Pith. sign in

REVIEW 2 major objections 5 minor 75 references

Muses composes fantasy 3D creatures by fusing skeletons of existing animals, without any training.

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 ↗

T0 review · deepseek-v4-flash

2026-08-03 12:19 UTC pith:YL2YA3HO

load-bearing objection A genuinely new skeleton-first recipe for composing fantasy 3D creatures without training, but the load-bearing skeleton classification heuristic is unvalidated and the evaluation is too thin for the SOTA claim. the 2 major comments →

arxiv 2601.03256 v2 pith:YL2YA3HO submitted 2026-01-06 cs.CV

Muses: Designing, Composing, Generating Nonexistent Fantasy 3D Creatures without Training

Hexiao Lu , Xiaokun Sun , Zeyu Cai , Hao Guo , Ying Tai , Jian Yang , Zhenyu Zhang This is my paper
classification cs.CV
keywords 3D generationfantasy creaturesskeletontraining-freetext-to-3Dstructured latentpart compositionappearance editing
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.

Muses tackles the open problem of generating entirely new 3D creatures that mix parts of different animals (a tiger body, dragon wings, robot legs, fox tails). It claims the first training-free, feed-forward method that does this reliably: instead of optimizing or stitching surfaces, it designs an explicit 3D skeleton, composes source assets in a sparse voxel latent space guided by that skeleton, and then re-textures the result. If correct, a text prompt is enough to get a coherent, textured, style-consistent 3D asset in under a minute. The paper positions the 3D skeleton as the key bridge: a biological form's backbone makes part-level composition geometrically rational instead of arbitrary.

Core claim

On the paper's own terms, Muses establishes that the 3D skeleton—not 2D images, part affinities, or manual assembly—is the right intermediate representation for creative 3D creature generation. Given a compositional prompt, it first generates separate 3D assets for each mentioned concept, auto-rigs skeletons for them, and classifies each skeleton into body, legs, wings, tail, and head using graph heuristics. A large language model then decides how to translate, rotate, and scale these sub-skeletons into a coherent new layout. The composed skeleton guides a voxel-based interpolation of the source assets' structured latents, so geometry and texture are blended at the feature level rather than

What carries the argument

The central object is the 3D skeleton, a graph of joints and bones that captures the articulated structure of a creature. The paper treats it as a universal 'plug interface' between design and geometry: once sub-skeletons are classified (body, legs, wings, tail, head) and reassembled by an LLM-guided layout reasoner, the skeleton provides an explicit correspondence that maps each semantic region onto the sparse voxels of a structured latent code (SLAT), a representation where each active voxel carries a feature vector. Skinning weights from auto-rigging transfer these regions into the latent space, where voxel interpolation fuses geometry and texture. This is what carries the argument: the s

Load-bearing premise

The whole pipeline depends on the hand-written grammar that classifies every source skeleton into body/leg/wing/tail/head; if a source asset's skeleton is noisy or does not fit the articulated-animal template, the design and composition stages fail, as the paper's own failure cases (a peacock, a failed auto-rig) show.

What would settle it

Take a diverse set of, say, 50 generated 3D assets spanning articulated and non-articulated forms, auto-rig them, run the graph classifier, and compare the predicted body/leg/wing/tail/head labels against manual annotation. If the classifier mislabels a substantial fraction—or if a single mislabel cascades into a broken final creature—the central claim of a reliable training-free pipeline is undermined. The paper already shows two failure cases; a larger-scale accuracy measurement would settle how general the method is.

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

If this is right

  • A user can generate a new, coherent 3D creature from a single text prompt—like 'tiger body, dragon wings, robot legs'—in under a minute on one GPU, without any per-prompt optimization or training.
  • The same skeleton-driven pipeline supports disentangled geometry editing: adding a head or tail modifies only the intended region, leaving the rest of the creature unchanged.
  • Because composition happens in a structured latent space rather than by cutting meshes, the result avoids the seams, voids, and misalignments that plague part-level assembly.
  • Style-consistent appearance modeling lets the user re-skin an existing creature in different artistic styles (e.g., steampunk, porcelain) while preserving its geometry.
  • The framework is backbone-agnostic: it adapts to any modern structured-latent 3D generator, not just the one used in the experiments.

