pith. machine review for the scientific record. sign in

arxiv: 2604.11808 · v2 · submitted 2026-04-13 · 💻 cs.CV

Recognition: 2 theorem links

· Lean Theorem

Pair2Scene: Learning Local Object Relations for Procedural Scene Generation

Xingjian Ran , Shujie Zhang , Weipeng Zhong , Li Luo , Bo Dai
Authors on Pith no claims yet

Pith reviewed 2026-05-12 01:51 UTC · model grok-4.3

classification 💻 cs.CV
keywords 3D scene generationprocedural modelingobject relation learningsupport and functional relationshierarchical generationindoor environmentscollision-aware sampling3D-Pairs dataset
0
0 comments X

The pith

Object placements in 3D indoor scenes can be modeled through learned local support and functional relations to generate complex environments beyond the training distribution.

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

The paper shows that high-fidelity 3D scene generation does not require full global scene distributions or large language models for spatial reasoning. Instead, it builds scenes from local inter-object rules that capture how one object supports or relates functionally to another. A network learns to predict position distributions for dependent objects given an anchor object's position and shape. These local predictions are then applied recursively inside a scene hierarchy, with collision-aware rejection sampling to produce overall coherent layouts. A reader would care because this sidesteps data scarcity by reusing pair-wise patterns across many configurations and produces physically stable, semantically sensible results even in dense setups not seen during training.

Core claim

Pair2Scene trains a network on a curated 3D-Pairs dataset to capture two main types of local relations: support relations that follow physical hierarchies and functional relations that reflect semantic links. At inference time the framework starts with root objects and recursively places dependent objects according to the learned position distributions, using collision-aware rejection sampling inside the hierarchy to resolve local decisions into globally plausible scenes that maintain physical stability and semantic consistency while exceeding the complexity of the original training data.

What carries the argument

A neural network that estimates the spatial position distribution of a dependent object conditioned only on the position and geometry of one or more anchor objects, representing support and functional relations.

If this is right

  • Scenes can be built hierarchically starting from a small set of seed objects and scaling to arbitrary density without retraining.
  • Physical plausibility is enforced by support relations and explicit collision rejection at each placement step.
  • Semantic coherence is achieved through functional relations without needing global context or external reasoning modules.
  • The same trained model generalizes to scene layouts whose overall statistics differ from the training distribution.
  • Generation remains procedural and controllable by varying the hierarchy depth and rejection thresholds.

Where Pith is reading between the lines

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

  • The same local-pair approach could be tested on 2D floor-plan generation or robotic workspace arrangement where full-scene examples are also scarce.
  • Because recursion depth controls complexity, the method might allow users to generate scenes at multiple scales from the same trained weights.
  • Combining the learned local rules with high-level semantic guidance from other models could produce task-specific scenes while retaining the spatial precision shown here.

Load-bearing premise

Local dependencies between objects are sufficient to produce coherent global scene layouts when applied recursively within hierarchies together with collision-aware rejection sampling.

What would settle it

Generate scenes with object counts and densities well above those in the training set and measure rates of physical instability (objects intersecting or falling) or semantic implausibility (objects placed in functionally invalid locations); high failure rates would falsify the claim that local rules suffice.

Figures

Figures reproduced from arXiv: 2604.11808 by Bo Dai, Li Luo, Shujie Zhang, Weipeng Zhong, Xingjian Ran.

