Pith. sign in

REVIEW 3 major objections 5 minor 22 cited by

TexVerse: A Universe of 3D Objects with High-Resolution Textures

T0 review · 3 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read This paper introduces TexVerse, a curated dataset of 858,669 unique 3D models with textures of at least 1024 pixels, including 158,518 PBR-material models, and argues that this scale and resolution were previously missing for training high-

desk verdict Useful dataset resource whose headline numbers rest entirely on Sketchfab metadata; the high-resolution guarantee needs file-level proof before I trust it. read the letter →

arxiv 2508.10868 v2 pith:5AOYUWYD submitted 2025-08-14 cs.CV

Yibo Zhang , Li Zhang , Rui Ma , Nan Cao This is my paper
classification cs.CV
keywords 3Ddatasethigh-resolutiontexturesPBRmaterialsSketchfabgenerationriggedmodelsanimationannotation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper introduces TexVerse, a curated dataset of 858,669 Sketchfab 3D models that all have texture resolutions of at least 1024 pixels, including 158,518 models with physically based rendering (PBR) materials. The authors argue that existing large datasets either cap textures at low resolution or lack scale, so end-to-end high-resolution texture and PBR generation has been held back by missing training data. TexVerse collects every high-resolution variant of each model, producing 1,661,101 instances, and preserves original skeleton and animation files for 69,138 rigged and 54,430 animated models. If the dataset is as described, it gives the community a large-scale resource for training texture and PBR generators without relying on post-hoc super-resolution.

What carries the argument

The curation pipeline is the central mechanism: it filters Sketchfab's freely downloadable models using metadata fields such as max-texture, pbr-type, is-rigged, animation-count, license, and NoAI tags, then packages the filtered models as .glb while retaining all high-resolution variants of each model. The pipeline re-fetches original model files for rigged and animated objects to avoid losing skeletons and animations, and a thumbnail-prompting step uses GPT-5 to produce fixed three-sentence annotations. This metadata-driven assembly is what turns a raw pool of about 1.6 million models into the claimed 858,669-model dataset.

What would settle it

Download a random sample of, say, 500 of the 858,669 .glb models and read the actual texture image dimensions inside each file instead of the metadata; if a meaningful fraction have a largest dimension below 1024 pixels or contain no texture map, the central high-resolution claim fails. Similarly, checking whether the claimed PBR models actually contain roughness/glossiness and metalness/specular maps would settle the PBR subset claim.

Watch

Extended reading notes

Core claim

The paper's central claim is that TexVerse is the first large-scale 3D object dataset in which every included model genuinely has texture resolution of at least 1024 pixels, with many at 2048, 4096, or 8192. It curates 858,669 unique models from Sketchfab, standardizes them to .glb, and additionally keeps every high-resolution variant, yielding 1,661,101 instances; 158,518 of the models carry PBR materials under metalness or specular workflows. For 69,138 rigged and 54,430 animated models, it preserves original user-uploaded files so skeletons and animations survive conversion. It also provides 856,312 GPT-5-generated structured annotations. The authors argue that this combination of scale,

Load-bearing premise

The headline counts and filters rest on Sketchfab's self-reported metadata, especially max-texture and pbr-type, being accurate for every model; the paper does not verify textures or PBR maps file-by-file.

Editorial extensions

If this is right

  • High-resolution texture generation can be trained directly on TexVerse without relying on super-resolution post-processing.
  • The 158,518-model PBR subset provides scale for learning roughness, metalness, specular, and normal maps jointly.
  • The multi-resolution variants give paired supervision for texture super-resolution and level-of-detail consistency.
  • The rigged and animated subsets open a path to generating models with usable skeletons and animations from the start.
  • The 856,312 structured annotations create a large image-to-description corpus for text-to-3D and asset captioning.

