Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

A unified 3D generation pipeline shows reconstruction fidelity, not the generative model, sets the quality ceiling, with compression errors alone accounting for 39.3% of generation error.

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

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

SDF grids reconstruct best, Dual Octrees score best on automatic generation metrics, but users prefer SDF output, and reconstruction plus compression errors make up a large share of generation error.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection Useful open-source 3D representation benchmark, but the headline error-decomposition numbers (12.9%/39.3%) don't survive scrutiny against the paper's own tables. the 4 major comments →

arxiv 2509.02474 v1 pith:77HDYR5D submitted 2025-09-02 cs.GR cs.CVcs.LG

Unifi3D: A Study on 3D Representations for Generation and Reconstruction in a Common Framework

Nina Wiedemann , Sainan Liu , Quentin Leboutet , Katelyn Gao , Benjamin Ummenhofer , Michael Paulitsch , Kai Yuan This is my paper
classification cs.GR cs.CVcs.LG
keywords 3D representationgenerationreconstructionlatent diffusionbenchmarksigned distance functionoctreeShapeNet
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

This paper builds one interchangeable pipeline for 3D generation—mesh to representation to latent to diffusion to mesh—so that six competing representations (voxel grids, SDF grids, point-cloud/occupancy sets, dual octrees, triplanes, and NeRFs) can be compared without the usual confounding differences in preprocessing, training, and metrics. Running reconstruction and generation on the same objects, it finds that the representation's reconstruction fidelity, not the generative model, sets the ceiling: reconstruction and compression errors together account for roughly half of the generation error. The empirical ranking is that an SDF-grid autoencoder reconstructs best (F-score 88.4), while a dual-octree variational autoencoder with a U-Net diffusion model generates best on automatic metrics (1-NNA 0.824), though human evaluators prefer SDF-generated surfaces. The paper's practical message is that reconstruction and generation must be evaluated jointly, and it releases the modular codebase so other representations can be plugged in.

Core claim

Unifi3D claims that the performance of latent-diffusion 3D generation is substantially bounded by how well the chosen representation survives the round trip through an encoder and decoder. On a controlled ShapeNet training setup, the SDF grid with a plain autoencoder gives the best reconstruction (F-score 88.4), but the dual octree with a variational autoencoder and U-Net diffusion gives the best unconditional generation on Chamfer-based metrics (1-NNA 0.824 vs 0.860 for SDF). A second dataset (Objaverse) reproduces the main rankings. The decomposition of the generation MMD shows reconstruction errors make up 12.9% and compression errors 39.3% of the total, so the authors conclude that repor

What carries the argument

The central object is the standardized four-stage pipeline: mesh conversion into a representation, autoencoder compression into a latent, latent diffusion (DiT or U-Net), and mesh reconstruction—with every representation forced through the same data split, training budget, and metric set. The load-bearing comparison device is the error decomposition: the generation MMD is set against the per-sample Chamfer distance of mesh-to-mesh reconstruction and of encode-decode reconstruction, which lower-bounds what a generator can achieve. Around this, a flood-fill preprocessing step defines inside/outside without thickening thin structures, and a Bradley-Terry user study converts 575 pairwise prefere

Load-bearing premise

The comparison is fair only if the observed differences come from the representation itself; but the paper uses a different encoder and diffusion backbone for each representation, and the NeRF results are explicitly affected by changes made to fit the common framework.

What would settle it

Retrain all six representations in the same harness with one shared diffusion backbone (one DiT for everything) and one shared tokenization, then check whether SDF still reconstructs best and DualOctree still generates best, and whether compression still accounts for 39.3% of the MMD; if swapping the backbone reorders the winners, the representation ranking is confounded by implementation.

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

If this is right

  • The SDF-grid autoencoder is the safest default when reconstruction fidelity matters, e.g., for CAD-style or quality-critical applications.
  • The dual octree's top automatic generation score with a small latent marks it as a promising direction for memory-limited and detail-hungry generation, since its spatial adaptivity packs more information per latent element.
  • Because compression alone eats 39.3% of the generation MMD, improving autoencoders—not just diffusion backbones—is a direct lever on generation quality.
  • Evaluation of unconditional 3D generation needs at least 200 samples; smaller sets inflate spread and make literature numbers incomparable.
  • Preprocessing matters: the flood-fill conversion avoids the surface-thickening bias of standard manifoldization, which distorts meshes and adds representation-independent error.

Where Pith is reading between the lines

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

  • If reconstruction error is a hard ceiling on generation, then the field's emphasis on ever-larger diffusion models may be mis-aimed: equally large gains could come from decoders tailored to each representation, or from fine-tuned reconstruction objectives.
  • The rankings pair representations with specific generators (DiT for SDF, U-Net for octree); a natural testable extension is to decouple the two and report a matrix of representation × generator results, since the paper's own ablations show DiT vs U-Net flips the winner for some representations.
  • The OOD gap (NeRF and DualOctree drop hardest when trained on chairs and tested on airplanes) suggests compact latents and MLP decoders overfit category-specific geometry; increasing latent capacity or adding regularization should be testable against the paper's benchmark.
  • The framework quantifies dataset quality via round-trip conversion error—a reusable diagnostic beyond this paper's ranking, e.g., for deciding whether a new 3D dataset is clean enough for generative training.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes Unifi3D, a unified pipeline for benchmarking 3D representations (voxel grids, SDF grids, 3DShape2VecSet, Dual Octrees, triplanes, and NeRFs) across representation conversion, compression with autoencoders, latent diffusion generation, and mesh reconstruction. Reconstruction quality is measured with Chamfer distance, F-score, and normal consistency on ShapeNet and Objaverse; unconditional generation is measured with COV, MMD, and 1-NNA, complemented by a user study. The central claims are that SDF AE gives the best reconstruction (F-score 88.4), DualOctree VAE UNet gives the best generation on automatic metrics (1-NNA 0.824), and that reconstruction/compression errors account for 12.9% and 39.3%, respectively, of the generation MMD, motivating joint evaluation of reconstruction and generation.