Where Pith is reading between the lines

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

  • If the skeleton-first approach is right, it suggests that the next generation of 3D generative models should expose their internal latent structure to user control; skeletons are a natural, topology-aware interface for that.
  • The method's dependence on auto-rigging and hand-written graph heuristics implies a cheap improvement path: learn the skeleton classifier from data rather than rules, which would extend Muses to non-articulated or exotic sources (mollusks, plants, machines) that do not fit the body/leg/wing/tail/head template.
  • The image-guided texture stage opens a two-way bridge: a user could take a hand-drawn or AI-generated 2D concept image and transfer its style onto the composed geometry, enabling 2D-to-3D creative workflows without retraining.
  • Because the pipeline is training-free and modular, it is a natural testbed for measuring how much of text-to-3D composition is a geometry problem versus a texture problem; swapping different latent backbones would directly quantify that split.

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

2 major / 5 minor

Summary. Muses is a training-free, feed-forward pipeline for generating 'fantastic 3D creatures' from compositional text prompts (e.g., tiger body + dragon wings + robot legs). The method first instantiates each mentioned concept with Trellis text-to-3D and auto-rigs the resulting meshes with Puppeteer. A graph heuristic in Sec. 4.1 classifies each skeleton into body/legs/wings/head/tail using the pelvis root, a trunk-junction node with degree >=4, symmetry, and height thresholds; an LLM then composes the extracted sub-skeletons into a new skeleton via Rotate/Translate/Scale operators. Sec. 4.2 maps the composed skeleton onto the SLAT latent representation through skinning weights and nearest-vertex inverse-distance weights, interpolates overlapping voxels in a coarse 16^3 grid, and decodes a coarse creature. Sec. 4.3 renders a reference image, edits it with FLUX.1 Kontext, and re-encodes it under the coarse geometry to produce a stylistically consistent final texture. The paper reports CLIP/VQAScore on 30 samples and a 10-example, 60-participant user study, claiming state-of-the-art visual fidelity and text alignment, and demonstrates part-level and texture editing applications.

Significance. If the claims hold, Muses is a practically useful contribution: it avoids per-instance SDS optimization and manual part stitching, and it introduces an explicit skeleton as a controllable intermediate representation for 3D asset composition. The idea of using skinning weights to bridge skeletons and SLAT latents is clean and is supported by the ablations, which show each component improves quantitative scores. The method is also genuinely training-free in the sense that no new model weights are learned, and the qualitative results on several hard prompts are compelling. The main cautions are that (i) the skeleton-decomposition heuristic is not validated despite being the backbone of the design stage, and (ii) the evaluation is too small and too poorly specified to support the strong state-of-the-art claim. The first is a correctness risk for a broad class of prompts; the second is an evidence risk for the headline claim. The pipeline is assembled from external tools and the metrics are independent of the construction, so I find no circularity in the main derivation.

