pith. sign in

arxiv: 2606.19019 · v1 · pith:LFNUTX6Mnew · submitted 2026-06-17 · 💻 cs.CV

FlowObject: Flow Steering for Bridging Generative Priors and Reconstruction Fidelity

Yuchen Rao , Xuqian Ren , Yinyu Nie , Sayan Deb Sarkar , Biao Zhang , Vincent Lepetit , Friedrich Fraundorfer This is my paper

Pith reviewed 2026-06-26 21:40 UTC · model grok-4.3

classification 💻 cs.CV
keywords flow-matchingsparse-view 3D reconstructiongenerative priors3D Gaussian SplattingODE guidanceinverse problemview-dependent fidelity
0
0 comments X

The pith

Dual-space guidance steers flow-matching ODEs to complete 3D geometry from sparse views while matching observations exactly.

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

Sparse-view 3D reconstruction faces a tradeoff where generative models produce complete but biased or inconsistent assets and optimization methods like 3D Gaussian Splatting deliver high fidelity only on visible surfaces. FlowObject reframes the task as a training-free inverse problem that applies dual-space guidance to the ODE trajectory of a pre-trained flow-matching model, allowing learned priors to fill unseen regions without overriding input images. A subsequent 3DGS refinement stage converts the steered output into photorealistic renderings. Benchmarks on synthetic and real datasets with severe occlusions show the approach improves both geometric completeness and view-dependent fidelity over pure generative and pure optimization baselines.

Core claim

Reformulating sparse-view 3D reconstruction as a training-free guided inverse problem on the flow-matching ODE, with dual-space guidance enforcing consistency to observations while permitting priors to complete unseen parts, followed by 3DGS refinement, yields reconstructions that are simultaneously more complete and more faithful to input views than current state-of-the-art methods.

What carries the argument

Dual-space guidance strategy applied to the ODE trajectory of a flow-matching model, which steers generation toward both the generative prior and the observed images.

If this is right

  • State-of-the-art generative models and optimization frameworks each fail to achieve both geometric completeness and observational consistency under heavy occlusions.
  • Bridging generative flow models with a 3DGS refinement stage reduces synthetic bias while preserving high visual fidelity.
  • The training-free nature allows the method to leverage existing pre-trained flow-matching models without additional fine-tuning.
  • The resulting assets exhibit improved performance on both synthetic benchmarks and real-world captures with partial observations.

Where Pith is reading between the lines

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

  • The same steering approach may generalize to other inverse problems where a generative prior must be balanced against direct measurements.
  • Flow-matching models appear more amenable to ODE-level guidance than diffusion models for this class of reconstruction tasks.
  • Extending the method to time-varying scenes or larger environments would test whether the guidance remains stable at scale.

Load-bearing premise

The dual-space guidance on the flow-matching ODE trajectory reliably completes unseen regions through learned priors while strictly enforcing consistency with observations without introducing new inconsistencies.

What would settle it

A test case in which the completed geometry, even after 3DGS refinement, produces renderings that visibly contradict any input view or shows no measurable gain in occluded-region completeness over baselines.

Figures

Figures reproduced from arXiv: 2606.19019 by Biao Zhang, Friedrich Fraundorfer, Sayan Deb Sarkar, Vincent Lepetit, Xuqian Ren, Yinyu Nie, Yuchen Rao.