Significance. The paper provides a useful open-source benchmark and a broad empirical comparison of six representations under a common pipeline. Strengths include the modular codebase, the novel DualOctree generative baseline, the Objaverse replication, the preprocessing study, and the analysis of metric stability with respect to sample size. If the central attribution claims are correct, the paper would be an important reference for practitioners choosing 3D representations. However, the load-bearing quantitative decomposition in Fig. 6 is not reproducible from the reported tables, and the fairness assumption that observed differences are inherent to the representation is weakened by representation-specific encoders and diffusion backbones. These issues need to be resolved before the main conclusions can be relied upon.

major comments (4)
  1. [§4.5, Fig. 6] The 39.3% compression share of MMD cannot be reproduced from Tables 2 and 3. Table 3 reports the average CD for DualOctree, SDF, Shape2VecSet, Triplane, and Voxel as 0.020, 0.012, 0.023, 0.028, and 0.016. If these are actual CD values, their mean (0.0198) divided by the mean Table 2 MMD (0.0344) is ~58%, not 39.3%. If the Table 3 header '(*1e-4)' means the values are scaled by 1e-4, the ratio is below 1%. The 39.3% figure matches a chair-only computation using Table 6 and Table 7, but §4.5 states 'averaging over all representations' and Fig. 6 does not indicate chair-only scope. The 12.9% reconstruction share is attributed to Fig. 7b, which plots F-score, not Chamfer distance, so it cannot be recomputed from the cited evidence. Please clarify the units, the exact per-sample computation, and the scope (all categories vs chair-only), and make the decomposition reproducible from the tables
  2. [§4.5, Fig. 6] The error decomposition double-counts the mesh→representation→mesh conversion error. The compression errors taken from Table 3 are full-pipeline errors (mesh→representation→latent→representation→mesh), which already include the conversion error used for the 'reconstruction' share. Presenting 12.9% and 39.3% as two stacked components of the MMD therefore conflates overlapping error sources. The caveat 'these errors are not simply additive' does not resolve this structural double counting; the paper should either decompose the stages sequentially (e.g., conversion error first, then incremental compression error) or present the percentages as shares of a non-additive upper bound with a clear statement of what is included in each term.
  3. [§2, §2.2, §4.2] The central attribution claim—that the standardized pipeline ensures 'any observed performance differences are inherent to the representations themselves'—is not supported because the encoder and diffusion architectures are not held constant. SDF and Shape2VecSet use DiT; DualOctree uses a two-layer 3D U-Net; Triplane uses a U-Net with 3D-aware convolutions; NeRF uses a modified Shap-E encoder. Section 4.2 explicitly says NeRF performs worst 'probably due to modifications in our implementation for the sake of comparability.' With different backbones per representation, the observed generation rankings (Table 2) and even reconstruction rankings could be driven by implementation choices rather than the representation. Either add controlled comparisons that vary only the representation, or explicitly reframe the conclusions as comparisons of the authors' best pipeline instantiation per repr
  4. [§4.1, Table 2] The generation metric table reports no confidence intervals or multiple-seed variance. The differences between adjacent methods (MMD 0.031 vs 0.032 vs 0.033; 1-NNA 0.824 vs 0.860) are small relative to the spread shown in Fig. 2 for set sizes around 400, where the paper itself demonstrates that MMD and 1-NNA have substantial sampling variability. Without repeated runs or bootstrap confidence intervals, the ranking of DualOctree over SDF and Shape2VecSet is not statistically established. Please report variance or otherwise quantify the uncertainty of the headline generation rankings.
minor comments (5)
  1. [Table 3] The header 'CD (*1e-4)' is ambiguous. If the values are actual Chamfer distances (~0.012), the notation is wrong; if they are scaled by 1e-4, the table should show appropriate decimal notation. This ambiguity propagates to the error-decomposition discussion in §4.5.
  2. [Abstract / Contributions] The introduction states 'Reconstruction errors are as high as 20% of the generation error,' while §4.5 reports 12.9% and 39.3%. These numbers should be reconciled or the contribution statement should match the final quantitative claim.
  3. [Appendix C] Typo: 'Iversion bracket' should be 'Iverson bracket'.
  4. [Table 4] The Objaverse CD values include standard deviations such as '0.016 ± 10.51' and '0.009 ± 4.948', which appear to have inconsistent scaling relative to the mean. Check whether these are percentage-scaled or absolute deviations.
  5. [§4.5] The phrase 'reconstruction errors are taken from the analysis in Figure 7b' is confusing because Fig. 7b reports F-score for SDF grid conversion under different preprocessing methods, not per-representation reconstruction errors. Clarify what data were actually used to compute the 12.9% share.