Reading between the lines

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

  • Because each unique model ships with all of its high-resolution variants, TexVerse can serve as a ready-made paired multi-resolution dataset for texture super-resolution, even though the paper does not frame it that way.
  • The rigged and animated subsets preserve skeletons and animations in the original formats, suggesting downstream work on skeleton-aware generation and retargeting rather than only static texture synthesis.
  • If the metadata-based filter is later combined with automatic texture-quality checks, the dataset could be re-released as a smaller but cleaner core with verified per-file guarantees; the paper explicitly leaves this verification to future work.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. TexVerse is a large-scale 3D dataset curated from Sketchfab, claiming 858,669 unique high-resolution textured models, 158,518 with PBR materials, and 1,661,101 total instances when all resolution variants are included. The dataset also includes TexVerse-Skeleton (69,138 rigged models) and TexVerse-Animation (54,430 animated models), preserving original upload formats, plus 856,312 GPT-5-generated annotations. The authors filter Sketchfab metadata for models with max texture resolution at least 1024 pixels, exclude NoAI-tagged content, and retain only distributable Creative Commons licenses, distributing results in .glb format.

Significance. If the dataset is as described, it would fill a genuine gap: existing large-scale datasets such as Objaverse provide mostly low-resolution textures, while high-quality PBR datasets such as DTC are too small. The scale, license filtering, preservation of rigs/animations, and PBR subset are potentially valuable contributions for texture synthesis, PBR material generation, and 3D vision. The paper is transparent about the curation pipeline and includes helpful statistics and examples. However, the central quantitative claims rest entirely on Sketchfab metadata with no file-level verification, and the paper itself acknowledges that metadata may contain annotation errors. This makes the current evidence for 'genuine high-resolution textures' and reliable PBR labelling weaker than the headline claims suggest.

major comments (3)
  1. [Section 3, first paragraph; Table 1] The central claim that all 858,669 models have high-resolution textures is based exclusively on Sketchfab's `max-texture` metadata, not on inspection of the downloaded .glb files. The Limitations section concedes 'occasional annotation errors' and states that 'further filtering and cleaning are needed to ensure consistent geometric quality and texture clarity.' Because Sketchfab's conversion to .glb can alter or repackage textures, the metadata may not reflect the actual distributed files. Please provide validation statistics, e.g., actual maximum texture dimensions and channel presence computed from a random sample of downloaded files, and a manifest or checksums. Table 1's unqualified high-resolution checkmark should be qualified accordingly.
  2. [Section 3.2, PBR material paragraph] The PBR definition ('each material must include a texture in the roughness or glossiness channel, as well as in either the metalness or specular channel') is applied to Sketchfab's `pbr-type` metadata, not to the actual material textures in the files. The paper does not report how many models satisfy this definition when the files are examined, nor does it provide rendered validation or a distribution of channel combinations. Since 158,518 PBR models are a headline contribution, this needs file-level verification on a representative sample.
  3. [Section 3.2, Fig. 3 and Objaverse comparison] The only evidence that TexVerse provides 'genuine high-resolution versions' where Objaverse does not is two thumbnail examples (Fig. 3). The comparison with Objaverse is a central motivation, but it is not backed by aggregate measurements from the downloaded files. Similarly, the claim that 'nearly 60% of TexVerse models are novel' is based on uid overlap; please define this metric precisely and report overlap statistics rather than leaving it implicit. A sample-based histogram of actual texture resolutions in TexVerse and Objaverse would directly support the main claim and is within scope.
minor comments (5)
  1. [Title and throughout] The name 'TexVerse' is inconsistently spaced as 'TexV erse' in the title and elsewhere; please fix.
  2. [Section 3.3] The GPT-5 annotations are generated only from thumbnails. No human evaluation or quality metric is reported. Please add a small human study or at least an error analysis to gauge annotation reliability.
  3. [Section 3] The description of collecting 'all of its high-resolution variants (e.g., the 4096 and 1024 versions of a model with a maximum resolution of 8192)' is slightly ambiguous. Clarify how variants are identified and why some are included and others not.
  4. [Section 3.1] The metadata list includes `face-count` and `vertex-count` but the paper does not report statistics on these. A short summary of geometric complexity distributions would help users assess suitability for downstream tasks.
  5. [Dataset availability] Only a GitHub repository link is provided. Please specify the actual data hosting, download procedure, and any planned versioning or maintenance, as dataset papers should make distribution mechanisms explicit.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: TexVerse is a dataset curation paper whose headline counts follow from its stated Sketchfab metadata filters; there is no derivation, prediction, or self-citation chain that reduces to its own inputs.