Figure 1
Figure 1. Figure 1: Overview of FlowObject. FlowObject provides a complete and high-fidelity 3D reconstruction of objects from few casual RGB-D images. Existing optimization-based 3D reconstruction methods fail to recover the unobserved geometry; generation-based methods produce complete object representations but in general a geometry that differs from the real object. FlowObject brings the best of the two worlds by combinin… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the Guided Voxel Generation pipeline. Partial depth inputs are fused into a voxel occupancy grid V, denoted as the partial observation, from which the guiding latent zv is extracted. During the structure flow process, our framework employs a combination of latent (Limplv ) and observation (Lexplv ) constraints to iteratively refine the latent state z, ultimately reconstructing a complete and hi… view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the Guided 3DGS Generation pipeline. Our Gaussian synthesis pipeline consolidates sparse-view RGB images {Ii} into a photometric-derived guiding la￾tent zs , built on the pre-generated voxel grid. By integrating latent (Limpls ) and photometric (Lexpls ) constraints during the appearance flow process, the framework iteratively optimizes the latent state z to reconstruct a complete, view-consist… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative Comparison of Object Reconstruction on the 3D-FRONT. Our method leverages generative priors to recover complete structures under heavy occlusion. In contrast, 3DGS optimization fails to reconstruct severely occluded furniture parts, while diffusion-based methods struggle to preserve structural integrity and topological consistency. ShapeR Evaluation Dataset [40]. This benchmark presents a chall… view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative Comparison of Object Reconstruction on Real-World Datasets. The first three samples (top row) originate from the ScanNet++ dataset [49], while the re￾maining three (bottom row) are drawn from ShapeR Evaluation Dataset [40]. Our method maintains global structural and topological consistency, whereas other baselines exhibit sig￾nificant geometric incompleteness or perceptual artifacts on real-wor… view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative Rendering Comparison on Challenging Real-World Samples. The first two samples (top row) originate from the ScanNet++ dataset [49], while the remaining two (bottom row) are drawn from ShapeR Evaluation Dataset [40]. Our method achieves superior appearance reconstruction with high-fidelity texture consistency. In contrast, base￾lines exhibit significant observational gaps and produce disjointed t… view at source ↗
read the original abstract

Recovering complete 3D representations of objects from few casual image captures remains a significant challenge. Recent 3D generative models, particularly those based on Flow-Matching (FM), can synthesize high-quality textured assets; however, they often suffer from ''synthetic bias'' where learned priors override observational evidence, alongside a lack of alignment with the observed instance. Conversely, optimization-based methods like 3D Gaussian Splatting (3DGS) provide high fidelity on visible surfaces but fail to reason about unobserved geometry. In this paper, we present FlowObject, a framework that reformulates sparse-view 3D reconstruction as a training-free, guided inverse problem. Our approach applies a dual-space guidance strategy to steer the Ordinary Differential Equation (ODE) trajectory of a flow-matching model, enabling the completion of unseen regions through learned generative priors while enforcing strict consistency with real-world observations. By integrating a 3DGS refinement stage, FlowObject further bridges the gap between ''synthetic-looking'' generative outputs and photorealistic reconstructions. Comprehensive benchmarks on synthetic and real-world datasets demonstrate that current state-of-the-art methods often struggle to achieve geometric completeness and observational consistency simultaneously, especially under severe occlusions. In contrast, our method significantly outperforms state-of-the-art generative models and optimization-based frameworks in both geometric completeness and view-dependent appearance fidelity.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript presents FlowObject, a training-free framework that reformulates sparse-view 3D reconstruction as a guided inverse problem. It applies a dual-space guidance strategy to steer the ODE trajectory of a flow-matching model, completing unseen regions via generative priors while enforcing strict consistency with observations, followed by a 3DGS refinement stage. The central claim is that this approach significantly outperforms state-of-the-art generative models and optimization-based methods in geometric completeness and view-dependent appearance fidelity on synthetic and real-world datasets.

Significance. If the dual-space guidance can be shown to achieve strict observational consistency without introducing new inconsistencies, the work would meaningfully advance the integration of generative priors with high-fidelity reconstruction for object capture under occlusion.

major comments (2)
  1. [Abstract] Abstract: the claim that the method 'significantly outperforms state-of-the-art generative models and optimization-based frameworks' is unsupported by any quantitative numbers, error bars, tables, or benchmark details, leaving the central empirical claim unverifiable from the provided text.
  2. [Abstract] Abstract: no equation, derivation, or proof sketch is supplied for the dual-space guidance terms on the flow-matching ODE, so it is impossible to verify whether the combined guidance produces a trajectory whose endpoint projects exactly onto the observed views (the minimal condition for the 'strict consistency' claim under severe occlusion).
minor comments (1)
  1. [Abstract] Abstract: inconsistent quotation style ('synthetic bias' vs. ''synthetic-looking'') and minor typographic issues with emphasis.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and indicate planned revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that the method 'significantly outperforms state-of-the-art generative models and optimization-based frameworks' is unsupported by any quantitative numbers, error bars, tables, or benchmark details, leaving the central empirical claim unverifiable from the provided text.