Figure 1
Figure 1. Figure 1: Scene Generation Results. (Top) When trained solely on curated 3D-Front data (avg. 4.07 objects for bedroom), learning-based methods remain confined to its limited distribution. In contrast, Pair2Scene beyond dataset distribution by leveraging local object relations to procedurally generate more complex scenes. (Bottom) When trained on full proposed 3D-Pairs dataset, Pair2Scene maintains superior spatial l… view at source ↗
Figure 2
Figure 2. Figure 2: Architecture of Pair2Scene Model. The pipeline processes the dependent asset Adep and anchor objects (support Osup and functional Ofnc) to predict the spatial distribution of the dependent object. Point cloud encoders extract latent geometric features z geo, while the bounding box encoder transforms spatial configurations into embeddings e bbox. These are processed through cascaded Transformer-like blocks … view at source ↗
Figure 3
Figure 3. Figure 3: Scene Assembly Pipeline. The framework parses text or room type into support and functional Trees, serialized as an ordered relation sequence. The Pair2Scene model then predicts spatial distributions based on these relations. Finally, rejection sampling resolves global collisions to produce the final 3D scene. predicted MoL distribution. Specifically, we minimize the Negative Log-Likelihood (NLL) loss func… view at source ↗
Figure 4
Figure 4. Figure 4: Data Curation Pipeline. Our data curation workflow involves Physical Validation to filter unstable objects, Support Extraction to identify support dependencies, and LLM-Driven Functional Distillation to extract functional relations. This process results in the 3D-Pair Dataset, which captures fine-grained object interactions. prevent inter-object collisions, we define the target distribu￾tion pglobal(x) as … view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative comparison under the 3D-Front only setting. Infinigen-Indoors Holodeck LayoutVLM FactoredScenes Ours view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative comparison under the multi-source setting view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative ablation study on the multi-source setting. where objects are logically grounded. This is evidenced by our significantly higher scores in Scene Complexity and Se￾mantic Alignment, where human evaluators find our layouts more functionally dense and consistent with commonsense. Overall, our framework excels in balancing intricate detail with overall scene coherence, achieving the highest Contex￾t… view at source ↗
Figure 8
Figure 8. Figure 8: More visualization results on new scene types view at source ↗
Figure 9
Figure 9. Figure 9: Screenshot of the user study interface view at source ↗
Figure 10
Figure 10. Figure 10: Distribution of dependents for floor as support anchor. 16 view at source ↗
Figure 11
Figure 11. Figure 11: Distribution of dependents for cabinet as support anchor view at source ↗
Figure 12
Figure 12. Figure 12: Distribution of dependents for floor as support anchor and bed as functional anchor view at source ↗
Figure 13
Figure 13. Figure 13: Distribution of dependents for floor as support anchor and cabinet as functional anchor. 17 view at source ↗
Figure 14
Figure 14. Figure 14: Distribution of dependents for floor as support anchor and table as functional anchor view at source ↗
Figure 15
Figure 15. Figure 15: Distribution of dependents for table as support anchor view at source ↗
Figure 16
Figure 16. Figure 16: Distribution of dependents for table as support anchor and book as functional anchor. 18 view at source ↗
Figure 17
Figure 17. Figure 17: Distribution of dependents for table as support anchor and bowl as functional anchor view at source ↗
Figure 18
Figure 18. Figure 18: Distribution of dependents for table as support anchor and computer as functional anchor. 19 view at source ↗
read the original abstract

Generating high-fidelity 3D indoor scenes remains a significant challenge due to data scarcity and the complexity of modeling intricate spatial relations. Current methods often struggle to scale beyond training distribution to dense scenes or rely on LLMs/VLMs that lack the ability for precise spatial reasoning. Building on top of the observation that object placement relies mainly on local dependencies instead of information-redundant global distributions, in this paper, we propose Pair2Scene, a novel procedural generation framework that integrates learned local rules with scene hierarchies and physics-based algorithms. These rules mainly capture two types of inter-object relations, namely support relations that follow physical hierarchies, and functional relations that reflect semantic links. We model these rules through a network, which estimates spatial position distributions of dependent objects conditioned on position and geometry of the anchor ones. Accordingly, we curate a dataset 3D-Pairs from existing scene data to train the model. During inference, our framework can generate scenes by recursively applying our model within a hierarchical structure, leveraging collision-aware rejection sampling to align local rules into coherent global layouts. Extensive experiments demonstrate that our framework outperforms existing methods in generating complex environments that go beyond training data while maintaining physical and semantic plausibility.

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

3 major / 2 minor

Summary. The paper proposes Pair2Scene, a procedural 3D indoor scene generation framework that learns local support and functional relations between object pairs from a curated 3D-Pairs dataset. A neural network models conditional spatial position distributions for dependent objects given anchor geometry and position. Scenes are generated recursively by applying the model in an (unspecified) hierarchical structure combined with collision-aware rejection sampling. The central claim is that this local-rule approach outperforms prior methods on complex, out-of-distribution scenes while preserving physical and semantic plausibility.