Circularity Check

0 steps flagged

No significant circularity: the benchmark is externally anchored; the only self-citation is not load-bearing.

full rationale

This paper is an empirical benchmark rather than a derivation. Its central claims—representation rankings for reconstruction/generation and the reported fractions of generation MMD attributable to reconstruction/compression errors—are computed from measured errors on external datasets (ShapeNet, Objaverse) and a user study, not derived from the definitions of the representations themselves. The autoencoders are trained for reconstruction and then evaluated with geometry metrics, but the paper does not specify that the training losses are identical to the CD/F-score/NC evaluation metrics, so the ranking is not a fitted parameter renamed as a prediction. The only self-citation (Leboutet et al. 2024, in the Outlook section) is not load-bearing. The error-decomposition percentages in Fig. 6 are not straightforwardly reproducible from Tables 2 and 3 and may double-count conversion error, but that is a correctness/reproducibility concern rather than circularity: the claimed shares are ratios of independent measurements, and the paper explicitly notes the errors are "not simply additive." No step reduces to its own input by construction, and no author-imposed uniqueness theorem is invoked to force a choice. Therefore, no significant circularity is present.

Axiom & Free-Parameter Ledger

6 free parameters · 4 axioms · 0 invented entities

The central comparison rests less on mathematical assumptions and more on modeling choices: fixed grid resolution, truncation, tree depth, and evaluation set size, plus the domain assumptions that meshes are ground truth and that latent diffusion is representative. No new physical entities are introduced.

free parameters (6)
  • Grid resolution 64^3 for voxel/SDF = 64^3
    Fixed by hand for all grid-based representations; higher resolutions improve reconstruction (Appendix H) but are memory-limited, so the choice affects the ranking.
  • SDF truncation cutoff = 0.2
    Chosen to balance expressiveness and memory; used for grid SDF conversion in Section 2.2.
  • Octree depth = 6
    Chosen in all Dual Octree experiments (Section 2.2); depth controls resolution and latent size.
  • Latent patch size for transformer tokenization = 4^3
    Chosen for tokenizing latent tensors for DiT (Section 2.2); affects transformer receptive field.
  • Evaluation set size for generation metrics = 400 (ShapeNet), 183 (Objaverse)
    Number of generated and reference samples used for COV/MMD/1-NNA; the paper itself shows these metrics are unstable below about 200 samples, and the Objaverse run uses 183.
  • Learning rates and schedules per model = not listed in main text
    Found by hyperparameter search (Appendix D); differences in tuning could favor some representations.
axioms (4)
  • domain assumption Meshes are the ground truth for evaluating 3D representation quality.
    Section 2 states 'The target representation in this study are meshes' and selects datasets that provide meshes. This excludes representations centered on rendering, like Gaussian splatting, from the comparison by design.
  • domain assumption Latent diffusion is representative enough that conclusions generalize to other generators (GANs, autoregressive).
    Section 2.1 says the main findings are generic despite using only diffusion models; this is asserted rather than shown.
  • ad hoc to paper Standardized pipeline isolates the representation as the source of performance differences.
    The central comparability assumption (Section 2) is necessary to interpret rankings, but it is not independently verified and is weakened by differing encoders and diffusers.
  • domain assumption Chamfer Distance, F-score, Normal Consistency, COV/MMD/1-NNA capture meaningful shape quality.
    Used throughout; the paper itself notes CD fails to capture perceptual detail and runs a user study to compensate, implying the metrics are imperfect.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Unifi3D: A Study on 3D Representations for Generation and Reconstruction in a Common Framework." pith.science (2026). https://pith.science/paper/77HDYR5D

@misc{pith2026250902474,
  author       = {Pith},
  title        = {Pith review of: Unifi3D: A Study on 3D Representations for Generation and Reconstruction in a Common Framework},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/77HDYR5D}},
  note         = {Machine review of arXiv:2509.02474}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Following rapid advancements in text and image generation, research has increasingly shifted towards 3D generation. Unlike the well-established pixel-based representation in images, 3D representations remain diverse and fragmented, encompassing a wide variety of approaches such as voxel grids, neural radiance fields, signed distance functions, point clouds, or octrees, each offering distinct advantages and limitations. In this work, we present a unified evaluation framework designed to assess the performance of 3D representations in reconstruction and generation. We compare these representations based on multiple criteria: quality, computational efficiency, and generalization performance. Beyond standard model benchmarking, our experiments aim to derive best practices over all steps involved in the 3D generation pipeline, including preprocessing, mesh reconstruction, compression with autoencoders, and generation. Our findings highlight that reconstruction errors significantly impact overall performance, underscoring the need to evaluate generation and reconstruction jointly. We provide insights that can inform the selection of suitable 3D models for various applications, facilitating the development of more robust and application-specific solutions in 3D generation. The code for our framework is available at https://github.com/isl-org/unifi3d.

Figures

Figures reproduced from arXiv: 2509.02474 by Benjamin Ummenhofer, Kai Yuan, Katelyn Gao, Michael Paulitsch, Nina Wiedemann, Quentin Leboutet, Sainan Liu.

