Pith. sign in

REVIEW 4 major objections 5 minor 59 references

Scene-SAM3D: Multi-View Scene Asset Generation Without Fine-Tuning

T0 review · 4 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read Scene-SAM3D turns a frozen single-view 3D generator into a multi-view scene generator without fine-tuning.

desk verdict A solid training-free multi-view extension of SAM3D with a genuinely new selection–fusion–alignment combination and honest component ablations, but the abstract overstates the headline comparison and the method silently depends on depth it never specifies. read the letter →

arxiv 2607.16805 v1 pith:5SMIX3WY submitted 2026-07-18 cs.CV

Yuqi Zhang , Yadan Luo , Xiangyu Sun , Fengyi Zhang , Zi Huang , Xin Tan This is my paper · ORCID
classification cs.CV
keywords multi-view3DgenerationsceneassetviewselectionvelocityfusionGaussianlayoutrefinementtraining-freeflowmatchingcalibratedcameras
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

The paper argues that calibrated multi-view scene asset generation can be solved by composing three training-free modules on top of a frozen single-view object generator: occupancy-coverage view selection, early-stop sign-consensus velocity fusion, and rigid-object Gaussian layout refinement. The central claim is that this selection–fusion–alignment decomposition lets a strong object prior (SAM3D) produce complete, layout-coherent indoor scenes despite occlusion, redundant observations, and cross-view inconsistencies. On two indoor datasets, the paper reports instance-level gains in Chamfer distance and completeness over single-view and MultiDiffusion baselines, scene-level CD reductions of 43.8% on Replica and 30.9% on ScanNet++ when the proposed Gaussian refinement replaces its own ICP alignment, and roughly 20% savings in flow-model FLOPs and wall-time. A sympathetic reader would care because the recipe is parameter-free: it upgrades existing feed-forward 3D generators to multi-view capture without retraining.

What carries the argument

The load-bearing machinery is a set of three operators. (1) Occupancy-coverage view selection (Eqs. 1–3): masked depth points are voxelized in a shared world frame; the anchor maximizes merged occupied volume and each helper greedily adds the most uncovered voxels, so views are judged by 3D coverage rather than 2D mask area. (2) Step-efficient sign-consensus fusion (Eqs. 8–11): helper-view velocities are computed only before a midpoint Tmid, and within each latent cell only velocities whose sign matches the anchor (in anchor-observed cells) or a majority vote (elsewhere) are averaged, preventing incompatible canonical shapes from being mixed. (3) Rigid-object Gaussian layout refinement (Eqs.

What would settle it

Run Scene-SAM3D on a calibrated RGB-only capture without dataset-provided depth or masks and measure scene-level Chamfer distance against single-view and MultiDiffusion baselines; if the occupancy-based view selection and Gaussian layout refinement no longer help, the central claim fails. Alternatively, reproduce the scene-level comparison with the baseline stated in the abstract — SAM3D with ICP alignment using the same coverage-selected views — and check whether the 43.8%/30.9% reductions still hold when the reference is SAM3D+ICP rather than the pipeline's own ICP refinement.

Watch

Extended reading notes

Core claim

The central discovery is that multi-view evidence can be exploited in a frozen flow-matching generator without any training: pick one anchor view that maximizes world-space occupancy coverage, add helper views that add new voxels, fuse their velocity fields only in early sampling steps and only where per-cell signs agree with the anchor (or with a majority vote in anchor-unobserved regions), then leave geometry untouched and optimize only each object's rotation, scale, and translation in a differentiable Gaussian proxy against multi-view masks, RGB, depth, boundaries, and inter-object penalties. The paper claims this suppresses duplicated or layered structures from uniform velocity averaging

Load-bearing premise

The pipeline assumes a reliable per-pixel depth map D_i and an instance mask are available for every view used in view selection and layout refinement; the paper never specifies how depth is obtained for arbitrary calibrated RGB video, since on the two benchmark datasets it is read directly from the data.

Editorial extensions

If this is right

  • A single-view 3D generator can produce complete scene assets from calibrated multi-view video without any fine-tuning, as long as depth and masks are available.
  • Occupancy-based view selection improves reconstruction even for single-view baselines, indicating that world-space 3D coverage is a better view-quality signal than 2D mask size.
  • Early-stop, sign-consensus fusion avoids the duplicated or layered geometry that uniform multi-view velocity averaging produces, while still recovering anchor-unobserved regions.
  • Refining only object-level similarity transforms (not geometry) yields large scene-layout gains — 43.8%/30.9% lower scene-level CD and 40.8%/42.5% higher IoU — and converges within 200 iterations.
  • Under ten-view inference, the step-efficient fusion reduces flow-model FLOPs and wall-time by nearly 20% versus exhaustive multi-view fusion.