full rationale

The paper reports a curated dataset, not a derivation. Its central numbers (858,669 models, 158,518 PBR models, 1,661,101 instances) are aggregate counts obtained by applying Sketchfab metadata filters (max-texture >= 1024, license restrictions, NoAI exclusion) to Sketchfab's catalog. No equation, fitted parameter, or generative model is involved, so there is no derivation chain in which an output is defined as its own input. The only candidate concern is that the 'high-resolution' and PBR labels inherit Sketchfab metadata without file-level verification; the paper itself states in Limitations: 'We determine model resolution based on metadata from Sketchfab, which may contain occasional annotation errors.' That is an empirical reliability limitation, not circularity: the dataset construction and its advertised property are the same curation operation, but this is a dataset report rather than a derivation. There are no load-bearing self-citations and no imported uniqueness theorems; citations to Objaverse, Objaverse-XL, and DTC are contextual comparisons. Accordingly, no circular step is present.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

No fitted parameters exist. The paper is a dataset curation effort, so its quantitative claims rest on Sketchfab metadata, license interpretation, and GPT-5 annotation quality rather than on mathematical axioms. TexVerse is a dataset artifact, not a theoretically postulated entity, so no invented entities are introduced.

assumptions (6)
  • domain assumption Sketchfab metadata fields (texture resolution, tags, license, rig flags, animation flags) are accurate enough to define dataset membership.
    All counts in Section 3.2 are computed from this metadata; the Limitations section concedes that resolution metadata may contain errors and that texture clarity needs cleaning.
  • domain assumption Creative Commons license classification via Sketchfab's API is legally sufficient for redistribution.
    Section 3.2 License states that all models are under distributable Creative Commons licenses, relying entirely on Sketchfab's license field.
  • domain assumption String matching for 'NoAI' tags and descriptions reliably excludes models whose creators object to AI use.
    Section 3 states models tagged or described with NoAI terms are excluded; no validation of tag completeness or false negatives is reported.
  • domain assumption Preserving original user-uploaded files prevents the loss of skeletons and animations during Sketchfab's .glb conversion.
    Section 3.2 Rigged and animated models argues this without showing a verification that preserved files are loadable and complete.
  • domain assumption GPT-5 thumbnail-based annotations are accurate enough to serve as dataset annotations.
    Section 3.3 generates 856,312 annotations with no reported human spot-check or quality evaluation.
  • domain assumption Texture resolution of 1024 pixels is a reasonable threshold for 'high-resolution'.
    Section 1 and Section 3 use 1024 pixels as the cutoff; many downstream texture generation works treat 1024 as the low-resolution baseline.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TexVerse: A Universe of 3D Objects with High-Resolution Textures." pith.science (2026). https://pith.science/paper/5AOYUWYD

@misc{pith2026250810868,
  author       = {Pith},
  title        = {Pith review of: TexVerse: A Universe of 3D Objects with High-Resolution Textures},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5AOYUWYD}},
  note         = {Machine review of arXiv:2508.10868}
}
read the original abstract

We introduce TexVerse, a large-scale 3D dataset featuring high-resolution textures. While recent advances in large-scale 3D datasets have enhanced high-resolution geometry generation, creating high-resolution textures end-to-end remains underexplored due to the lack of suitable datasets. TexVerse fills this gap with a curated collection of over 858K unique high-resolution 3D models sourced from Sketchfab, including more than 158K models with physically based rendering (PBR) materials. Each model encompasses all of its high-resolution variants, bringing the total to 1.6M 3D instances. TexVerse also includes specialized subsets: TexVerse-Skeleton, with 69K rigged models, and TexVerse-Animation, with 54K animated models, both preserving original skeleton and animation data uploaded by the user. We also provide detailed model annotations describing overall characteristics, structural components, and intricate features. TexVerse offers a high-quality data resource with wide-ranging potential applications in texture synthesis, PBR material development, animation, and various 3D vision and graphics tasks.