Figure 1
Figure 1. Figure 1: Overview of the steps involved in a standard 3D generation pipeline: a) the mesh is transformed into a suitable [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: We select 100 random subsets of different sizes of the train and test split of the ShapeNet airplane category [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Rankings of 3D repre￾sentations based on generation and reconstruction metrics. The outer circle indicates the top rank. This chart compares reconstruction met￾rics (CD, F-score, NC), reconstruc￾tion generalization on an OOD task, and reconstruction efficiency (mem￾ory footprint, encoding size, and in￾ference time) as well as the metrics for unconditional generation per￾formance with user study rankings (1… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative results for mesh generation. We show results for each representation using the best encoder [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Reconstruction quality by runtime for inference [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: We compare different methods for converting a mesh to an SDF. [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative results for mesh reconstruction. [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative results for the Airplane category [PITH_FULL_IMAGE:figures/full_fig_p024_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative results for the Car category [PITH_FULL_IMAGE:figures/full_fig_p025_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Comparison of different mesh conversion methods in terms of Chamfer distance. [PITH_FULL_IMAGE:figures/full_fig_p025_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Reconstruction quality when converting to a sampled SDF and back [PITH_FULL_IMAGE:figures/full_fig_p026_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Distribution of F-scores between original and reconstructed mesh using an SDF grid of size [PITH_FULL_IMAGE:figures/full_fig_p026_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Effect of the grid resolution on the round trip conversion errors from mesh to SDF grid and back with our [PITH_FULL_IMAGE:figures/full_fig_p026_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Complexity analysis. Best viewed zoomed in. [PITH_FULL_IMAGE:figures/full_fig_p027_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Mandelbulb reconstruction with p as the vector of scores and yi as the preference labels collected from the participants. The results are visualized in [PITH_FULL_IMAGE:figures/full_fig_p027_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Surface-to-volume ratio for generated objects and ShapeNet, using categories [PITH_FULL_IMAGE:figures/full_fig_p027_17.png] view at source ↗

discussion (0)

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

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. A Cross-Model VLM-Judge Protocol for Single-Image 3D Mesh Quality (and Why Cheap Proxies Fall Short)

    cs.LG 2026-06 unverdicted novelty 6.0

    A reproducible VLM-judge protocol with position-bias correction is validated as superior to CLIP similarity and geometry-validity proxies for assessing single-image 3D mesh quality.

Reference graph

Works this paper leans on

108 extracted references · 44 canonical work pages · cited by 1 Pith paper · 2 internal anchors

  1. [1]

    Learning representations and generative models for 3d point clouds

    Panos Achlioptas, Olga Diamanti, Ioannis Mitliagkas, and Leonidas Guibas. Learning representations and generative models for 3d point clouds. In International conference on machine learning, pages 40--49. PMLR, 2018

    2018

  2. [2]

    Sal: Sign agnostic learning of shapes from raw data

    Matan Atzmon and Yaron Lipman. Sal: Sign agnostic learning of shapes from raw data. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2565--2574, 2020

    2020

  3. [3]

    Rank analysis of incomplete block designs: I

    Ralph Allan Bradley and Milton E Terry. Rank analysis of incomplete block designs: I. the method of paired comparisons. Biometrika, 39 0 (3/4): 0 324--345, 1952

    1952

  4. [4]

    Generative and discriminative voxel modeling with convolutional neural networks

    Andrew Brock, Theodore Lim, James M Ritchie, and Nick Weston. Generative and discriminative voxel modeling with convolutional neural networks. arXiv preprint arXiv:1608.04236, 2016

    Pith/arXiv arXiv 2016

  5. [5]

    Learning gradient fields for shape generation

    Ruojin Cai, Guandao Yang, Hadar Averbuch-Elor, Zekun Hao, Serge Belongie, Noah Snavely, and Bharath Hariharan. Learning gradient fields for shape generation. In Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part III 16, pages 364--381. Springer, 2020

    2020

  6. [6]

    A survey on generative diffusion models

    Hanqun Cao, Cheng Tan, Zhangyang Gao, Yilun Xu, Guangyong Chen, Pheng-Ann Heng, and Stan Z Li. A survey on generative diffusion models. IEEE transactions on knowledge and data engineering, 36 0 (7): 0 2814--2830, 2024

    2024

  7. [7]

    Efficient geometry-aware 3d generative adversarial networks

    Eric R Chan, Connor Z Lin, Matthew A Chan, Koki Nagano, Boxiao Pan, Shalini De Mello, Orazio Gallo, Leonidas J Guibas, Jonathan Tremblay, Sameh Khamis, et al. Efficient geometry-aware 3d generative adversarial networks. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 16123--16133, 2022

    2022

  8. [8]

    Shapenet: An information-rich 3d model repository

    Angel X Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio Savarese, Manolis Savva, Shuran Song, Hao Su, et al. Shapenet: An information-rich 3d model repository. arXiv preprint arXiv:1512.03012, 2015

    Pith/arXiv arXiv 2015

  9. [9]

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

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

    2023

  10. [10]

    Fantasia3d: Disentangling geometry and appearance for high-quality text-to-3d content creation

    Rui Chen, Yongwei Chen, Ningxin Jiao, and Kui Jia. Fantasia3d: Disentangling geometry and appearance for high-quality text-to-3d content creation. In Proceedings of the IEEE/CVF international conference on computer vision, pages 22246--22256, 2023 b

    2023

  11. [11]

    Dora: Sampling and benchmarking for 3d shape variational auto-encoders

    Rui Chen, Jianfeng Zhang, Yixun Liang, Guan Luo, Weiyu Li, Jiarui Liu, Xiu Li, Xiaoxiao Long, Jiashi Feng, and Ping Tan. Dora: Sampling and benchmarking for 3d shape variational auto-encoders. arXiv preprint arXiv:2412.17808, 2024 a