Significance. If the empirical claims hold, the work would be significant for procedural scene synthesis: it offers a data-efficient alternative to global distribution models or LLM-based methods by showing that curated local pair relations can compose into plausible layouts beyond the training distribution. The integration of learned distributions with physics-based rejection sampling is a concrete strength that could influence future hybrid procedural pipelines.

major comments (3)
  1. [Abstract and Experiments] Abstract and Experiments section: the claim that the framework 'outperforms existing methods' and produces 'plausible' results in complex OOD scenes is asserted without any reported quantitative metrics, baselines, ablation studies, or error analysis. This absence directly undermines verification of the generalization and plausibility claims that constitute the paper's main contribution.
  2. [Method and Inference] Method and Inference sections: the recursive application of the learned pair model 'within a hierarchical structure' plus collision-aware rejection sampling is presented as sufficient to produce globally coherent layouts, yet no analysis, failure-case enumeration, or ablation is given for cases with overlapping relations or dense packing. This step is load-bearing for the central claim that local dependencies alone suffice for OOD scenes.
  3. [§3] §3 (dataset curation): the 3D-Pairs dataset is described as capturing 'mainly' local dependencies, but no quantitative comparison is supplied against global scene statistics or alternative conditioning schemes to justify that local pair modeling is informationally adequate for the claimed generalization regime.
minor comments (2)
  1. [Method] Notation for the conditional distribution p(·|anchor) and the network architecture should be formalized with explicit equations rather than prose descriptions.
  2. [Experiments] Figure captions and experimental protocol details (scene sizes, object counts, rejection-sampling parameters) are insufficiently specified for reproducibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the thoughtful and constructive review. We agree that the empirical claims require stronger quantitative support and that additional analysis of the inference process and dataset rationale would strengthen the manuscript. We outline revisions below to address each major comment.

read point-by-point responses

  1. Referee: [Abstract and Experiments] Abstract and Experiments section: the claim that the framework 'outperforms existing methods' and produces 'plausible' results in complex OOD scenes is asserted without any reported quantitative metrics, baselines, ablation studies, or error analysis. This absence directly undermines verification of the generalization and plausibility claims that constitute the paper's main contribution.

    Authors: We acknowledge that the current Experiments section relies primarily on qualitative visual comparisons. To strengthen the claims, we will add quantitative evaluations including baseline comparisons using metrics such as object placement accuracy and scene coherence scores, along with ablation studies on the pair model and collision sampling. These will be reported with error analysis in the revised Experiments section. revision: yes

  2. Referee: [Method and Inference] Method and Inference sections: the recursive application of the learned pair model 'within a hierarchical structure' plus collision-aware rejection sampling is presented as sufficient to produce globally coherent layouts, yet no analysis, failure-case enumeration, or ablation is given for cases with overlapping relations or dense packing. This step is load-bearing for the central claim that local dependencies alone suffice for OOD scenes.

    Authors: We agree that further validation of the recursive inference is warranted. In the revision we will add a new subsection with failure-case enumeration for overlapping relations and dense packing, plus an ablation isolating the contribution of collision-aware sampling. This will demonstrate how local rules compose into coherent global layouts. revision: yes

  3. Referee: [§3] §3 (dataset curation): the 3D-Pairs dataset is described as capturing 'mainly' local dependencies, but no quantitative comparison is supplied against global scene statistics or alternative conditioning schemes to justify that local pair modeling is informationally adequate for the claimed generalization regime.