Figures

Figures reproduced from arXiv: 2508.10868 by the authors.

Figure 1
Figure 1. We introduce TexVerse, a large-scale 3D asset dataset featuring high-resolution textures. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. TexVerse statistics. (a) Distribution of high-level TexVerse categories; (b) Word cloud of metadata tags; (c) Metadata statistics, including metadata comparison between TexVerse and Objaverse, texture resolution distribution, and PBR material distribution. 3.1 Metadata For each object in the TexVerse, we provide metadata extracted from the information sup￾plied by its creator upon uploading to Sketchfab: uid, name, … view at source ↗
Figure 3
Figure 3. Objaverse only provides versions up to 1024 resolution for objects labeled as having [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Data example with 4K resolution PBR materials. Please zoom in to check the detail. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5 [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 22 Pith papers

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

  1. Every9D-21M: Large-Scale Real-World 9D Canonicalization of Everyday Objects

    cs.CV 2026-05 conditional novelty 8.0 of 10

    Every9D-21M supplies 21.8M real-world 9D pose annotations for 700 everyday categories by propagating manual canonical poses through cross-instance alignment in object-centric videos and verifying them multiview.

  2. SpatialBench: Is Your Spatial Foundation Model an All-Round Player?

    cs.CV 2026-05 unverdicted novelty 8.0 of 10

    SpatialBench evaluates 41 spatial foundation models across 6 paradigms and 5 task suites, finds they are not all-round players, and introduces the DA-Next-5M dataset plus DA-Next baseline model.

  3. Engine-Native Editable 3D World Reconstruction with Objects and Lighting

    cs.CV 2026-07 conditional novelty 7.0 of 10

    A UE5-derived dataset (Lumera-2K) and VLM pipeline parse single-image scenes into oriented object boxes and parametric light tuples, establishing a measurable benchmark for editable, light-aware 3D reconstruction.

  4. Compact Representation of Mipmapped SVBRDFs via Shared Gaussians

    cs.GR 2026-07 conditional novelty 6.0 of 10

    Shared 2D Gaussians across mip levels and SVBRDF maps beat ASTC on quality-per-byte while keeping random-access, non-neural GPU decoding.

  5. Ink3D: Sculpting 3D Assets with Extremely Complex Textures via Video Generative Models

    cs.CV 2026-07 unverdicted novelty 6.0 of 10

    Ink3D decouples geometry from texture by generating dense orbit videos with a conditional video model and baking them via a neural optimizer to produce complex 3D textures.

  6. Mesh BDF: Barycentric Dominance Field for 3D Native Mesh Generation

    cs.CV 2026-06 unverdicted novelty 6.0 of 10

    Barycentric Dominance Field converts discrete mesh connectivity into a continuous surface signal that diffusion models can use directly for higher-quality native 3D mesh generation.

  7. HAT-4D: Lifting Monocular Video for 4D Multi-Object Interactions via Human-Agent Collaboration

    cs.CV 2026-06 unverdicted novelty 6.0 of 10

    HAT-4D presents an agentic VLM-plus-human-in-the-loop pipeline for monocular 4D multi-object interaction reconstruction and releases the MVOIK-4D benchmark.

  8. Extracting Neural Materials from Multi-view Images

    cs.CV 2026-06 unverdicted novelty 6.0 of 10

    NeuMatEx combines a Large Material Reconstruction Model for initialization and uncertainty-guided inverse path tracing to extract spatially varying neural materials from multi-view images.