Reading between the lines

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

  • The same selection–fusion–alignment split is a general recipe: any flow- or diffusion-based single-view 3D generator could be wrapped in these three stages, so the contribution is not tied to SAM3D specifically.
  • Because the experiments read depth directly from the datasets, a natural next test is to substitute monocular depth estimation and run the pipeline on RGB-only footage; if the gains survive, the method becomes applicable to casual phone or robot captures.
  • The decoupling of shape from layout implies the Gaussian refinement stage could be reused as a post-process for any object-centric generator, independent of the fusion module that precedes it.
  • The sign-consensus rule suggests that spatially selective averaging of velocity fields (rather than uniform averaging) could improve other multi-condition latent-space generation tasks, such as text-guided or multi-image 3D generation.
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. Scene-SAM3D proposes a training-free pipeline that adapts the frozen single-image generator SAM3D to calibrated multi-view scene asset generation. The method selects a compact anchor-helper view set via world-space occupancy coverage, fuses helper-view latent velocities only in early flow steps with an anchor-guided sign-consensus operator, and then refines the object-level rotations, scales, and translations through a rigid-object Gaussian layout optimization. Experiments on Replica and ScanNet++ report instance-level improvements over SAM3D and a MultiDiffusion-based baseline, and large scene-level CD reductions attributed to the Gaussian layout refinement. The paper also reports roughly 20% savings in flow-model FLOPs and latency under ten-view inference.

Significance. If the claims are reproducible, the paper offers a practically useful design pattern for lifting object-centric 3D generators to multi-view scenes: view selection by 3D occupancy, early-stop conflict-aware fusion, and layout refinement via a frozen Gaussian proxy are all sensible, and the component-wise ablations (Tables 1, 3, 5) support the internal design choices. However, the headline scene-level claim is not supported by the tables as currently worded, and the method's unstated dependence on per-pixel depth is a load-bearing gap. With corrections to the comparisons, explicit handling of depth, and proper statistical reporting, the contribution would be significant for the 3D generation community.

major comments (4)
  1. [Abstract & Sec. 4.1, Tables 2, 6] The abstract states that Scene-SAM3D reduces scene-level CD by 43.8% on Replica and 30.9% on ScanNet++ 'over SAM3D with ICP alignment.' In Tables 2 and 6, the rows compared are 'Ours ICP' and 'Ours GS' under the same coverage-selected views and fusion strategy. On Replica, (8.64-4.86)/8.64 = 0.4375; on ScanNet++, (6.89-4.76)/6.89 = 0.3091. Thus the percentages are an ablation of the GS layout refinement against the authors' own ICP baseline, not a comparison with SAM3D+ICP. The claim as worded is unsupported. Please either add the missing SAM3D+ICP comparison or change the claim to 'compared with our ICP-based refinement.'
  2. [Sec. 2 and Sec. 3, Eq. (1), Eq. (7), Eq. (13)] The problem definition lists observations as (I_i, M_i, C_i, K) with no depth map, and the paper frames the task as 'calibrated multi-view scene asset generation.' However, D_i is used in Eq. (1) to build world-space point sets for view selection, in Eq. (7) to map anchor pixels into the canonical latent grid, and in Eq. (13) as L_depth supervision in layout refinement. The source of D_i is never specified; on Replica and ScanNet++ it is the dataset ground-truth depth. Without a depth estimator, a practitioner with RGB-only calibrated video cannot execute view selection or layout refinement, and the reported results are conditional on an input the method does not state. Please specify and evaluate the depth source, or explicitly scope the method and claims to RGB-D input.
  3. [Sec. 4.2 and Appendix B (hyperparameters)] The implementation details in Appendix B set M=3, Tmid=15, and the layout loss coefficients; Sec. 4.2 states that the fusion-step ablation was conducted on Replica. The main Replica results in Tables 1-2 are on the same dataset. If the same scenes were used to choose Tmid and the loss weights, the reported improvements are partially a result of tuning on the evaluation set. Please report a separate validation split or show that the results are stable across a range of hyperparameters.
  4. [All tables] No standard deviations or number of seeds are reported for any experiment. The generation process is stochastic (flow matching with random initialization), and several reported differences are small (e.g., CD 39.37 vs 40.98 in Table 1, NC 75.17 vs 74.86). Please report at least three seeds with mean and std, and state whether the differences are significant.
minor comments (5)
  1. [Appendix B] The two lists of loss coefficients are inconsistent. The first list gives λb=2.0 and λp=0.04; the second gives λb=0.04 and λp=0.5. Eq. (13) also includes L_reg with no coefficient. Please unify.
  2. [Sec. 3.2, Eq. (10)-(11)] If the sum of signs in Eq. (10) is zero, Iunobs(u) may be empty and the average in Eq. (11) is undefined. Add a tie-breaking rule (e.g., fall back to the anchor or to uniform averaging).
  3. [Fig. 2 caption] 'Based on the selected views, its performs step-efficient...' should read 'it performs.'
  4. [Table 4] Please specify whether the reported TFLOPs and time are for the full pipeline or only the flow model, and state the view-selection setting. The text says 'flow-model sampling' but the table header lacks units and scope.
  5. [Sec. 4.1] The sentence 'Compared with ICP, GS reduces CD by 43.8%/30.9%' is correct as an ablation, but it conflicts with the abstract's wording. Please harmonize the claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the pipeline is a composition over a frozen external model; no reported metric is fixed by construction from its inputs.