    Pith/arXiv arXiv 2024

  12. [12]

    Meshxl: Neural coordinate field for generative 3d foundation models

    Sijin Chen, Xin Chen, Anqi Pang, Xianfang Zeng, Wei Cheng, Yijun Fu, Fukun Yin, Billzb Wang, Jingyi Yu, Gang Yu, Bin Fu, and Tao Chen. Meshxl: Neural coordinate field for generative 3d foundation models. In NeurIPS, 2024 b

    2024

  13. [13]

    gdna: Towards generative detailed neural avatars

    Xu Chen, Tianjian Jiang, Jie Song, Jinlong Yang, Michael J Black, Andreas Geiger, and Otmar Hilliges. gdna: Towards generative detailed neural avatars. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20427--20437, 2022

    2022

  14. [14]

    Autoregressive 3d shape generation via canonical mapping

    An-Chieh Cheng, Xueting Li, Sifei Liu, Min Sun, and Ming-Hsuan Yang. Autoregressive 3d shape generation via canonical mapping. In European Conference on Computer Vision, pages 89--104. Springer, 2022

    2022

  15. [15]

    SDFusion : Multimodal 3d shape completion, reconstruction, and generation

    Yen-Chi Cheng, Hsin-Ying Lee, Sergey Tulyakov, Alexander G Schwing, and Liang-Yan Gui. SDFusion : Multimodal 3d shape completion, reconstruction, and generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4456--4465, 2023

    2023

  16. [16]

    3d-r2n2: A unified approach for single and multi-view 3d object reconstruction

    Christopher B Choy, Danfei Xu, JunYoung Gwak, Kevin Chen, and Silvio Savarese. 3d-r2n2: A unified approach for single and multi-view 3d object reconstruction. In Computer Vision--ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11-14, 2016, Proceedings, Part VIII 14, pages 628--644. Springer, 2016

    2016

  17. [17]

    Scalable high-resolution pixel-space image synthesis with hourglass diffusion transformers

    Katherine Crowson, Stefan Andreas Baumann, Alex Birch, Tanishq Mathew Abraham, Daniel Z Kaplan, and Enrico Shippole. Scalable high-resolution pixel-space image synthesis with hourglass diffusion transformers. In Forty-first International Conference on Machine Learning, 2024

    2024

  18. [18]

    Scancomplete: Large-scale scene completion and semantic segmentation for 3d scans

    Angela Dai, Daniel Ritchie, Martin Bokeloh, Scott Reed, J \"u rgen Sturm, and Matthias Nie ner. Scancomplete: Large-scale scene completion and semantic segmentation for 3d scans. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 4578--4587, 2018

    2018

  19. [19]

    Objaverse: A universe of annotated 3d objects

    Matt Deitke, Dustin Schwenk, Jordi Salvador, Luca Weihs, Oscar Michel, Eli VanderBilt, Ludwig Schmidt, Kiana Ehsani, Aniruddha Kembhavi, and Ali Farhadi. Objaverse: A universe of annotated 3d objects. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13142--13153, 2023

    2023

  20. [20]

    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 Voleti, Samir Yitzhak Gadre, et al. Objaverse-xl: A universe of 10m+ 3d objects. Advances in Neural Information Processing Systems, 36, 2024

    2024

  21. [21]

    Boost your nerf: A model-agnostic mixture of experts framework for high quality and efficient rendering

    Francesco Di Sario, Riccardo Renzulli, Enzo Tartaglione, and Marco Grangetto. Boost your nerf: A model-agnostic mixture of experts framework for high quality and efficient rendering. In European Conference on Computer Vision, pages 176--192. Springer, 2024

    2024

  22. [22]

    Plenoxels: Radiance fields without neural networks

    Sara Fridovich-Keil, Alex Yu, Matthew Tancik, Qinhong Chen, Benjamin Recht, and Angjoo Kanazawa. Plenoxels: Radiance fields without neural networks. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5501--5510, 2022

    2022

  23. [23]

    Representing scenes as compositional generative neural feature fields based on giraffe for 3d reconstruction of classroom scenes

    Tianlin Fu, Renyu Deng, Yue Gao, and Fuquan Zhang. Representing scenes as compositional generative neural feature fields based on giraffe for 3d reconstruction of classroom scenes. In International Conference on Intelligent Information Hiding and Multimedia Signal Processing, pages 227--237. Springer, 2022

    2022

  24. [24]

    Get3d: A generative model of high quality 3d textured shapes learned from images

    Jun Gao, Tianchang Shen, Zian Wang, Wenzheng Chen, Kangxue Yin, Daiqing Li, Or Litany, Zan Gojcic, and Sanja Fidler. Get3d: A generative model of high quality 3d textured shapes learned from images. Advances In Neural Information Processing Systems, 35: 0 31841--31854, 2022

    2022

  25. [25]

    Sdm-net: Deep generative network for structured deformable mesh

    Lin Gao, Jie Yang, Tong Wu, Yu-Jie Yuan, Hongbo Fu, Yu-Kun Lai, and Hao Zhang. Sdm-net: Deep generative network for structured deformable mesh. ACM Transactions on Graphics (TOG), 38 0 (6): 0 1--15, 2019

    2019

  26. [26]

    A review on deep learning approaches for 3d data representations in retrieval and classifications