    Authors: To justify the local modeling choice, we will augment Section 3 with quantitative comparisons of local versus global relation statistics extracted from the source scene datasets, as well as a brief comparison of model performance under local versus global conditioning. This will support the claim that local pair relations are informationally adequate. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on independent training and procedural application

full rationale

The paper trains a network on a curated 3D-Pairs dataset extracted from existing scene data to learn local support and functional relations, then applies the model recursively within hierarchies plus collision-aware rejection sampling. No equations, predictions, or claims in the abstract or described framework reduce outputs to fitted inputs by construction, nor do any load-bearing steps rely on self-citations, imported uniqueness theorems, or smuggled ansatzes. The central premise (local dependencies suffice for global coherence) is presented as an observation to be validated empirically rather than a definitional equivalence. This is a standard non-circular setup for learned procedural generation.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on training a neural network whose parameters are fitted to the curated 3D-Pairs dataset and on the domain assumption that local relations suffice for global coherence. No new physical entities are postulated.

free parameters (1)
  • neural network parameters
    Weights of the model that estimates spatial position distributions conditioned on anchor object geometry and position; fitted during training on 3D-Pairs.
axioms (1)
  • domain assumption Local object relations dominate scene structure over global distributions
    Explicitly stated as the foundational observation for building the Pair2Scene framework.

pith-pipeline@v0.9.0 · 5517 in / 1375 out tokens · 71795 ms · 2026-05-12T01:51:31.346140+00:00 · methodology

Review history (3 revisions) →

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

36 extracted references · 36 canonical work pages · 2 internal anchors

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

    work page

  2. [2]

    Kenny Jones, Qiuhong Anna Wei, Kailiang Fu, and Daniel Ritchie

    Aguina-Kang, R., Gumin, M., Han, D. H., Morris, S., Yoo, S. J., Ganeshan, A., Jones, R. K., Wei, Q. A., Fu, K., and Ritchie, D. Open-universe indoor scene generation using llm program synthesis and uncurated object databases. arXiv preprint arXiv:2403.09675, 2024

    work page arXiv 2024

  3. [3]

    J., Arbel, M., and Gretton, A

    Bi \'n kowski, M., Sutherland, D. J., Arbel, M., and Gretton, A. Demystifying mmd gans. In ICLR, 2018

    work page 2018

  4. [4]

    I-design: Personalized llm interior designer

    C elen, A., Han, G., Schindler, K., Van Gool, L., Armeni, I., Obukhov, A., and Wang, X. I-design: Personalized llm interior designer. arXiv preprint arXiv:2404.02838, 2024

    work page arXiv 2024

  5. [5]

    Global-local tree search in vlms for 3d indoor scene generation

    Deng, W., Qi, M., and Ma, H. Global-local tree search in vlms for 3d indoor scene generation. In CVPR, 2025

    work page 2025

  6. [6]

    E., and Wang, W

    Feng, W., Zhu, W., Fu, T.-j., Jampani, V., Akula, A., He, X., Basu, S., Wang, X. E., and Wang, W. Y. Layoutgpt: Compositional visual planning and generation with large language models. In NeurIPS, 2023

    work page 2023

  7. [7]

    Casagpt: cuboid arrangement and scene assembly for interior design

    Feng, W., Zhou, H., Liao, J., Cheng, L., and Zhou, W. Casagpt: cuboid arrangement and scene assembly for interior design. In CVPR, 2025

    work page 2025

  8. [8]

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

    Fu, H., Cai, B., Gao, L., Zhang, L.-X., Wang, J., Li, C., Zeng, Q., Sun, C., Jia, R., Zhao, B., et al. 3d-front: 3d furnished rooms with layouts and semantics. In ICCV, 2021

    work page 2021

  9. [9]

    Anyhome: Open-vocabulary generation of structured and textured 3d homes

    Fu, R., Wen, Z., Liu, Z., and Sridhar, S. Anyhome: Open-vocabulary generation of structured and textured 3d homes. In ECCV, 2024

    work page 2024

  10. [10]

    Gemini 3 flash: frontier intelligence built for speed, 2025

    Google . Gemini 3 flash: frontier intelligence built for speed, 2025. URL https://blog.google/products/gemini/gemini-3-flash/

    work page 2025

  11. [11]