full rationale

The derivation chain is self-contained in the circularity sense. Scene-SAM3D consumes a frozen, externally trained SAM3D model plus calibrated multi-view observations and composes three modules: occupancy-coverage selection (Eqs. 1-3), step-efficient sign-consensus fusion (Eqs. 4-11), and rigid-object Gaussian layout refinement (Eqs. 12-13). No equation is definitionally identical to a reported metric. Eq. 1 uses a depth map D_i to build world-space point sets, but Chamfer Distance, Completeness, and F-score are computed on downstream generated geometry against GT meshes, so selecting high-coverage views does not by itself set the evaluation numbers; the frozen SAM3D prior and the fusion step sit between selection and evaluation. Eq. 7 maps anchor pixels into the canonical latent grid using the SAM3D-predicted transform, but that transform is an intermediate latent-space quantity, not the final evaluated scene layout. Eq. 13 optimizes object-level RST parameters against multi-view masks/depth/RGB renderings, and scene-level CD is subsequently measured on GT meshes; this is test-time optimization, not a fitted parameter renamed as a prediction. There are no load-bearing self-citations: the paper references external work (SAM3D, MultiDiffusion, TRELLIS, etc.) and invokes no uniqueness theorem from the authors' own prior work. The abstract's '43.8%/30.9% over SAM3D with ICP' claim is not actually supported by Tables 2 and 6, where the reductions correspond to Ours-GS vs Ours-ICP (8.64 to 4.86 on Replica; 6.89 to 4.76 on ScanNet++), and the method assumes per-pixel depth in Eqs. 1, 7, and 13 although the Sec. 2 problem definition lists only (I_i, M_i, C_i, K). Those are correctness/scope issues, not circularity. Hyperparameters such as Tmid=15 and loss weights were chosen by ablations on Replica and reused on the same benchmark, which is a mild evaluation-set tuning concern but not a circular reduction, since the ablated values do not reconstruct the reported metrics by definition.

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

The method relies on the frozen SAM3D model and dataset-provided RGB-D evidence; no new physical entities are introduced. The main ledger items are hand-set hyperparameters (M, Tmid, loss weights) and implicit domain assumptions about depth/mask availability and canonical-space compatibility of velocity fields.

free parameters (4)
  • helper view count M = 3
    Appendix B sets M=3; no sensitivity analysis is reported.
  • fusion midpoint Tmid = 15
    Chosen from the Fig. 6 ablation on Replica and used for the main Replica results.
  • layout loss weights = lambda_m=2.0, lambda_rgb=2.5, lambda_d=2.0, lambda_b=0.04, lambda_c=0.05, lambda_p=0.5 (lambda_r=0.05 listed separately)
    Appendix B lists weights with inconsistent duplicate entries; no sensitivity study is provided.
  • Gaussian proxy downsample size / voxel resolution = not stated
    The downsampled object-centric Gaussian proxy Pk and the voxel grid used for anchor-observed regions are unspecified; both affect fusion and refinement behavior.
assumptions (4)
  • domain assumption Per-view depth maps D_i and instance masks are available for every observation used in selection and refinement.
    Eq. (1) projects masked depth into world occupancy and Section 3.3 supervises with masks, RGB, and depth. The paper never specifies a depth estimator for RGB-only captures.
  • domain assumption Calibrated camera poses C_i and shared intrinsics K are known.
    The problem definition in Section 2 assumes calibrated multi-view video; real captures may have pose drift that is not modeled.
  • domain assumption SAM3D's predicted camera-relative transform (R^c, s^c, t^c) is accurate enough to define anchor-observed canonical regions.
    Eq. (7) maps anchor pixels to canonical object frame using SAM3D's Stage-1 pose; if the pose is wrong, the sign-consensus masking is misaligned.
  • ad hoc to paper Multi-view conditional velocities live in a shared canonical latent grid where per-cell sign agreement separates compatible and conflicting evidence.
    Fuse(.) in Section 3.2 operates elementwise on the 16x16x16xd structured velocity; no theoretical guarantee is given that velocity fields from different views are comparable in this grid.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Scene-SAM3D: Multi-View Scene Asset Generation Without Fine-Tuning." pith.science (2026). https://pith.science/paper/5SMIX3WY

@misc{pith2026260716805,
  author       = {Pith},
  title        = {Pith review of: Scene-SAM3D: Multi-View Scene Asset Generation Without Fine-Tuning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5SMIX3WY}},
  note         = {Machine review of arXiv:2607.16805}
}
read the original abstract

High-quality 3D scene assets are critical for embodied applications such as robotic manipulation, navigation, and simulation. Despite their strong object priors, recent single-image 3D generation models such as SAM3D remain insufficient for real-world scenes, where severe occlusions, redundant observations, and cross-view inconsistencies make reliable scene generation challenging. We introduce Scene-SAM3D, a training-free framework that extends SAM3D from single-view object generation to calibrated multi-view scene asset generation. Scene-SAM3D selects a compact set of complementary views, reducing observation redundancy while providing additional evidence for regions occluded in individual views. Based on the selected views, it performs step-efficient latent velocity fusion to integrate multi-view evidence and suppress cross-view conflicts in canonical space. Finally, a lightweight rigid-object Gaussian optimization refines the scene layout within 200 iterations while preserving the generated object geometry. Experiments on Replica and ScanNet++ demonstrate consistent improvements at both instance and scene levels, with our method reducing scene-level CD by 43.8% on Replica and 30.9% on ScanNet++, while cutting flow-model sampling FLOPs and wall-time latency by nearly 20% under the same multi-view setting. Code will be released at https://github.com/xibi777/Scene-SAM3D.