    Abubakar Sulaiman Gezawa, Yan Zhang, Qicong Wang, and Lei Yunqi. A review on deep learning approaches for 3d data representations in retrieval and classifications. IEEE access, 8: 0 57566--57593, 2020

    2020

  27. [27]

    3dgen: Triplane latent diffusion for textured mesh generation

    Anchit Gupta, Wenhan Xiong, Yixin Nie, Ian Jones, and Barlas O g uz. 3dgen: Triplane latent diffusion for textured mesh generation. arXiv preprint arXiv:2303.05371, 2023

    Pith/arXiv arXiv 2023

  28. [28]

    Sparseflex: High-resolution and arbitrary-topology 3d shape modeling

    Xianglong He, Zi-Xin Zou, Chia-Hao Chen, Yuan-Chen Guo, Ding Liang, Chun Yuan, Wanli Ouyang, Yan-Pei Cao, and Yangguang Li. Sparseflex: High-resolution and arbitrary-topology 3d shape modeling. arXiv preprint arXiv:2503.21732, 2025

    Pith/arXiv arXiv 2025

  29. [29]

    Clip goes 3d: Leveraging prompt tuning for language grounded 3d recognition

    Deepti Hegde, Jeya Maria Jose Valanarasu, and Vishal Patel. Clip goes 3d: Leveraging prompt tuning for language grounded 3d recognition. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 2028--2038, 2023

    2028

  30. [30]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33: 0 6840--6851, 2020

    2020

  31. [31]

    Robust watertight manifold surface generation method for shapenet models

    Jingwei Huang, Hao Su, and Leonidas Guibas. Robust watertight manifold surface generation method for shapenet models. arXiv preprint arXiv:1802.01698, 2018

    Pith/arXiv arXiv 2018

  32. [32]

    Manifoldplus: A robust and scalable watertight manifold surface generation method for triangle soups

    Jingwei Huang, Yichao Zhou, and Leonidas Guibas. Manifoldplus: A robust and scalable watertight manifold surface generation method for triangle soups. arXiv preprint arXiv:2005.11621, 2020

    Pith/arXiv arXiv 2005

  33. [33]

    Make-a-shape: a ten-million-scale 3d shape model

    Ka-Hei Hui, Aditya Sanghi, Arianna Rampini, Kamal Rahimi Malekshan, Zhengzhe Liu, Hooman Shayani, and Chi-Wing Fu. Make-a-shape: a ten-million-scale 3d shape model. In Forty-first International Conference on Machine Learning, 2024

    2024

  34. [34]

    A survey on text-to-3d contents generation in the wild

    Chenhan Jiang. A survey on text-to-3d contents generation in the wild. arXiv preprint arXiv:2405.09431, 2024

    Pith/arXiv arXiv 2024

  35. [35]

    Shap-e: Generating conditional 3d implicit functions

    Heewoo Jun and Alex Nichol. Shap-e: Generating conditional 3d implicit functions. arXiv preprint arXiv:2305.02463, 2023

    Pith/arXiv arXiv 2023

  36. [36]

    Holodiffusion: Training a 3d diffusion model using 2d images

    Animesh Karnewar, Andrea Vedaldi, David Novotny, and Niloy J Mitra. Holodiffusion: Training a 3d diffusion model using 2d images. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 18423--18433, 2023

    2023

  37. [37]

    3d gaussian splatting for real-time radiance field rendering

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk \"u hler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph., 42 0 (4): 0 139--1, 2023

    2023

  38. [38]

    Softflow: Probabilistic framework for normalizing flow on manifolds

    Hyeongju Kim, Hyeonseung Lee, Woo Hyun Kang, Joun Yeop Lee, and Nam Soo Kim. Softflow: Probabilistic framework for normalizing flow on manifolds. Advances in Neural Information Processing Systems, 33: 0 16388--16397, 2020

    work page 2020

  39. [39]

    Neuralfield-ldm: Scene generation with hierarchical latent diffusion models

    Seung Wook Kim, Bradley Brown, Kangxue Yin, Karsten Kreis, Katja Schwarz, Daiqing Li, Robin Rombach, Antonio Torralba, and Sanja Fidler. Neuralfield-ldm: Scene generation with hierarchical latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8496--8506, 2023

    work page 2023

  40. [40]

    Discrete point flow networks for efficient point cloud generation

    Roman Klokov, Edmond Boyer, and Jakob Verbeek. Discrete point flow networks for efficient point cloud generation. In European Conference on Computer Vision, pages 694--710. Springer, 2020

    work page 2020

  41. [41]

    Gaussiananything: Interactive point cloud flow matching for 3d object generation

    Yushi Lan, Shangchen Zhou, Zhaoyang Lyu, Fangzhou Hong, Shuai Yang, Bo Dai, Xingang Pan, and Chen Change Loy. Gaussiananything: Interactive point cloud flow matching for 3d object generation. arXiv preprint arXiv:2411.08033, 2024

    Pith/arXiv arXiv 2024

  42. [42]

    Midgard: Modular interpretable diffusion over graphs for articulated designs

    Quentin Leboutet, Nina Wiedemann, Zhipeng Cai, Michael Paulitsch, and Kai Yuan. Midgard: Modular interpretable diffusion over graphs for articulated designs. In Advances in Neural Information Processing Systems, 2024

    work page 2024

  43. [43]