major comments (2)
  1. [§4.1, Eqs. (1)–(2)] The semantic skeleton decomposition is the linchpin of the pipeline, but it is unvalidated and brittle. Eq. (2) defines d as the argmin over all nodes with deg(v)>=4 of ⟨v, δ̂⟩, whereas the prose says the search starts at b and proceeds along δ; the equation imposes no path constraint, so the 'trunk junction' can be selected on the wrong side of the body. The rule set assumes an articulated animal-like tree; the octopus, jellyfish, and snail examples in Fig. 5 do not fit this template, and the degenerate assignment G_body = G̃ covers only fish-like shapes. Sec. 5.5 concedes that Trellis and Puppeteer failures block the design stage. No per-part accuracy or failure rates are reported. Since every later stage (e.g., Sec. 4.2, Eq. (4)) assumes the partition is correct, a misclassified wing/leg/head propagates into the composed skeleton and the final asset. Please add per-part precision/reca
  2. [Table 1 and §5.1] The state-of-the-art claim rests on Table 1, but the evaluation in §5.1 is too thin. The 'Visual fidelity' and 'Text alignment' numbers come from a 10-example user study with 60 volunteers; no confidence intervals, inter-rater agreement, or significance tests are reported. The gap 66.67 vs 12.62 (visual fidelity vs OmniPart) is impossible to assess without knowing whether these are preference win rates, mean ratings, and how the 10 items were aggregated. Automatic metrics use only 30 samples with no standard deviations. The OmniPart baseline is manually assembled, conflating method quality with manual effort, and the claimed <1-minute runtime is not compared with any baseline. Please specify the protocol, add variance/statistics, increase the number of examples, publish the prompt set and per-example scores, and match baseline conditions.
minor comments (5)
  1. [§4.2] Superscript formatting is ambiguous in the printed text: '64^3 SLAT space' and '16^3 voxel space' appear as '643' and '163'. Please use proper superscripts.
  2. [Table 1 / §5.2] Method naming is inconsistent: Table 1 lists 'GaussianDreamer' while the text refers to 'DreamGaussian [49]'. Similarly, Fig. 5 uses 'UNO+Trellis' while Table 1 writes 'UNO [56] + Trellis [58]'.
  3. [§5.1] The Appendix referenced for user-study details is not present in the submitted version. Please include it, as the exact rating format is needed to interpret the headline numbers.
  4. [§4.1, Eq. (3)] The exact output format of the LLM operator sequence is not shown. One worked example of the Rot/Trans/Scale sequence for a full prompt would significantly improve reproducibility.
  5. [Table 2] VQA1 and VQA2 are defined only in the table caption ('VQA-Score based on CLIP-FlanT5' and 'ShareGPT4V'), not in the main text. Please introduce both metrics in §5.1.

Circularity Check

0 steps flagged

No significant circularity: the pipeline is assembled from external, independently pretrained components and no predicted quantity is defined in terms of its own inputs.

full rationale

I walked the derivation chain and found no step where a claimed prediction is equivalent, by construction or by definition, to an input or fitted parameter. The skeleton classification in Sec. 4.1 (Eqs. 1–2) is a hand-written heuristic operating on skeletons produced by the external Puppeteer tool; it is not a fitted function of the final evaluation metrics, and no output quantity is defined as the argmin/argmax that produced it. The SLAT composition in Sec. 4.2 (Eqs. 4–6) is a weighted interpolation/averaging of latent vectors and skinning weights from external assets; it contains no learned parameters and no prediction target. The texture stage in Sec. 4.3 (Eqs. 7–8) uses FLUX.1 Kontext and the second-stage Trellis decoder on the composed geometry and an edited reference image; the final asset is therefore generated from its inputs rather than secretly being one of its inputs renamed. Evaluation metrics (CLIP, VQAScore, and the user study) are external to the method's construction and are not used to fit any component. I also checked the self-citation patterns: the load-bearing external tools (Trellis, Puppeteer, Qwen-plus, FLUX.1 Kontext) are cited as independent works and no uniqueness theorem or load-bearing claim is reduced to a self-citation by the present authors. The acknowledged limitations in Sec. 5.5 are failures of upstream external generators and skeletonization, which is a robustness/correctness concern, not circularity. The 'first training-free feed-forward' claim is a comparative positioning statement, not a derived result. In short, the paper may have correctness or generalization risks, but its derivation is self-contained and non-circular.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

Muses introduces no new physical or representational entities; it uses existing skeletons, SLAT latents, and pretrained generators. The main unverified assumptions are that skeleton classification, skinning transfer, linear latent interpolation, and image-guided restyling all behave reliably across out-of-domain inputs. These are domain assumptions, not fabricated entities.