  9. Extracting Neural Materials from Multi-view Images

    cs.CV 2026-06 unverdicted novelty 6.0 of 10

    NeuMatEx extracts spatially varying neural materials from multi-view images by using a learned LMRM prior for initialization followed by uncertainty-guided inverse path tracing optimization.

  10. Helix4D: Complex 4D Mesh Generation

    cs.CV 2026-05 unverdicted novelty 6.0 of 10

    Helix4D generates high-quality dynamic 4D meshes from videos by extending Trellis2 with sliding-window cross-frame attention anchored on the first frame and a repurposed 4D temporal encoding.

  11. H-OmniStereo: Zero-Shot Omnidirectional Stereo Matching with Heading-Aligned Normal Priors

    cs.CV 2026-05 unverdicted novelty 6.0 of 10

    H-OmniStereo trains a stereo matcher on 2.8 million synthetic equirectangular pairs and adds a heading-aligned normal prior to improve zero-shot accuracy and generalization on out-of-domain and real omnidirectional data.

  12. Camera Control for Text-to-Image Generation via Learning Viewpoint Tokens

    cs.CV 2026-04 unverdicted novelty 6.0 of 10

    Viewpoint tokens learned on a mixed 3D-rendered and photorealistic dataset enable precise camera control in text-to-image generation while factorizing geometry from appearance and transferring to unseen object categories.

  13. Geometric Context Transformer for Streaming 3D Reconstruction

    cs.CV 2026-04 unverdicted novelty 6.0 of 10

    LingBot-Map is a streaming 3D reconstruction model built on a geometric context transformer that combines anchor context, pose-reference window, and trajectory memory to deliver accurate, drift-resistant results at 20...

  14. Native and Compact Structured Latents for 3D Generation

    cs.CV 2025-12 unverdicted novelty 6.0 of 10

    Introduces O-Voxel omni-voxel representation and Sparse Compression VAE for structured native 3D latents, enabling efficient training of large flow-matching models that produce higher-quality geometry and materials th...

  15. NI-Tex: Non-isometric Image-based Garment Texture Generation

    cs.CV 2025-11 conditional novelty 6.0 of 10

    A training framework that makes image-to-garment texture transfer robust to pose and topology mismatch, using simulated garment videos, AI image editing, and uncertainty-guided multi-view baking.

  16. Droplet3D: Commonsense Priors from Videos Facilitate 3D Generation

    cs.CV 2025-08 conditional novelty 6.0 of 10

    A video diffusion backbone fine-tuned on 4M densely captioned 360-degree renderings generates spatially consistent multi-view images for 3D assets from image plus detailed text input.

  17. Sat2City v2: Native 3D City Asset Generation from a Single Satellite Image

    cs.CV 2026-06 unverdicted novelty 5.0 of 10

    Sat2City v2 adapts a pretrained native 3D latent model to generate controllable textured 3D city assets from satellite images via geometry flow fine-tuning and anchored texturing on a collected real dataset.

  18. High-Fidelity 4D Hand-Object Capture via Multi-View Spatiotemporal Tracking and Physics-Aware Gaussians

    cs.CV 2026-06 unverdicted novelty 5.0 of 10

    A multi-view feed-forward transformer provides initial poses and geometry from calibrated videos, followed by physics-aware Gaussian optimization with tetrahedral and collision constraints to produce robust 4D hand-ob...

  19. Artiverse: A Diverse and Physically Grounded Dataset for Articulated Objects

    cs.CV 2026-05 unverdicted novelty 5.0 of 10

    Artiverse is a new dataset of 5.4K human-authored articulated 3D objects with detailed annotations for parts, multi-DoF joints, interior structures, and physical attributes to enable functional modeling and physics-ba...