Figures

Figures reproduced from arXiv: 2607.16805 by the authors.

Figure 1
Figure 1. Our preliminary study on extending SAM3D with multi-view fusion. Left: stochastic conditioning is efficient but unstable, whereas MultiDiffusion improves geometry at a higher cost. Right: different views produce different predictions of rotation, scale, and translation. stochastic conditioning [18]. However, even multiple observations may not fully cover an object, leaving missing geometry or inducing hallucinated s… view at source ↗
Figure 2
Figure 2. Overview of Scene-SAM3D view selection and multi-view fusion. Given calibrated multi￾view observations, Scene-SAM3D first selects a compact anchor-helper view set using world-space occupancy coverage. Based on the selected views, its performs step-efficient conflict-aware multiview fusion: the anchor view provides the main generation trajectory, while helper views are early-stopped and fused to inject complementary … view at source ↗
Figure 3
Figure 3. Rigid-object Gaussian layout refinement. Scene-SAM3D refines the scene by optimizing object-level similarity transforms in a foreground-only Gaussian representation. - Case B Anchor-unobserved region: For u ∈ Ω unobs a , the anchor no longer has privileged evidence. We instead let all selected views vote for the dominant direction: Iunobs(u) = {i ∈ Q | sgn(Zi(u)) = s(u)} , s(u) = sgn X i∈Q sgn(Zi(u))! , (10) and ave… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Qualitative instance-level results on the Replica and ScanNet++ datasets. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Qualitative scene-level results on the Replica datasets. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Ablation study on the multiview fusion steps. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Selected views visualization on Replica. [PITH_FULL_IMAGE:figures/full_fig_p016_7.png]
Figure 8
Figure 8. Figure 8: Qualitative visualization on Replica. In the error maps, warmer colors indicate larger [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]
Figure 9
Figure 9. Figure 9: Qualitative visualization on ScanNet++. In the error maps, warmer colors indicate larger [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

59 extracted references · 8 linked inside Pith

  1. [1]

    Jones, Rajinder Sodhi, Michael Murdock, Ravish Mehra, Hrvoje Benko, Andrew D

    Brett R. Jones, Rajinder Sodhi, Michael Murdock, Ravish Mehra, Hrvoje Benko, Andrew D. Wilson, Eyal Ofek, Blair MacIntyre, Nikunj Raghuvanshi, and Lior Shapira. Roomalive: Magical experiences enabled by scalable, adaptive projector-camera units. InThe 27th Annual ACM Symposium on User Interface Software and Technology (UIST), pages 637–644. ACM, 2014

    2014

  2. [2]

    Loop, Qin Cai, Philip A

    Sergio Orts-Escolano, Christoph Rhemann, Sean Ryan Fanello, Wayne Chang, Adarsh Kowdle, Yury Degtyarev, David Kim, Philip Davidson, Sameh Khamis, Mingsong Dou, Vladimir Tankovich, Charles T. Loop, Qin Cai, Philip A. Chou, Sarah Mennicken, Julien P. C. Valentin, Vivek Pradeep, Shenlong Wang, Sing Bing Kang, Pushmeet Kohli, Yuliya Lutchyn, Cem Keskin, and S...

    2016

  3. [3]

    Newcombe, Shahram Izadi, Otmar Hilliges, David Molyneaux, David Kim, An- drew J

    Richard A. Newcombe, Shahram Izadi, Otmar Hilliges, David Molyneaux, David Kim, An- drew J. Davison, Pushmeet Kohli, Jamie Shotton, Steve Hodges, and Andrew Fitzgibbon. Kinectfusion: Real-time dense surface mapping and tracking. In10th IEEE International Symposium on Mixed and Augmented Reality (ISMAR), pages 127–136. IEEE, 2011

    2011

  4. [4]

    Oasis: Procedurally generated social virtual spaces from 3d scanned real spaces.IEEE Trans

    Misha Sra, Sergio Garrido-Jurado, and Pattie Maes. Oasis: Procedurally generated social virtual spaces from 3d scanned real spaces.IEEE Trans. Vis. Comput. Graph., 24(12):3174–3187, 2018

    2018

  5. [5]

    VR-NeRF: high- fidelity virtualized walkable spaces

    Linning Xu, Vasu Agrawal, William Laney, Tony Garcia, Aayush Bansal, Changil Kim, Samuel Rota Bulò, Lorenzo Porzi, Peter Kontschieder, Aljaž Boži ˇc, et al. VR-NeRF: high- fidelity virtualized walkable spaces. InSIGGRAPH Asia Conference Papers, pages 1–12, 2023

    2023

  6. [6]