free parameters (4)
  • k (nearest mesh vertices in Eq. 5)
    Hand-chosen neighborhood size for transferring skinning weights to SLAT voxels; no sensitivity analysis is provided.
  • Trunk-junction degree threshold deg(v)>=4 (Eq. 2) = deg >= 4
    Hand-written rule for locating the body/head/wing junction; changes the skeleton decomposition and downstream composition.
  • Root-selection threshold deg(r)>=3 (Eq. 1) = deg >= 3
    Hand-written heuristic for choosing the anatomical root; affects which branches are classified as legs vs tail.
  • Trellis inference settings = CFG 5.0, 25 steps
    Classifier-free guidance scale and sampling steps used for all asset generation and refinement; standard hyperparameters, not fitted to the target result.
axioms (5)
  • domain assumption Linear interpolation of SLAT voxels from different objects yields geometrically coherent shapes with smooth seams.
    The core of Sec. 4.2: the composed latent z' is formed by weighted interpolation of separate object latents in 16^3 space; if SLAT is not semantically aligned across objects, the fused geometry can be incoherent.
  • domain assumption Puppeteer auto-rigging produces anatomically reliable skeletons and skinning weights for arbitrary generated 3D assets.
    Sec. 4.1 relies on Puppeteer for skeleton extraction and Sec. 4.2 for region mapping; the paper's Sec. 5.5 shows Puppeteer failures that block the pipeline.
  • domain assumption Qwen-plus LLM can reliably infer spatial layout, scale, and orientation from natural-language assembly requests.
    The LLM-guided assembly in Sec. 4.1 turns free-text prompts into primitive Rot/Trans/Scale operations; no systematic evaluation of LLM failure rate is reported.
  • domain assumption FLUX Kontext image editing preserves 3D geometry while restyling texture.
    Sec. 4.3 uses the rendered coarse creature as conditioning image and relies on FLUX Kontext to output a style-edited image that is geometrically aligned for the second SLAT pass.
  • domain assumption Trellis generates source assets of sufficient quality and diversity for the concept decomposition.
    The entire pipeline starts from Trellis-generated 3D assets; the paper's Sec. 5.5 admits Trellis failures (e.g., peacock) that make skeleton extraction impossible.

pith-pipeline@v1.3.0-alltime-deepseek · 14679 in / 10841 out tokens · 102423 ms · 2026-08-03T12:19:04.527810+00:00 · methodology

0 comments
read the original abstract

We present Muses, the first training-free method for fantastic 3D creature generation in a feed-forward paradigm. Previous methods, which rely on part-aware optimization, manual assembly, or 2D image generation, often produce unrealistic or incoherent 3D assets due to the challenges of intricate part-level manipulation and limited out-of-domain generation. In contrast, Muses leverages the 3D skeleton, a fundamental representation of biological forms, to explicitly and rationally compose diverse elements. This skeletal foundation formalizes 3D content creation as a structure-aware pipeline of design, composition, and generation. Muses begins by constructing a creatively composed 3D skeleton with coherent layout and scale through graph-constrained reasoning. This skeleton then guides a voxel-based assembly process within a structured latent space, integrating regions from different objects. Finally, image-guided appearance modeling under skeletal conditions is applied to generate a style-consistent and harmonious texture for the assembled shape. Extensive experiments establish Muses' state-of-the-art performance in terms of visual fidelity and alignment with textual descriptions, and potential on flexible 3D object editing. Project page: https://luhexiao.github.io/Muses.github.io/.

Figures

Figures reproduced from arXiv: 2601.03256 by Hao Guo, Hexiao Lu, Jian Yang, Xiaokun Sun, Ying Tai, Zeyu Cai, Zhenyu Zhang.

Figure 1
Figure 1. Figure 1: Generated nonexistent fantastic 3D creatures including animals, humanoids, and fictional characters by Muses. Driven by a 3D [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Compared with (a) methods that distill part-level affinity [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of Muses. Our framework automates fantastic creature generation through a 3D skeleton-driven pipeline of design, [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: , given 3D assets {X}M m=1 and their associated skele￾tons {G = (V, E)}M m=1 where M is the number of crea￾ [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparison with the state-of-the-art methods. Note that DreamBeast [ [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Ablation study of Muses. We observe that each component of our method contributes to performance. The assets without [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Applications of geometry and texture editing. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Failure cases. 5.5. Limitation The limitations of our method mainly come from two as￾pects: failed generations of Trellis and improper initializa￾tion of the 3D skeleton. As illustrated in [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

75 extracted references · 28 linked inside Pith

  1. [1]