  20. EVA01: Unified Native 3D Understanding and Generation via Mixture-of-Transformers

    cs.CV 2026-05 unverdicted novelty 5.0 of 10

    EVA01 introduces a Mixture-of-Transformers model that natively adds 3D mesh understanding, generation, and multi-turn editing to MLLMs by decoupling understanding and generation experts with shared global self-attention.

  21. SplatPainter: Interactive Authoring of 3D Gaussians from 2D Edits via Test-Time Training

    cs.CV 2025-12 conditional novelty 5.0 of 10

    A test-time-trained feedforward model that propagates 2D edits onto 3D Gaussian attributes at interactive speeds.

  22. Advances in Neural 3D Mesh Texturing: A Survey

    cs.CV 2026-05 unverdicted novelty 2.0 of 10

    A literature survey that organizes neural 3D mesh texturing methods into a taxonomy spanning early GAN-based approaches to modern diffusion pipelines, while reviewing architectures, datasets, evaluation, and open challenges.

Reference graph

Works this paper leans on

20 extracted references · 20 canonical work pages · cited by 21 Pith papers

  1. [1]

    https://sketchfab.com/

    Sketchfab. https://sketchfab.com/. 2

    work page

  2. [2]

    https://openai.com/gpt-5/, 2025

    Gpt-5. https://openai.com/gpt-5/, 2025. 2, 5

    work page 2025

  3. [3]

    Chang, Thomas A

    Angel X. Chang, Thomas A. Funkhouser, Leonidas J. Guibas, Pat Hanrahan, Qi-Xing Huang, Zimo Li, Silvio Savarese, Manolis Savva, Shuran Song, Hao Su, Jianxiong Xiao, Li Yi, and Fisher Yu. Shapenet: An information-rich 3d model repository. arXiv preprint, 2015. 3

    work page 2015

  4. [4]

    Ultra3d: Efficient and high-fidelity 3d generation with part attention

    Yiwen Chen, Zhihao Li, Yikai Wang, Hu Zhang, Qin Li, Chi Zhang, and Guosheng Lin. Ultra3d: Efficient and high-fidelity 3d generation with part attention. arXiv preprint, 2025. 2

    work page 2025

  5. [5]

    Mvpaint: Synchronized multi-view diffusion for painting anything 3d

    Wei Cheng, Juncheng Mu, Xianfang Zeng, Xin Chen, Anqi Pang, Chi Zhang, Zhibin Wang, Bin Fu, Gang Yu, Ziwei Liu, and Liang Pan. Mvpaint: Synchronized multi-view diffusion for painting anything 3d. In CVPR, 2025. 2

    work page 2025

  6. [6]

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

    Jasmine Collins, Shubham Goel, Kenan Deng, Achleshwar Luthra, Leon Xu, Erhan Gundogdu, Xi Zhang, Tomas F. Yago Vicente, Thomas Dideriksen, Himanshu Arora, Matthieu Guillaumin, and Jitendra Malik. ABO: dataset and benchmarks for real-world 3d object understanding. In CVPR, 2022. 3

    work page 2022

  7. [7]

    Objaverse-xl: A universe of 10m+ 3d objects

    Matt Deitke, Ruoshi Liu, Matthew Wallingford, Huong Ngo, Oscar Michel, Aditya Kusupati, Alan Fan, Christian Laforte, Vikram V oleti, Samir Yitzhak Gadre, Eli VanderBilt, Aniruddha Kembhavi, Carl V ondrick, Georgia Gkioxari, Kiana Ehsani, Ludwig Schmidt, and Ali Farhadi. Objaverse-xl: A universe of 10m+ 3d objects. In NeurIPS, 2023. 2, 3

    work page 2023

  8. [8]

    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 CVPR, 2023. 2, 3

    work page 2023

Show all 20 references
  1. [9]

    Newcombe

    Zhao Dong, Ka Chen, Zhaoyang Lv, Hong-Xing Yu, Yunzhi Zhang, Cheng Zhang, Yufeng Zhu, Stephen Tian, Zhengqin Li, Geordie Moffatt, Sean Christofferson, James Fort, Xiaqing Pan, Mingfei Yan, Jiajun Wu, Carl Yuheng Ren, and Richard A. Newcombe. Digital twin catalog: A large-scale...