    Authors: We agree that the abstract, being a concise summary, does not include specific numerical results. The full manuscript provides detailed quantitative benchmarks in Section 4, including tables with PSNR, SSIM, LPIPS, and geometric completeness metrics with error bars across synthetic and real-world datasets. To address this, we will revise the abstract to incorporate key performance highlights supporting the outperformance claim. revision: yes

  2. Referee: [Abstract] Abstract: no equation, derivation, or proof sketch is supplied for the dual-space guidance terms on the flow-matching ODE, so it is impossible to verify whether the combined guidance produces a trajectory whose endpoint projects exactly onto the observed views (the minimal condition for the 'strict consistency' claim under severe occlusion).

    Authors: The equations and derivation of the dual-space guidance terms on the flow-matching ODE are provided in Section 3.2 of the full manuscript, with additional analysis in the supplementary material. The guidance combines prior and observation terms to steer the trajectory toward consistency. We will revise the abstract to reference this formulation. We note that the guidance enforces observational consistency at each step but does not include a formal proof of exact endpoint projection under all occlusion conditions; empirical validation supports the strict consistency claim. revision: partial

Circularity Check

0 steps flagged

No derivation chain or equations presented; method described at conceptual level only

full rationale

The abstract and provided text contain no equations, parameter fittings, uniqueness theorems, or derivation steps. The approach is characterized as a training-free dual-space guidance on an FM ODE trajectory plus a 3DGS refinement stage, with performance claims resting on empirical benchmarks rather than any mathematical reduction. No self-definitional mappings, fitted inputs renamed as predictions, or load-bearing self-citations appear. The derivation chain is absent, so no circularity can be exhibited; the description is self-contained as an inference-time procedure.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review prevents enumeration of specific free parameters or axioms; the central claim implicitly rests on unstated assumptions about the steerability of flow-matching ODEs and the compatibility of generative priors with observed data.

pith-pipeline@v0.9.1-grok · 5790 in / 1177 out tokens · 14419 ms · 2026-06-26T21:40:37.002833+00:00 · methodology

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

55 extracted references · 15 canonical work pages · 7 internal anchors

  1. [1]

    Scan2cad: Learning cad model alignment in rgb-d scans

    Armen Avetisyan, Manuel Dahnert, Angela Dai, Manolis Savva, Angel X Chang, and Matthias Nießner. Scan2cad: Learning cad model alignment in rgb-d scans. InCVPR, pages 2614–2623, 2019

    2019

  2. [2]

    Universal guidance for diffusion models

    Arpit Bansal, Hong-Min Chu, Avi Schwarzschild, Soumyadip Sengupta, Micah Goldblum, Jonas Geiping, and Tom Goldstein. Universal guidance for diffusion models. InCVPR, pages 843–852, 2023

    2023

  3. [3]

    Reconviagen: Towards accurate multi-view 3d object reconstruction via generation.arXiv preprint arXiv: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.arXiv preprint arXiv:2510.23306, 2025

    work page arXiv 2025

  4. [4]

    PGSR: Planar-Based Gaussian Splatting for Efficient and High-Fidelity Surface Reconstruction .IEEE TVCG, 31(09), 2025

    Danpeng Chen, Hai Li, Weicai Ye, Yifan Wang, Weijian Xie, Shangjin Zhai, Nan Wang, Haomin Liu, Hujun Bao, and Guofeng Zhang. PGSR: Planar-Based Gaussian Splatting for Efficient and High-Fidelity Surface Reconstruction .IEEE TVCG, 31(09), 2025

    2025

  5. [5]

    Text2tex: Text-driven texture synthesis via diffusion models

    Dave Zhenyu Chen, Yawar Siddiqui, Hsin-Ying Lee, Sergey Tulyakov, and Matthias Nießner. Text2tex: Text-driven texture synthesis via diffusion models. InICCV, pages 18558–18568, 2023. 16RAO ET AL.: FLOWOBJECT

    2023

  6. [6]

    VCR- GauS: View Consistent Depth-Normal Regularizer for Gaussian Surface Reconstruction

    Hanlin Chen, Fangyin Wei, Chen Li, Tianxin Huang, Yunsong Wang, and Gim Hee Lee. VCR- GauS: View Consistent Depth-Normal Regularizer for Gaussian Surface Reconstruction. In NeurIPS, 2024