    Mesatask: Towards task-driven tabletop scene generation via 3d spatial reasoning

    Hao, J., Liang, N., Luo, Z., Xu, X., Zhong, W., Yi, R., Jin, Y., Lyu, Z., Zheng, F., Ma, L., et al. Mesatask: Towards task-driven tabletop scene generation via 3d spatial reasoning. arXiv preprint arXiv:2509.22281, 2025

    work page arXiv 2025

  12. [12]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium

    Heusel, M., Ramsauer, H., Unterthiner, T., Nessler, B., and Hochreiter, S. Gans trained by a two time-scale update rule converge to a local nash equilibrium. In NeurIPS, 2017

    work page 2017

  13. [13]

    Hsu, J., Jin, E., Wu, J., and Mitra, N. J. From programs to poses: Factored real-world scene generation via learned program libraries. arXiv preprint arXiv:2510.10292, 2025

    work page arXiv 2025

  14. [14]

    arXiv preprint arXiv:2506.16504 , year=

    Lai, Z., Zhao, Y., Liu, H., Zhao, Z., Lin, Q., Shi, H., Yang, X., Yang, M., Yang, S., Feng, Y., et al. Hunyuan3d 2.5: Towards high-fidelity 3d assets generation with ultimate details. arXiv preprint arXiv:2506.16504, 2025

    work page arXiv 2025

  15. [15]

    Proc-GS : Procedural building generation for city assembly with 3d gaussians

    Li, Y., Ran, X., Xu, L., Lu, T., Yu, M., Wang, Z., Xiangli, Y., Lin, D., and Dai, B. Proc-GS : Procedural building generation for city assembly with 3d gaussians. In CVPR, 2025

    work page 2025

  16. [16]

    and Mu, Y

    Lin, C. and Mu, Y. Instructscene: Instruction-driven 3d indoor scene synthesis with semantic graph prior. In ICLR, 2024

    work page 2024

  17. [17]

    E., Liu, W., Tian, Y., and Yuan, L

    Pang, Y., Wang, W., Tay, F. E., Liu, W., Tian, Y., and Yuan, L. Masked autoencoders for point cloud self-supervised learning. In ECCV, 2022

    work page 2022

  18. [18]

    Atiss: Autoregressive transformers for indoor scene synthesis

    Paschalidou, D., Kar, A., Shugrina, M., Kreis, K., Geiger, A., and Fidler, S. Atiss: Autoregressive transformers for indoor scene synthesis. In NeurIPS, 2021

    work page 2021

  19. [19]

    Pun, H. I. D., Tam, H. I. I., Wang, A. T., Huo, X., Chang, A. X., and Savva, M. Hsm: Hierarchical scene motifs for multi-scale indoor scene generation. arXiv preprint arXiv:2503.16848, 2025

    work page arXiv 2025

  20. [20]

    Infinigen indoors: Photorealistic indoor scenes using procedural generation

    Raistrick, A., Mei, L., Kayan, K., Yan, D., Zuo, Y., Han, B., Wen, H., Parakh, M., Alexandropoulos, S., Lipson, L., et al. Infinigen indoors: Photorealistic indoor scenes using procedural generation. In CVPR, 2024

    work page 2024

  21. [21]

    Direct numerical layout generation for 3d indoor scene synthesis via spatial reasoning

    Ran, X., Li, Y., Xu, L., Yu, M., and Dai, B. Direct numerical layout generation for 3d indoor scene synthesis via spatial reasoning. arXiv preprint arXiv:2506.05341, 2025

    work page arXiv 2025

  22. [22]

    Salimans, T., Karpathy, A., Chen, X., and Kingma, D. P. Pixelcnn++: Improving the pixelcnn with discretized logistic mixture likelihood and other modifications. In ICLR, 2017

    work page 2017

  23. [23]

    Layoutvlm: Differentiable optimization of 3d layout via vision-language models

    Sun, F.-Y., Liu, W., Gu, S., Lim, D., Bhat, G., Tombari, F., Li, M., Haber, N., and Wu, J. Layoutvlm: Differentiable optimization of 3d layout via vision-language models. In CVPR, 2025 a

    work page 2025

  24. [24]

    Hierarchically-structured open-vocabulary indoor scene synthesis with pre-trained large language model

    Sun, W., Li, X., Li, M., Xu, K., Meng, X., and Meng, L. Hierarchically-structured open-vocabulary indoor scene synthesis with pre-trained large language model. In AAAI, 2025 b

    work page 2025

  25. [25]