    Nap: Neural 3d articulated object prior

    Jiahui Lei, Congyue Deng, William B Shen, Leonidas J Guibas, and Kostas Daniilidis. Nap: Neural 3d articulated object prior. Advances in Neural Information Processing Systems, 36: 0 31878--31894, 2023

    work page 2023

  44. [44]

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

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

    work page 2023

  45. [45]

    Advances in 3d generation: A survey

    Xiaoyu Li, Qi Zhang, Di Kang, Weihao Cheng, Yiming Gao, Jingbo Zhang, Zhihao Liang, Jing Liao, Yan-Pei Cao, and Ying Shan. Advances in 3d generation: A survey. arXiv preprint arXiv:2401.17807, 2024

    Pith/arXiv arXiv 2024

  46. [46]

    Triposg: High-fidelity 3d shape synthesis using large-scale rectified flow models

    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. arXiv preprint arXiv:2502.06608, 2025

    Pith/arXiv arXiv 2025

  47. [47]

    Magic3d: High-resolution text-to-3d content 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 content creation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 300--309, 2023

    work page 2023

  48. [48]

    A comprehensive survey on 3d content generation

    Jian Liu, Xiaoshui Huang, Tianyu Huang, Lu Chen, Yuenan Hou, Shixiang Tang, Ziwei Liu, Wanli Ouyang, Wangmeng Zuo, Junjun Jiang, et al. A comprehensive survey on 3d content generation. arXiv preprint arXiv:2402.01166, 2024 a

    Pith/arXiv arXiv 2024

  49. [49]

    Cage: Controllable articulation generation

    Jiayi Liu, Hou In Ivan Tam, Ali Mahdavi-Amiri, and Manolis Savva. Cage: Controllable articulation generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17880--17889, 2024 b

    work page 2024

  50. [50]

    Neural sparse voxel fields

    Lingjie Liu, Jiatao Gu, Kyaw Zaw Lin, Tat-Seng Chua, and Christian Theobalt. Neural sparse voxel fields. Advances in Neural Information Processing Systems, 33: 0 15651--15663, 2020

    work page 2020

  51. [51]

    One-2-3-45++: Fast single image to 3d objects with consistent multi-view generation and 3d diffusion

    Minghua Liu, Ruoxi Shi, Linghao Chen, Zhuoyang Zhang, Chao Xu, Xinyue Wei, Hansheng Chen, Chong Zeng, Jiayuan Gu, and Hao Su. One-2-3-45++: Fast single image to 3d objects with consistent multi-view generation and 3d diffusion. arXiv preprint arXiv:2311.07885, 2023 a

    Pith/arXiv arXiv 2023

  52. [52]

    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. Advances in Neural Information Processing Systems, 36, 2024 c

    work page 2024

  53. [53]

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

    Ruoshi Liu, Rundi Wu, Basile Van Hoorick, Pavel Tokmakov, Sergey Zakharov, and Carl Vondrick. Zero-1-to-3: Zero-shot one image to 3d object. In Proceedings of the IEEE/CVF international conference on computer vision, pages 9298--9309, 2023 b

    work page 2023

  54. [54]

    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. In CVPR , pages 9970--9980. IEEE , 2024

    work page 2024

  55. [55]

    Lorensen and Harvey E

    William E. Lorensen and Harvey E. Cline. Marching cubes: A high resolution 3d surface construction algorithm. In SIGGRAPH , pages 163--169. ACM , 1987

    work page 1987

  56. [56]

    Surfgen: Adversarial 3d shape synthesis with explicit surface discriminators

    Andrew Luo, Tianqin Li, Wen-Hao Zhang, and Tai Sing Lee. Surfgen: Adversarial 3d shape synthesis with explicit surface discriminators. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 16238--16248, 2021

    work page 2021

  57. [57]

    Diffusion probabilistic models for 3d point cloud generation

    Shitong Luo and Wei Hu. Diffusion probabilistic models for 3d point cloud generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2837--2845, 2021

    work page 2021

  58. [58]

    A conditional point diffusion-refinement paradigm for 3d point cloud completion

    Zhaoyang Lyu, Zhifeng Kong, Xudong Xu, Liang Pan, and Dahua Lin. A conditional point diffusion-refinement paradigm for 3d point cloud completion. arXiv preprint arXiv:2112.03530, 2021

    Pith/arXiv arXiv 2021

  59. [59]

    Voxnet: A 3d convolutional neural network for real-time object recognition

    Daniel Maturana and Sebastian Scherer. Voxnet: A 3d convolutional neural network for real-time object recognition. In 2015 IEEE/RSJ international conference on intelligent robots and systems (IROS), pages 922--928. IEEE, 2015

    work page 2015

  60. [60]

    Occupancy networks: Learning 3d reconstruction in function space

    Lars Mescheder, Michael Oechsle, Michael Niemeyer, Sebastian Nowozin, and Andreas Geiger. Occupancy networks: Learning 3d reconstruction in function space. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4460--4470, 2019

    2019

  61. [61]

    Nerf: Representing scenes as neural radiance fields for view synthesis

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view synthesis. Communications of the ACM, 65 0 (1): 0 99--106, 2021

    2021

  62. [62]

    Autosdf: Shape priors for 3d completion, reconstruction and generation

    Paritosh Mittal, Yen-Chi Cheng, Maneesh Singh, and Shubham Tulsiani. Autosdf: Shape priors for 3d completion, reconstruction and generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 306--315, 2022

    work page 2022

  63. [63]