    2024

  7. [7]

    SAM 3D: 3Dfy Anything in Images

    Xingyu Chen, Fu-Jen Chu, Pierre Gleize, Kevin J Liang, Alexander Sax, Hao Tang, Weiyao Wang, Michelle Guo, Thibaut Hardin, Xiang Li, et al. Sam 3d: 3dfy anything in images.arXiv preprint arXiv:2511.16624, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  8. [8]

    Mesh2nerf: Direct mesh supervision for neural radiance field representation and generation

    Yujin Chen, Yinyu Nie, Benjamin Ummenhofer, Reiner Birkl, Michael Paulitsch, Matthias Müller, and Matthias Nießner. Mesh2nerf: Direct mesh supervision for neural radiance field representation and generation. InECCV, pages 173–191. Springer, 2024

    2024

  9. [9]

    Diffusion-sdf: Conditional generative modeling of signed distance functions

    Gene Chou, Yuval Bahat, and Felix Heide. Diffusion-sdf: Conditional generative modeling of signed distance functions. InICCV, pages 2262–2272, 2023

    2023

  10. [10]

    Diffusion Posterior Sampling for General Noisy Inverse Problems

    Hyungjin Chung, Jeongsol Kim, Michael T Mccann, Marc L Klasky, and Jong Chul Ye. Diffusion posterior sampling for general noisy inverse problems.arXiv preprint arXiv:2209.14687, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  11. [11]

    Depth-regularized optimization for 3d gaussian splatting in few-shot images

    Jaeyoung Chung, Jeongtaek Oh, and Kyoung Mu Lee. Depth-regularized optimization for 3d gaussian splatting in few-shot images. InCVPR, pages 811–820, 2024

    2024

  12. [12]

    Shape completion using 3d-encoder- predictor cnns and shape synthesis

    Angela Dai, Charles Ruizhongtai Qi, and Matthias Nießner. Shape completion using 3d-encoder- predictor cnns and shape synthesis. InCVPR, pages 5868–5877, 2017

    2017

  13. [13]

    High- quality Surface Reconstruction using Gaussian Surfels

    Pinxuan Dai, Jiamin Xu, Wenxiang Xie, Xinguo Liu, Huamin Wang, and Weiwei Xu. High- quality Surface Reconstruction using Gaussian Surfels. InSIGGRAPH Asia, 2024

    2024

  14. [14]

    3d-front: 3d furnished rooms with layouts and semantics

    Huan Fu, Bowen Cai, Lin Gao, Ling-Xiao Zhang, Jiaming Wang, Cao Li, Qixun Zeng, Chengyue Sun, Rongfei Jia, Binqiang Zhao, et al. 3d-front: 3d furnished rooms with layouts and semantics. InICCV, pages 10933–10942, 2021

    2021

  15. [15]

    SuGaR: Surface-Aligned Gaussian Splatting for Efficient 3D Mesh Reconstruction and High-Quality Mesh Rendering

    Antoine Guédon and Vincent Lepetit. SuGaR: Surface-Aligned Gaussian Splatting for Efficient 3D Mesh Reconstruction and High-Quality Mesh Rendering. InCVPR, 2024

    2024

  16. [16]

    MILo: Mesh-In-the-Loop Gaussian Splatting for Detailed and Efficient Surface Reconstruction.ACM TOG, 44(6), 2025

    Antoine Guédon, Diego Gomez, Nissim Maruani, Bingchen Gong, George Drettakis, and Maks Ovsjanikov. MILo: Mesh-In-the-Loop Gaussian Splatting for Detailed and Efficient Surface Reconstruction.ACM TOG, 44(6), 2025

    2025

  17. [17]

    Roca: Robust cad model retrieval and alignment from a single image

    Can Gümeli, Angela Dai, and Matthias Nießner. Roca: Robust cad model retrieval and alignment from a single image. InCVPR, pages 4022–4031, 2022

    2022

  18. [18]

    Denoising diffusion probabilistic models.NeurIPS, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models.NeurIPS, 33:6840–6851, 2020

    2020

  19. [19]

    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

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  20. [20]

    G3splat: Geometrically consistent generalizable gaussian splatting.arXiv preprint arXiv:2512.17547, 2025