    Litereality: Graphics-ready 3d scene reconstruction from RGB-D scans.Ad- vances in Neural Information Processing Systems, 38:162794–162827, 2026

    Zhening Huang, Xiaoyang Wu, Fangcheng Zhong, Hengshuang Zhao, Matthias Nießner, and Joan Lasenby. Litereality: Graphics-ready 3d scene reconstruction from RGB-D scans.Ad- vances in Neural Information Processing Systems, 38:162794–162827, 2026

    2026

  7. [7]

    Robogen: Towards unleashing infinite data for automated robot learning via generative simulation

    Yufei Wang, Zhou Xian, Feng Chen, Tsun-Hsuan Wang, Yian Wang, Katerina Fragkiadaki, Zackory Erickson, David Held, and Chuang Gan. Robogen: Towards unleashing infinite data for automated robot learning via generative simulation. InForty-first International Conference on Machine Learning (ICML), Proceedings of Machine Learning Research, pages 51936–51983. P...

    2024

  8. [8]

    Maniskill2: A unified benchmark for generalizable manipulation skills

    Jiayuan Gu, Fanbo Xiang, Xuanlin Li, Zhan Ling, Xiqiang Liu, Tongzhou Mu, Yihe Tang, Stone Tao, Xinyue Wei, Yunchao Yao, Xiaodi Yuan, Pengwei Xie, Zhiao Huang, Rui Chen, and Hao Su. Maniskill2: A unified benchmark for generalizable manipulation skills. InThe Eleventh International Conference on Learning Representations (ICLR). OpenReview.net, 2023

    2023

Show all 59 references
  1. [9]

    Chang, and Manolis Savva

    Mukul Khanna, Yongsen Mao, Hanxiao Jiang, Sanjay Haresh, Brennan Shacklett, Dhruv Batra, Alexander Clegg, Eric Undersander, Angel X. Chang, and Manolis Savva. Habitat synthetic scenes dataset (HSSD-200): An analysis of 3d scene scale and realism tradeoffs for objectgoal naviga...

    2024

  2. [10]

    Turner, Oleksandr Maksymets, Zsolt Kira, Mrinal Kalakrishnan, Jitendra Malik, Devendra Singh Chaplot, Unnat Jain, Dhruv Batra, Ak- shara Rai, and Roozbeh Mottaghi

    Xavier Puig, Eric Undersander, Andrew Szot, Mikael Dallaire Cote, Tsung-Yen Yang, Rus- lan Partsey, Ruta Desai, Alexander Clegg, Michal Hlavac, So Yeon Min, Vladimir V ondrus, Théophile Gervet, Vincent-Pierre Berges, John M. Turner, Oleksandr Maksymets, Zsolt Kira, Mrinal Kala...

    2024

  3. [11]

    SQA3D: situated question answering in 3d scenes

    Xiaojian Ma, Silong Yong, Zilong Zheng, Qing Li, Yitao Liang, Song-Chun Zhu, and Siyuan Huang. SQA3D: situated question answering in 3d scenes. InThe Eleventh International Conference on Learning Representations (ICLR). OpenReview.net, 2023. 10

    2023

  4. [12]

    Openeqa: Embodied question answering in the era of foundation models

    Arjun Majumdar, Anurag Ajay, Xiaohan Zhang, Pranav Putta, Sriram Yenamandra, Mikael Henaff, Sneha Silwal, Paul McVay, Oleksandr Maksymets, Sergio Arnaud, Karmesh Yadav, Qiyang Li, Ben Newman, Mohit Sharma, Vincent-Pierre Berges, Shiqi Zhang, Pulkit Agrawal, Yonatan Bisk, Dhruv...

    2024

  5. [13]

    Structured 3d latents for scalable and versatile 3d generation

    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. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 21469– 21480, 2025

    2025

  6. [14]

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

    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...

    2025 arXiv

  7. [15]

    TriposG: high-fidelity 3d shape synthesis using large-scale rectified flow models.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

    Yangguang Li, Zi-Xin Zou, Zexiang Liu, Dehu Wang, Yuan Liang, Zhipeng Yu, Xingchao Liu, Yuan-Chen Guo, Ding Liang, Wanli Ouyang, et al. TriposG: high-fidelity 3d shape synthesis using large-scale rectified flow models.IEEE Transactions on Pattern Analysis and Machine Intellige...

    2025

  8. [16]

    Xingyu Chen, Fu-Jen Chu, Pierre Gleize, Kevin J. Liang, Alexander Sax, Hao Tang, Weiyao Wang, Michelle Guo, Thibaut Hardin, Xiang Li, Aohan Lin, Jiawei Liu, Ziqi Ma, Anushka Sagar, Bowen Song, Xiaodong Wang, Jianing Yang, Bowen Zhang, Piotr Dollár, Georgia Gkioxari, Matt Feisz...

    2025 arXiv

  9. [17]

    Multidiffusion: Fusing diffusion paths for controlled image generation

    Omer Bar-Tal, Lior Yariv, Yaron Lipman, and Tali Dekel. Multidiffusion: Fusing diffusion paths for controlled image generation. In Andreas Krause, Emma Brunskill, Kyunghyun Cho, Barbara Engelhardt, Sivan Sabato, and Jonathan Scarlett, editors,International Conference on Machin...

