

    International Conference on Learning Representations , volume=

    Sam 2: Segment anything in images and videos , author=. International Conference on Learning Representations , volume=

  49. [49]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Amodal3r: Amodal 3d reconstruction from occluded 2d images , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  50. [50]

    arXiv:1801.09847 , year =

    Qian-Yi Zhou and Jaesik Park and Vladlen Koltun , title =. arXiv:1801.09847 , year =

    Pith/arXiv arXiv

  51. [51]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    A point set generation network for 3d object reconstruction from a single image , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  52. [52]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    What do single-view 3d reconstruction networks learn? , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  53. [53]

    European conference on computer vision , pages=

    3d-r2n2: A unified approach for single and multi-view 3d object reconstruction , author=. European conference on computer vision , pages=. 2016 , organization=

    2016

  54. [54]

    arXiv preprint arXiv:1711.05101 , year=

    Decoupled weight decay regularization , author=. arXiv preprint arXiv:1711.05101 , year=

    Pith/arXiv arXiv

  55. [55]

    arXiv preprint arXiv:2207.12598 , year=

    Classifier-free diffusion guidance , author=. arXiv preprint arXiv:2207.12598 , year=

    Pith/arXiv arXiv

  56. [56]

    SIAM journal on control and optimization , volume=

    Acceleration of stochastic approximation by averaging , author=. SIAM journal on control and optimization , volume=. 1992 , publisher=

    1992

  57. [57]

    Advances in Neural Information Processing Systems , volume=

    Efficient part-level 3d object generation via dual volume packing , author=. Advances in Neural Information Processing Systems , volume=

  58. [58]

    arXiv preprint arXiv:2511.16624 , year=

    Sam 3d: 3dfy anything in images , author=. arXiv preprint arXiv:2511.16624 , year=

    Pith/arXiv arXiv

  59. [59]

    arXiv preprint arXiv:2601.13664 , year=

    VIAFormer: Voxel-Image Alignment Transformer for High-Fidelity Voxel Refinement , author=. arXiv preprint arXiv:2601.13664 , year=

    arXiv

  60. [60]

    arXiv preprint arXiv:2507.17745 , year=

    Ultra3d: Efficient and high-fidelity 3d generation with part attention , author=. arXiv preprint arXiv:2507.17745 , year=

    Pith/arXiv arXiv