    Mehdi Hosseinzadeh, Shin-Fang Chng, Yi Xu, Simon Lucey, Ian Reid, and Ravi Garg. G3splat: Geometrically consistent generalizable gaussian splatting.arXiv preprint arXiv:2512.17547, 2025

    work page arXiv 2025

  21. [21]

    2D Gaussian Splat- ting for Geometrically Accurate Radiance Fields

    Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao. 2D Gaussian Splat- ting for Geometrically Accurate Radiance Fields. InSIGGRAPH, 2024. RAO ET AL.: FLOWOBJECT17

    2024

  22. [22]

    Cupid: Gen- erative 3d reconstruction via joint object and pose modeling.arXiv preprint arXiv:2510.20776, 2025

    Binbin Huang, Haobin Duan, Yiqun Zhao, Zibo Zhao, Yi Ma, and Shenghua Gao. Cupid: Gen- erative 3d reconstruction via joint object and pose modeling.arXiv preprint arXiv:2510.20776, 2025

    work page arXiv 2025

  23. [23]

    3d gaussian splatting for real-time radiance field rendering.ACM Trans

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, George Drettakis, et al. 3d gaussian splatting for real-time radiance field rendering.ACM Trans. Graph., 42(4):139–1, 2023

    2023

  24. [24]

    Hunyuan3D 2.5: Towards High-Fidelity 3D Assets Generation with Ultimate Details

    Zeqiang Lai, Yunfei Zhao, Haolin Liu, Zibo Zhao, Qingxiang Lin, Huiwen Shi, Xianghui Yang, Mingxin Yang, Shuhui Yang, Yifei Feng, et al. Hunyuan3d 2.5: Towards high-fidelity 3d assets generation with ultimate details.arXiv preprint arXiv:2506.16504, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  25. [25]

    Magic3d: High-resolution text-to-3d con- tent creation

    Chen-Hsuan Lin, Jun Gao, Luming Tang, Towaki Takikawa, Xiaohui Zeng, Xun Huang, Karsten Kreis, Sanja Fidler, Ming-Yu Liu, and Tsung-Yi Lin. Magic3d: High-resolution text-to-3d con- tent creation. InCVPR, pages 300–309, 2023

    2023

  26. [26]

    Depth Anything 3: Recovering the Visual Space from Any Views

    Haotong Lin, Sili Chen, Junhao Liew, Donny Y Chen, Zhenyu Li, Guang Shi, Jiashi Feng, and Bingyi Kang. Depth anything 3: Recovering the visual space from any views.arXiv preprint arXiv:2511.10647, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  27. [27]

    Prompting depth anything for 4k resolution accurate metric depth estimation

    Haotong Lin, Sida Peng, Jingxiao Chen, Songyou Peng, Jiaming Sun, Minghuan Liu, Hujun Bao, Jiashi Feng, Xiaowei Zhou, and Bingyi Kang. Prompting depth anything for 4k resolution accurate metric depth estimation. InCVPR, pages 17070–17080, 2025

    2025

  28. [28]

    Flow Matching for Generative Modeling

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow match- ing for generative modeling.arXiv preprint arXiv:2210.02747, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  29. [29]

    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. InICCV, pages 9298–9309, 2023

    2023

  30. [30]

    Orientation matters: Making 3d generative models orientation-aligned.arXiv preprint arXiv:2506.08640, 2025

    Yichong Lu, Yuzhuo Tian, Zijin Jiang, Yikun Zhao, Yuanbo Yang, Hao Ouyang, Haoji Hu, Huimin Yu, Yujun Shen, and Yiyi Liao. Orientation matters: Making 3d generative models orientation-aligned.arXiv preprint arXiv:2506.08640, 2025

    work page arXiv 2025

  31. [31]

    Text2mesh: Text- driven neural stylization for meshes

    Oscar Michel, Roi Bar-On, Richard Liu, Sagie Benaim, and Rana Hanocka. Text2mesh: Text- driven neural stylization for meshes. InCVPR, pages 13492–13502, 2022

    2022

  32. [32]

    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. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 6022–6033, 2025

    2025

  33. [33]

    Learning 3d scene priors with 2d supervision

    Yinyu Nie, Angela Dai, Xiaoguang Han, and Matthias Nießner. Learning 3d scene priors with 2d supervision. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 792–802, 2023