    2023

  10. [18]

    Reconviagen: Towards accurate multi-view 3d object reconstruction via generation.CoRR, abs/2510.23306, 2025

    Jiahao Chang, Chongjie Ye, Yushuang Wu, Yuantao Chen, Yidan Zhang, Zhongjin Luo, Chenghong Li, Yihao Zhi, and Xiaoguang Han. Reconviagen: Towards accurate multi-view 3d object reconstruction via generation.CoRR, abs/2510.23306, 2025

    2025

  11. [19]

    Besl and Neil D

    Paul J. Besl and Neil D. McKay. A method for registration of 3-d shapes.IEEE Trans. Pattern Anal. Mach. Intell., 14(2):239–256, 1992

    1992

  12. [20]

    Julian Straub, Thomas Whelan, Lingni Ma, Yufan Chen, Erik Wijmans, Simon Green, Jakob J. Engel, Raul Mur-Artal, Carl Yuheng Ren, Shobhit Verma, Anton Clarkson, Mingfei Yan, Brian Budge, Yajie Yan, Xiaqing Pan, June Yon, Yuyang Zou, Kimberly Leon, Nigel Carter, Jesus Briales, T...

    1906 arXiv

  13. [21]

    Scannet++: A high-fidelity dataset of 3d indoor scenes

    Chandan Yeshwanth, Yueh-Cheng Liu, Matthias Nießner, and Angela Dai. Scannet++: A high-fidelity dataset of 3d indoor scenes. InIEEE/CVF International Conference on Computer Vision (ICCV), pages 12–22. IEEE, 2023. 11

    2023

  14. [22]

    Amodal3r: Amodal 3d reconstruction from occluded 2d images

    Tianhao Wu, Chuanxia Zheng, Frank Guan, Andrea Vedaldi, and Tat-Jen Cham. Amodal3r: Amodal 3d reconstruction from occluded 2d images. InProceedings of the IEEE/CVF Interna- tional Conference on Computer Vision (ICCV), pages 9181–9193, 2025

    2025

  15. [23]

    In- stantmesh: Efficient 3d mesh generation from a single image with sparse-view large reconstruc- tion models.CoRR, abs/2404.07191, 2024

    Jiale Xu, Weihao Cheng, Yiming Gao, Xintao Wang, Shenghua Gao, and Ying Shan. In- stantmesh: Efficient 3d mesh generation from a single image with sparse-view large reconstruc- tion models.CoRR, abs/2404.07191, 2024

    2024 arXiv

  16. [24]

    Pixel2mesh: Generating 3d mesh models from single RGB images

    Nanyang Wang, Yinda Zhang, Zhuwen Li, Yanwei Fu, Wei Liu, and Yu-Gang Jiang. Pixel2mesh: Generating 3d mesh models from single RGB images. InComputer Vision - ECCV 2018, pages 55–71. Springer, 2018

    2018

  17. [25]

    Soft rasterizer: Differentiable rendering for unsupervised single-view mesh reconstruction.CoRR, abs/1901.05567, 2019

    Shichen Liu, Weikai Chen, Tianye Li, and Hao Li. Soft rasterizer: Differentiable rendering for unsupervised single-view mesh reconstruction.CoRR, abs/1901.05567, 2019

    1901 arXiv

  18. [26]

    Neural 3d mesh renderer

    Hiroharu Kato, Yoshitaka Ushiku, and Tatsuya Harada. Neural 3d mesh renderer. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 3907–3916, 2018

    2018

  19. [27]

    DISN: deep implicit surface network for high-quality single-view 3d reconstruction

    Qiangeng Xu, Weiyue Wang, Duygu Ceylan, Radomír Mech, and Ulrich Neumann. DISN: deep implicit surface network for high-quality single-view 3d reconstruction. InAdvances in Neural Information Processing Systems (NeurIPS), pages 490–500, 2019

    2019

  20. [28]

    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. InThe Twelfth International Conference on Learning Representations (ICLR). OpenReview.net, 2024

    2024

  21. [29]

    Direct3d: Scalable image-to-3d generation via 3d latent diffusion transformer

    Shuang Wu, Youtian Lin, Yifei Zeng, Feihu Zhang, Jingxi Xu, Philip Torr, Xun Cao, and Yao Yao. Direct3d: Scalable image-to-3d generation via 3d latent diffusion transformer. InAdvances in Neural Information Processing Systems (NeurIPS), 2024

    2024

  22. [30]

    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 Advances in Neural Information Processing Systems (NeurIPS), 2023

    2023

  23. [31]

    CRM: single image to 3d textured mesh with convolutional reconstruction model

    Zhengyi Wang, Yikai Wang, Yifei Chen, Chendong Xiang, Shuo Chen, Dajiang Yu, Chongxuan Li, Hang Su, and Jun Zhu. CRM: single image to 3d textured mesh with convolutional reconstruction model. InComputer Vision - ECCV 2024, pages 57–74. Springer, 2024

    2024

  24. [32]

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

    Xiaoxiao Long, Yuan-Chen Guo, Cheng Lin, Yuan Liu, Zhiyang Dou, Lingjie Liu, Yuexin Ma, Song-Hai Zhang, Marc Habermann, Christian Theobalt, and Wenping Wang. Wonder3d: Single image to 3d using cross-domain diffusion. InIEEE/CVF Conference on Computer Vision and Pattern Recogni...