    A morphable model for the synthesis of 3d faces

    V olker Blanz and Thomas Vetter. A morphable model for the synthesis of 3d faces. InSeminal Graphics Papers: Pushing the Boundaries, Volume 2, pages 157–164. 1999. 2

    1999

  2. [2]

    Partgen: Part-level 3d generation and reconstruction with multi-view diffusion models

    Minghao Chen, Roman Shapovalov, Iro Laina, Tom Mon- nier, Jianyuan Wang, David Novotny, and Andrea Vedaldi. Partgen: Part-level 3d generation and reconstruction with multi-view diffusion models. InProceedings of the Com- puter Vision and Pattern Recognition Conference, pages 5881–5892, 2025. 2, 3

    2025

  3. [3]

    Autopartgen: Autogres- sive 3d part generation and discovery.arXiv preprint arXiv:2507.13346, 2025

    Minghao Chen, Jianyuan Wang, Roman Shapovalov, Tom Monnier, Hyunyoung Jung, Dilin Wang, Rakesh Ranjan, Iro Laina, and Andrea Vedaldi. Autopartgen: Autogres- sive 3d part generation and discovery.arXiv preprint arXiv:2507.13346, 2025. 2, 3

    Pith/arXiv arXiv 2025

  4. [4]

    Ultra3d: Efficient and high- fidelity 3d generation with part attention.arXiv preprint arXiv:2507.17745, 2025

    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 arXiv:2507.17745, 2025. 3

    Pith/arXiv arXiv 2025

  5. [5]

    Meshcoder: Llm-powered structured mesh code generation from point clouds.arXiv preprint arXiv:2508.14879, 2025

    Bingquan Dai, Li Ray Luo, Qihong Tang, Jie Wang, Xinyu Lian, Hao Xu, Minghan Qin, Xudong Xu, Bo Dai, Hao- qian Wang, et al. Meshcoder: Llm-powered structured mesh code generation from point clouds.arXiv preprint arXiv:2508.14879, 2025. 3

    Pith/arXiv arXiv 2025

  6. [6]

    Objaverse-xl: A universe of 10m+ 3d objects.Advances in Neural Informa- tion Processing Systems, 36:35799–35813, 2023

    Matt Deitke, Ruoshi Liu, Matthew Wallingford, Huong Ngo, Oscar Michel, Aditya Kusupati, Alan Fan, Christian Laforte, Vikram V oleti, Samir Yitzhak Gadre, et al. Objaverse-xl: A universe of 10m+ 3d objects.Advances in Neural Informa- tion Processing Systems, 36:35799–35813, 2023. 2, 3

    2023

  7. [7]

    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. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 13142–13153, 2023. 2, 3

    2023

  8. [8]

    From one to more: Contex- tual part latents for 3d generation

    Shaocong Dong, Lihe Ding, Xiao Chen, Yaokun Li, Yuxin Wang, Yucheng Wang, Qi Wang, Jaehyeok Kim, Chenjian Gao, Zhanpeng Huang, et al. From one to more: Contex- tual part latents for 3d generation. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 8230–8240, 2025. 3

    2025

  9. [9]

    Ip-composer: Semantic composition of visual concepts

    Sara Dorfman, Dana Cohen-Bar, Rinon Gal, and Daniel Cohen-Or. Ip-composer: Semantic composition of visual concepts. InProceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers, pages 1–11, 2025. 3

    2025

  10. [10]

    Distribution-conditional generation: From class distribution to creative generation.arXiv preprint arXiv:2505.03667,

    Fu Feng, Yucheng Xie, Xu Yang, Jing Wang, and Xin Geng. Distribution-conditional generation: From class distribution to creative generation.arXiv preprint arXiv:2505.03667,

    Pith/arXiv arXiv

  11. [11]

    Redefining¡ creative¿ in dictionary: Towards an enhanced se- mantic understanding of creative generation

    Fu Feng, Yucheng Xie, Xu Yang, Jing Wang, and Xin Geng. Redefining¡ creative¿ in dictionary: Towards an enhanced se- mantic understanding of creative generation. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 18444–18454, 2025. 2, 3