    2023

  34. [34]

    SOF: Sorted Opacity Fields for Fast Unbounded Surface Reconstruction

    Lukas Radl, Felix Windisch, Thomas Deixelberger, Jozef Hladky, Michael Steiner, Dieter Schmalstieg, and Markus Steinberger. SOF: Sorted Opacity Fields for Fast Unbounded Surface Reconstruction. InSIGGRAPH Asia, 2025

    2025

  35. [35]

    Patchcomplete: Learning multi-resolution patch priors for 3d shape completion on unseen categories.NeurIPS, 35:34436–34450, 2022

    Yuchen Rao, Yinyu Nie, and Angela Dai. Patchcomplete: Learning multi-resolution patch priors for 3d shape completion on unseen categories.NeurIPS, 35:34436–34450, 2022

    2022

  36. [36]

    Lever- aging automatic cad annotations for supervised learning in 3d scene understanding.arXiv preprint arXiv:2504.13580, 2025

    Yuchen Rao, Stefan Ainetter, Sinisa Stekovic, Vincent Lepetit, and Friedrich Fraundorfer. Lever- aging automatic cad annotations for supervised learning in 3d scene understanding.arXiv preprint arXiv:2504.13580, 2025. 18RAO ET AL.: FLOWOBJECT

    work page arXiv 2025

  37. [37]

    Sam 2: Segment any- thing in images and videos

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman Rädle, Chloe Rolland, Laura Gustafson, et al. Sam 2: Segment any- thing in images and videos. InInternational Conference on Learning Representations, volume 2025, pages 28085–28128, 2025

    2025

  38. [38]

    Ags-mesh: Adaptive gaussian splatting and meshing with geometric priors for indoor room reconstruction using smartphones

    Xuqian Ren, Matias Turkulainen, Jiepeng Wang, Otto Seiskari, Iaroslav Melekhov, Juho Kannala, and Esa Rahtu. Ags-mesh: Adaptive gaussian splatting and meshing with geometric priors for indoor room reconstruction using smartphones. In2025 International Conference on 3D Vision (3DV), pages 1080–1090. IEEE, 2025

    2025

  39. [39]

    GuideFlow3D: Optimization-Guided Rectified Flow For Appearance Transfer

    Sayan Deb Sarkar, Sinisa Stekovic, Vincent Lepetit, and Iro Armeni. GuideFlow3D: Optimization-Guided Rectified Flow For Appearance Transfer. InNeurIPS, 2025

    2025

  40. [40]

    Shaper: Robust conditional 3d shape generation from casual captures,

    Yawar Siddiqui, Duncan Frost, Samir Aroudj, Armen Avetisyan, Henry Howard-Jenkins, Daniel DeTone, Pierre Moulon, Qirui Wu, Zhengqin Li, Julian Straub, et al. Shaper: Robust conditional 3d shape generation from casual captures.arXiv preprint arXiv:2601.11514, 2026

    work page arXiv 2026

  41. [41]

    Dn-splatter: Depth and normal priors for gaussian splatting and meshing

    Matias Turkulainen, Xuqian Ren, Iaroslav Melekhov, Otto Seiskari, Esa Rahtu, and Juho Kan- nala. Dn-splatter: Depth and normal priors for gaussian splatting and meshing. In2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 2421–2431. IEEE, 2025

    2025

  42. [42]

    Unilat3d: Geometry-appearance unified latents for single-stage 3d generation.arXiv preprint arXiv:2509.25079, 2025

    Guanjun Wu, Jiemin Fang, Chen Yang, Sikuang Li, Taoran Yi, Jia Lu, Zanwei Zhou, Jiazhong Cen, Lingxi Xie, Xiaopeng Zhang, et al. Unilat3d: Geometry-appearance unified latents for single-stage 3d generation.arXiv preprint arXiv:2509.25079, 2025

    work page arXiv 2025

  43. [43]

    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. InICCV, pages 9181–9193, 2025

    2025

  44. [44]

    Holoscene: Simulation-ready interactive 3d worlds from a single video.Advances in Neural Information Processing Systems, 38:32501–32524, 2026