    Instant neural graphics primitives with a multiresolution hash encoding

    Thomas M \"u ller, Alex Evans, Christoph Schied, and Alexander Keller. Instant neural graphics primitives with a multiresolution hash encoding. ACM transactions on graphics (TOG), 41 0 (4): 0 1--15, 2022

    work page 2022

  64. [64]

    3d-ldm: Neural implicit 3d shape generation with latent diffusion models

    Gimin Nam, Mariem Khlifi, Andrew Rodriguez, Alberto Tono, Linqi Zhou, and Paul Guerrero. 3d-ldm: Neural implicit 3d shape generation with latent diffusion models. arXiv preprint arXiv:2212.00842, 2022

    Pith/arXiv arXiv 2022

  65. [65]

    Polygen: An autoregressive generative model of 3d meshes

    Charlie Nash, Yaroslav Ganin, SM Ali Eslami, and Peter Battaglia. Polygen: An autoregressive generative model of 3d meshes. In International conference on machine learning, pages 7220--7229. PMLR, 2020

    work page 2020

  66. [66]

    Blockgan: Learning 3d object-aware scene representations from unlabelled images

    Thu H Nguyen-Phuoc, Christian Richardt, Long Mai, Yongliang Yang, and Niloy Mitra. Blockgan: Learning 3d object-aware scene representations from unlabelled images. Advances in neural information processing systems, 33: 0 6767--6778, 2020

    work page 2020

  67. [67]

    Point-e: A system for generating 3d point clouds from complex prompts

    Alex Nichol, Heewoo Jun, Prafulla Dhariwal, Pamela Mishkin, and Mark Chen. Point-e: A system for generating 3d point clouds from complex prompts. arXiv preprint arXiv:2212.08751, 2022

    Pith/arXiv arXiv 2022

  68. [68]

    Deepsdf: Learning continuous signed distance functions for shape representation

    Jeong Joon Park, Peter Florence, Julian Straub, Richard Newcombe, and Steven Lovegrove. Deepsdf: Learning continuous signed distance functions for shape representation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 165--174, 2019

    2019

  69. [69]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 4195--4205, 2023

    work page 2023

  70. [70]

    Convolutional occupancy networks

    Songyou Peng, Michael Niemeyer, Lars Mescheder, Marc Pollefeys, and Andreas Geiger. Convolutional occupancy networks. In Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part III 16, pages 523--540. Springer, 2020

    work page 2020

  71. [71]

    State of the art on diffusion models for visual computing

    Ryan Po, Wang Yifan, Vladislav Golyanik, Kfir Aberman, Jonathan T Barron, Amit Bermano, Eric Chan, Tali Dekel, Aleksander Holynski, Angjoo Kanazawa, et al. State of the art on diffusion models for visual computing. In Computer graphics forum, volume 43, page e15063. Wiley Online Library, 2024

    work page 2024

  72. [72]

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

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

    Pith/arXiv arXiv 2022

  73. [73]

    Xcube: Large-scale 3d generative modeling using sparse voxel hierarchies

    Xuanchi Ren, Jiahui Huang, Xiaohui Zeng, Ken Museth, Sanja Fidler, and Francis Williams. Xcube: Large-scale 3d generative modeling using sparse voxel hierarchies. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4209--4219, 2024

    work page 2024

  74. [74]

    High-resolution image synthesis with latent diffusion models

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

    2022

  75. [75]

    Zero123++: a single image to consistent multi-view diffusion base model

    Ruoxi Shi, Hansheng Chen, Zhuoyang Zhang, Minghua Liu, Chao Xu, Xinyue Wei, Linghao Chen, Chong Zeng, and Hao Su. Zero123++: a single image to consistent multi-view diffusion base model. arXiv preprint arXiv:2310.15110, 2023 a

    Pith/arXiv arXiv 2023

  76. [76]

    Mvdream: Multi-view diffusion for 3d generation

    Yichun Shi, Peng Wang, Jianglong Ye, Mai Long, Kejie Li, and Xiao Yang. Mvdream: Multi-view diffusion for 3d generation. arXiv preprint arXiv:2308.16512, 2023 b

    Pith/arXiv arXiv 2023

  77. [77]

    3d point cloud generative adversarial network based on tree structured graph convolutions

    Dong Wook Shu, Sung Woo Park, and Junseok Kwon. 3d point cloud generative adversarial network based on tree structured graph convolutions. In Proceedings of the IEEE/CVF international conference on computer vision, pages 3859--3868, 2019

    work page 2019

  78. [78]

    Meshgpt: Generating triangle meshes with decoder-only transformers

    Yawar Siddiqui, Antonio Alliegro, Alexey Artemov, Tatiana Tommasi, Daniele Sirigatti, Vladislav Rosov, Angela Dai, and Matthias Nie ner. Meshgpt: Generating triangle meshes with decoder-only transformers. arXiv preprint arXiv:2311.15475, 2023

    Pith/arXiv arXiv 2023

  79. [79]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In International conference on machine learning, pages 2256--2265. PMLR, 2015

    work page 2015

  80. [80]

    Direct voxel grid optimization: Super-fast convergence for radiance fields reconstruction

    Cheng Sun, Min Sun, and Hwann-Tzong Chen. Direct voxel grid optimization: Super-fast convergence for radiance fields reconstruction. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5459--5469, 2022

    work page 2022

Showing first 80 references.

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.