    2025

  2. [10]

    Google scanned objects: A high- quality dataset of 3d scanned household items

    Laura Downs, Anthony Francis, Nate Koenig, Brandon Kinman, Ryan Hickman, Krista Rey- mann, Thomas Barlow McHugh, and Vincent Vanhoucke. Google scanned objects: A high- quality dataset of 3d scanned household items. In ICRA, 2022. 3

    2022

  3. [11]

    Maybank, and Dacheng Tao

    Huan Fu, Rongfei Jia, Lin Gao, Mingming Gong, Binqiang Zhao, Stephen J. Maybank, and Dacheng Tao. 3d-future: 3d furniture shape with texture. IJCV, 2021. 3

    2021

  4. [12]

    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, 2025. 2

    2025

  5. [13]

    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, Sheng Zhang, Xin Huang, Di Luo, Fan Yang, Fang Yang, Lifu Wang, Sicong Liu, Yixuan Tang, Yulin Cai, Zebin He, Tian Liu, Yuhong Liu, Jie Jiang, Linu...

    2025

  6. [14]

    Sparc3d: Sparse rep- resentation and construction for high-resolution 3d shapes modeling

    Zhihao Li, Yufei Wang, Heliang Zheng, Yihao Luo, and Bihan Wen. Sparc3d: Sparse rep- resentation and construction for high-resolution 3d shapes modeling. arXiv preprint, 2025. 2

    2025

  7. [15]

    Gt23d-bench: A comprehensive general text-to-3d generation benchmark

    Sitong Su, Xiao Cai, Lianli Gao, Pengpeng Zeng, Qinhong Du, Mengqi Li, Heng Tao Shen, and Jingkuan Song. Gt23d-bench: A comprehensive general text-to-3d generation benchmark. arXiv preprint, 2024. 5

    2024

  8. [16]

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

    Kailu Wu, Fangfu Liu, Zhihan Cai, Runjie Yan, Hanyang Wang, Yating Hu, Yueqi Duan, and Kaisheng Ma. Unique3d: High-quality and efficient 3d mesh generation from a single image. In NeurIPS, 2024. 2 7

    2024

  9. [17]

    Direct3d-s2: Gigascale 3d generation made easy with spatial sparse attention

    Shuang Wu, Youtian Lin, Feihu Zhang, Yifei Zeng, Yikang Yang, Yajie Bao, Jiachen Qian, Siyu Zhu, Xun Cao, Philip Torr, and Yao Yao. Direct3d-s2: Gigascale 3d generation made easy with spatial sparse attention. arXiv preprint, 2025. 2

    2025

  10. [18]

    Omniobject3d: Large-vocabulary 3d object dataset for realistic perception, reconstruction and generation

    Tong Wu, Jiarui Zhang, Xiao Fu, Yuxin Wang, Jiawei Ren, Liang Pan, Wayne Wu, Lei Yang, Jiaqi Wang, Chen Qian, Dahua Lin, and Ziwei Liu. Omniobject3d: Large-vocabulary 3d object dataset for realistic perception, reconstruction and generation. In CVPR, 2023. 3

    2023

  11. [19]

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

    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 TOG, 2024. 2

    2024

  12. [20]

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

    Zibo Zhao, Zeqiang Lai, Qingxiang Lin, Yunfei Zhao, Haolin Liu, Shuhui Yang, Yifei Feng, Mingxin Yang, Sheng Zhang, Xianghui Yang, Huiwen Shi, Sicong Liu, Junta Wu, Yihang Lian, Fan Yang, Ruining Tang, Zebin He, Xinzhou Wang, Jian Liu, Xuhui Zuo, Zhuo Chen, Biwen Lei, Haohan W...

    2025

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.