    Hongchi Xia, Chih-Hao Lin, Hao-Yu Hsu, Quentin Leboutet, Katelyn Gao, Michael Paulitsch, Benjamin Ummenhofer, and Shenlong Wang. Holoscene: Simulation-ready interactive 3d worlds from a single video.Advances in Neural Information Processing Systems, 38:32501–32524, 2026

    2026

  45. [45]

    Native and Compact Structured Latents for 3D Generation

    Jianfeng Xiang, Xiaoxue Chen, Sicheng Xu, Ruicheng Wang, Zelong Lv, Yu Deng, Hongyuan Zhu, Yue Dong, Hao Zhao, Nicholas Jing Yuan, et al. Native and compact structured latents for 3d generation.arXiv preprint arXiv:2512.14692, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  46. [46]

    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. In CVPR, pages 21469–21480, 2025

    2025

  47. [47]

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

    Yinghao Xu, Zifan Shi, Wang Yifan, Hansheng Chen, Ceyuan Yang, Sida Peng, Yujun Shen, and Gordon Wetzstein. Grm: Large gaussian reconstruction model for efficient 3d reconstruction and generation. InECCV, pages 1–20. Springer, 2024

    2024

  48. [48]

    Instascene: Towards complete 3d instance decomposition and reconstruction from cluttered scenes

    Zesong Yang, Bangbang Yang, Wenqi Dong, Chenxuan Cao, Liyuan Cui, Yuewen Ma, Zhaopeng Cui, and Hujun Bao. Instascene: Towards complete 3d instance decomposition and reconstruction from cluttered scenes. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 7771–7781, 2025

    2025

  49. [49]

    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. InICCV, pages 12–22, 2023

    2023

  50. [50]

    Gaussian Opacity Fields: Efficient Adaptive Surface Reconstruction in Unbounded Scenes.ACM TOG, 43(6), 2024

    Zehao Yu, Torsten Sattler, and Andreas Geiger. Gaussian Opacity Fields: Efficient Adaptive Surface Reconstruction in Unbounded Scenes.ACM TOG, 43(6), 2024. RAO ET AL.: FLOWOBJECT19

    2024

  51. [51]

    Clay: A controllable large-scale generative model for creating high- quality 3d assets.ACM Transactions on Graphics (TOG), 43(4):1–20, 2024

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

    2024

  52. [52]

    Flow priors for linear inverse problems via iterative corrupted trajectory matching

    Yasi Zhang, Peiyu Yu, Yaxuan Zhu, Yingshan Chang, Feng Gao, Ying Nian Wu, and Oscar Leong. Flow priors for linear inverse problems via iterative corrupted trajectory matching. NeurIPS, 37:57389–57417, 2024

    2024

  53. [53]

    A water-tight and fully manifold 3D mesh of a [category] in [room type], no redundant parts, no missing parts, no holes, complete surface

    Zheng Zhou, Yu-Jie Xiong, Jia-Chen Zhang, Chun-Ming Xia, Xihe Qiu, and Hongjian Zhan. Gradient-direction-aware density control for 3d gaussian splatting.arXiv preprint arXiv:2508.09239, 2025. RAO ET AL.: FLOWOBJECT1 FlowObject: Flow Steering for Bridging Generative Priors and Reconstruction Fidelity Supplementary Material Abstract We provide more experime...

    work page arXiv 2025

  54. [54]

    Compared to GS-based and purely generative baselines, our method reconstructs structurally sound geometry in heavily occluded regions while maintaining con- sistent metric scale

    Ours GT Figure S2:Qualitative Object Reconstruction Results.The first three samples (top row) originate from the 3D-FRONT dataset [14], the subsequent three (middle row) are drawn from ScanNet++ [49], and the remaining three (bottom row) are drawn from ShapeR Eval- uation Dataset [40]. Compared to GS-based and purely generative baselines, our method recon...

  55. [55]

    Our method achieves superior appearance reconstruction with high-fidelity texture consistency across multi-view renders

    Ours GT Figure S3:Qualitative Rendering Comparison.The first three samples (top row) origi- nate from the 3D-FRONT dataset [14], the subsequent three (middle row) are drawn from ScanNet++ [49], and the remaining three (bottom row) are drawn from ShapeR Evaluation Dataset [40]. Our method achieves superior appearance reconstruction with high-fidelity textu...