    2024

  25. [33]

    Unique3d: High-quality and efficient 3d mesh generation from a single image

    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. InAdvances in Neural Information Processing Systems (NeurIPS), 2024

    2024

  26. [34]

    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, Eli VanderBilt, Aniruddha Kembhavi, Carl V ondrick, Georgia Gkioxari, Kiana Ehsani, Ludwig Schmidt, and Ali Farhadi. Objave...

    2023

  27. [35]

    Yago Vicente, Thomas Dideriksen, Himanshu Arora, Matthieu Guillaumin, and Jitendra Malik

    Jasmine Collins, Shubham Goel, Kenan Deng, Achleshwar Luthra, Leon Xu, Erhan Gundogdu, Xi Zhang, Tomas F. Yago Vicente, Thomas Dideriksen, Himanshu Arora, Matthieu Guillaumin, and Jitendra Malik. ABO: dataset and benchmarks for real-world 3d object understanding. In IEEE/CVF C...

    2022

  28. [36]

    MIDI: multi-instance diffusion for single image to 3d scene generation

    Zehuan Huang, Yuan-Chen Guo, Xingqiao An, Yunhan Yang, Yangguang Li, Zi-Xin Zou, Ding Liang, Xihui Liu, Yan-Pei Cao, and Lu Sheng. MIDI: multi-instance diffusion for single image to 3d scene generation. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), p...

    2025

  29. [37]

    Scenegen: Single-image 3d scene generation in one feedforward pass.CoRR, abs/2508.15769, 2025

    Yanxu Meng, Haoning Wu, Ya Zhang, and Weidi Xie. Scenegen: Single-image 3d scene generation in one feedforward pass.CoRR, abs/2508.15769, 2025

    2025

  30. [38]

    Scenemaker: Open-set 3d scene generation with decoupled de-occlusion and pose estimation model.CoRR, abs/2512.10957, 2025

    Yukai Shi, Weiyu Li, Zihao Wang, Hongyang Li, Xingyu Chen, Ping Tan, and Lei Zhang. Scenemaker: Open-set 3d scene generation with decoupled de-occlusion and pose estimation model.CoRR, abs/2512.10957, 2025

    2025

  31. [39]

    I-scene: 3d instance models are implicit generalizable spatial learners.CoRR, abs/2512.13683, 2025

    Lu Ling, Yunhao Ge, Yichen Sheng, and Aniket Bera. I-scene: 3d instance models are implicit generalizable spatial learners.CoRR, abs/2512.13683, 2025

    2025

  32. [40]

    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. InThe Twelfth International Conference on Learning Representations (ICLR). OpenReview.net, 2024

    2024

  33. [41]

    SV3D: novel multi-view synthesis and 3d generation from a single image using latent video diffusion

    Vikram V oleti, Chun-Han Yao, Mark Boss, Adam Letts, David Pankratz, Dmitry Tochilkin, Christian Laforte, Robin Rombach, and Varun Jampani. SV3D: novel multi-view synthesis and 3d generation from a single image using latent video diffusion. InComputer Vision - ECCV 2024, pages...

    2024

  34. [42]

    Ouroboros3d: Image- to-3d generation via 3d-aware recursive diffusion

    Hao Wen, Zehuan Huang, Yaohui Wang, Xinyuan Chen, and Lu Sheng. Ouroboros3d: Image- to-3d generation via 3d-aware recursive diffusion. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 21631–21641. Computer Vision Foundation / IEEE, 2025

    2025

  35. [43]

    Unirecgen: Unifying multi-view 3d reconstruction and generation.arXiv preprint arXiv:2604.01479, 2026

    Zhisheng Huang, Jiahao Chen, Cheng Lin, Chenyu Hu, Hanzhuo Huang, Zhengming Yu, Mengfei Li, Yuheng Liu, Zekai Gu, Zibo Zhao, et al. Unirecgen: Unifying multi-view 3d reconstruction and generation.arXiv preprint arXiv:2604.01479, 2026

    2026 arXiv

  36. [44]

    Frost, Samir Aroudj, Armen Avetisyan, Henry Howard-Jenkins, Daniel DeTone, Pierre Moulon, Qirui Wu, Zhengqin Li, Julian Straub, Richard A

    Yawar Siddiqui, Duncan P. Frost, Samir Aroudj, Armen Avetisyan, Henry Howard-Jenkins, Daniel DeTone, Pierre Moulon, Qirui Wu, Zhengqin Li, Julian Straub, Richard A. Newcombe, and Jakob J. Engel. Shaper: Robust conditional 3d shape generation from casual captures. CoRR, abs/260...

    2026

  37. [45]

    VGGT: visual geometry grounded transformer

    Jianyuan Wang, Minghao Chen, Nikita Karaev, Andrea Vedaldi, Christian Rupprecht, and David Novotný. VGGT: visual geometry grounded transformer. InIEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2025, Nashville, TN, USA, June 11-15, 2025, pages 5294–5306. C...