    Diffuscene: Denoising diffusion models for generative indoor scene synthesis

    Tang, J., Nie, Y., Markhasin, L., Dai, A., Thies, J., and Nie ner, M. Diffuscene: Denoising diffusion models for generative indoor scene synthesis. In CVPR, 2024

    work page 2024

  26. [26]

    Wavenet: A generative model for raw audio

    van den Oord, A., Dieleman, S., Zen, H., Simonyan, K., Vinyals, O., Graves, A., Kalchbrenner, N., Senior, A., and Kavukcuoglu, K. Wavenet: A generative model for raw audio. In Proc. SSW, 2016

    work page 2016

  27. [27]

    N., Kaiser, ., and Polosukhin, I

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, ., and Polosukhin, I. Attention is all you need. In NeurIPS, 2017

    work page 2017

  28. [28]

    arXiv preprint arXiv:2412.01506 (2024) 4

    Xiang, J., Lv, Z., Xu, S., Deng, Y., Wang, R., Zhang, B., Chen, D., Tong, X., and Yang, J. Structured 3d latents for scalable and versatile 3d generation. arXiv preprint arXiv:2412.01506, 2024

    work page arXiv 2024

  29. [29]

    J., Sanchez, V., and Zheng, F

    Yang, Y., Lu, J., Zhao, Z., Luo, Z., Yu, J. J., Sanchez, V., and Zheng, F. Llplace: The 3d indoor scene layout generation and editing via large language model. arXiv preprint arXiv:2406.03866, 2024 a

    work page arXiv 2024

  30. [30]

    Holodeck: Language guided generation of 3d embodied ai environments

    Yang, Y., Sun, F.-Y., Weihs, L., VanderBilt, E., Herrasti, A., Han, W., Wu, J., Haber, N., Krishna, R., Liu, L., et al. Holodeck: Language guided generation of 3d embodied ai environments. In CVPR, 2024 b

    work page 2024

  31. [31]

    P., Chen, D

    Zhai, G., \"O rnek, E. P., Chen, D. Z., Liao, R., Di, Y., Navab, N., Tombari, F., and Busam, B. Echoscene: Indoor scene generation via information echo over scene graph diffusion. In ECCV, 2024

    work page 2024

  32. [32]

    Clay: A controllable large-scale generative model for creating high-quality 3d assets

    Zhang, L., Wang, Z., Zhang, Q., Qiu, Q., Pang, A., Jiang, H., Yang, W., Xu, L., and Yu, J. Clay: A controllable large-scale generative model for creating high-quality 3d assets. In SIGGRAPH, 2024

    work page 2024

  33. [33]

    Hunyuan3D 2.0: Scaling Diffusion Models for High Resolution Textured 3D Assets Generation

    Zhao, Z., Lai, Z., Lin, Q., Zhao, Y., Liu, H., Yang, S., Feng, Y., Yang, M., Zhang, S., Yang, X., et al. Hunyuan3d 2.0: Scaling diffusion models for high resolution textured 3d assets generation. arXiv preprint arXiv:2501.12202, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  34. [34]

    InternScenes: A Large-scale Simulatable Indoor Scene Dataset with Realistic Layouts

    Zhong, W., Cao, P., Jin, Y., Luo, L., Cai, W., Lin, J., Wang, H., Lyu, Z., Wang, T., Dai, B., et al. Internscenes: A large-scale simulatable indoor scene dataset with realistic layouts. arXiv preprint arXiv:2509.10813, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  35. [35]

    Gala3d: Towards text-to-3d complex scene generation via layout-guided generative gaussian splatting

    Zhou, X., Ran, X., Xiong, Y., He, J., Lin, Z., Wang, Y., Sun, D., and Yang, M.-H. Gala3d: Towards text-to-3d complex scene generation via layout-guided generative gaussian splatting. In ICML, 2024

    work page 2024

  36. [36]

    On the continuity of rotation representations in neural networks

    Zhou, Y., Barnes, C., Lu, J., Yang, J., and Li, H. On the continuity of rotation representations in neural networks. In CVPR, 2019

    work page 2019