    2025

  38. [46]

    MV- SAM3D: adaptive multi-view fusion for layout-aware 3d generation.CoRR, abs/2603.11633, 2026

    Baicheng Li, Dong Wu, Jun Li, Shunkai Zhou, Zecui Zeng, Lusong Li, and Hongbin Zha. MV- SAM3D: adaptive multi-view fusion for layout-aware 3d generation.CoRR, abs/2603.11633, 2026

    2026 arXiv

  39. [47]

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

    Ruoshi Liu, Rundi Wu, Basile Van Hoorick, Pavel Tokmakov, Sergey Zakharov, and Carl V on- drick. Zero-1-to-3: Zero-shot one image to 3d object. InIEEE/CVF International Conference on Computer Vision (ICCV), pages 9264–9275. IEEE, 2023

    2023

  40. [48]

    Flex3d: Feed-forward 3d generation with flexible reconstruction model and input view curation

    Junlin Han, Jianyuan Wang, Andrea Vedaldi, Philip Torr, and Filippos Kokkinos. Flex3d: Feed-forward 3d generation with flexible reconstruction model and input view curation. In Forty-second International Conference on Machine Learning (ICML), Proceedings of Machine Learning Re...

    2025

  41. [49]

    A solution to the next best view problem for automated surface acquisition.IEEE Trans

    Richard Pito. A solution to the next best view problem for automated surface acquisition.IEEE Trans. Pattern Anal. Mach. Intell., 21(10):1016–1030, 1999

    1999

  42. [50]

    Gammell, and Paul Newman

    Rowan Border, Jonathan D. Gammell, and Paul Newman. Surface edge explorer (see): Planning next best views directly from 3d observations. In2018 IEEE International Conference on Robotics and Automation (ICRA), pages 1–8. IEEE, 2018

    2018

  43. [51]

    Rowan Border and Jonathan D. Gammell. The surface edge explorer (see): A measurement- direct approach to next best view planning.Int. J. Robotics Res., 43(10):1506–1532, 2024

    2024

  44. [52]

    Fisherrf: Active view selection and mapping with radiance fields using fisher information

    Wen Jiang, Boshu Lei, and Kostas Daniilidis. Fisherrf: Active view selection and mapping with radiance fields using fisher information. InComputer Vision - ECCV 2024, pages 422–440. Springer, 2024. 13

    2024

  45. [53]

    Pop-gs: Next best view in 3d- gaussian splatting with p-optimality

    Joey Wilson, Marcelino Almeida, Sachit Mahajan, Martin Labrie, Maani Ghaffari, Omid Ghasemalizadeh, Min Sun, Cheng-Hao Kuo, and Arnab Sen. Pop-gs: Next best view in 3d- gaussian splatting with p-optimality. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR...

    2025

  46. [54]

    Next-best view policy for 3d reconstruction

    Daryl Peralta, Joel Casimiro, Aldrin Michael Nilles, Justine Aletta Aguilar, Rowel Atienza, and Rhandley Cajote. Next-best view policy for 3d reconstruction. InComputer Vision - ECCV 2020 Workshops, pages 558–573. Springer, 2020

    2020

  47. [55]

    Gennbv: Generalizable next-best-view policy for active 3d reconstruction

    Xiao Chen, Quanyi Li, Tai Wang, Tianfan Xue, and Jiangmiao Pang. Gennbv: Generalizable next-best-view policy for active 3d reconstruction. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 16436–16445. IEEE, 2024

    2024

  48. [56]

    VIN-NBV: a view introspection network for next-best-view selection for resource-efficient 3d reconstruction.CoRR, abs/2505.06219, 2025

    Noah Frahm, Dongxu Zhao, Andrea Dunn Beltran, Ron Alterovitz, Jan-Michael Frahm, Junier Oliva, and Roni Sengupta. VIN-NBV: a view introspection network for next-best-view selection for resource-efficient 3d reconstruction.CoRR, abs/2505.06219, 2025

    2025 arXiv

  49. [57]

    Decompositional neural scene reconstruction with generative diffusion prior

    Junfeng Ni, Yu Liu, Ruijie Lu, Zirui Zhou, Song-Chun Zhu, Yixin Chen, and Siyuan Huang. Decompositional neural scene reconstruction with generative diffusion prior. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 6022–6033. Computer Vision Founda...

    2025

  50. [58]

    Xin Kong, Shikun Liu, Marwan Taher, and Andrew J. Davison. vmap: Vectorised object mapping for neural field SLAM. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 952–961. IEEE, 2023

    2023

  51. [59]

    k-means++: the advantages of careful seeding

    David Arthur and Sergei Vassilvitskii. k-means++: the advantages of careful seeding. In Nikhil Bansal, Kirk Pruhs, and Clifford Stein, editors,Proceedings of the Eighteenth Annual ACM-SIAM Symposium on Discrete Algorithms, SODA 2007, New Orleans, Louisiana, USA, January 7-9, 2...

    2007

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.