    2025

  12. [12]

    Seed3d 1.0: From images to high-fidelity simulation- ready 3d assets.arXiv preprint arXiv:2510.19944, 2025

    Jiashi Feng, Xiu Li, Jing Lin, Jiahang Liu, Gaohong Liu, Weiqiang Lou, Su Ma, Guang Shi, Qinlong Wang, Jun Wang, et al. Seed3d 1.0: From images to high-fidelity simulation- ready 3d assets.arXiv preprint arXiv:2510.19944, 2025. 2

    arXiv 2025

  13. [13]

    Multi-view stereo: A tutorial.Foundations and trends® in Computer Graphics and Vision, 9(1-2):1–148, 2015

    Yasutaka Furukawa, Carlos Hern ´andez, et al. Multi-view stereo: A tutorial.Foundations and trends® in Computer Graphics and Vision, 9(1-2):1–148, 2015. 2

    2015

  14. [14]

    An image is worth one word: Personalizing text-to- image generation using textual inversion.arXiv preprint arXiv:2208.01618, 2022

    Rinon Gal, Yuval Alaluf, Yuval Atzmon, Or Patash- nik, Amit H Bermano, Gal Chechik, and Daniel Cohen- Or. An image is worth one word: Personalizing text-to- image generation using textual inversion.arXiv preprint arXiv:2208.01618, 2022. 3

    Pith/arXiv arXiv 2022

  15. [15]

    Tokenverse: Versatile multi-concept personalization in token modulation space.ACM Transactions On Graphics (TOG), 44(4):1–11, 2025

    Daniel Garibi, Shahar Yadin, Roni Paiss, Omer Tov, Shiran Zada, Ariel Ephrat, Tomer Michaeli, Inbar Mosseri, and Tali Dekel. Tokenverse: Versatile multi-concept personalization in token modulation space.ACM Transactions On Graphics (TOG), 44(4):1–11, 2025. 3

    2025

  16. [16]

    Mv-adapter: Multi-view consistent image generation made easy

    Zehuan Huang, Yuan-Chen Guo, Haoran Wang, Ran Yi, Lizhuang Ma, Yan-Pei Cao, and Lu Sheng. Mv-adapter: Multi-view consistent image generation made easy. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 16377–16387, 2025. 2

    2025

  17. [17]

    Hunyuan3d 2.1: From images to high- fidelity 3d assets with production-ready pbr material.arXiv preprint arXiv:2506.15442, 2025

    Team Hunyuan3D, Shuhui Yang, Mingxin Yang, Yifei Feng, Xin Huang, Sheng Zhang, Zebin He, Di Luo, Haolin Liu, Yunfei Zhao, et al. Hunyuan3d 2.1: From images to high- fidelity 3d assets with production-ready pbr material.arXiv preprint arXiv:2506.15442, 2025. 3

    Pith/arXiv arXiv 2025

  18. [18]

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

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk ¨uhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering.ACM Trans. Graph., 42(4):139–1,

  19. [19]

    Black Forest Labs, Stephen Batifol, Andreas Blattmann, Frederic Boesel, Saksham Consul, Cyril Diagne, Tim Dock- horn, Jack English, Zion English, Patrick Esser, et al. Flux. 1 kontext: Flow matching for in-context image generation and editing in latent space.arXiv preprint arXiv:2506.15742,

    Pith/arXiv arXiv

  20. [20]

    Hunyuan3d 2.5: Towards high- fidelity 3d assets generation with ultimate details.arXiv preprint arXiv:2506.16504, 2025

    Zeqiang Lai, Yunfei Zhao, Haolin Liu, Zibo Zhao, Qingxi- ang Lin, Huiwen Shi, Xianghui Yang, Mingxin Yang, Shuhui Yang, Yifei Feng, et al. Hunyuan3d 2.5: Towards high- fidelity 3d assets generation with ultimate details.arXiv preprint arXiv:2506.16504, 2025. 2, 3

    Pith/arXiv arXiv 2025

  21. [21]

    Instant3d: Fast text-to-3d with sparse-view generation and large reconstruction model

    Jiahao Li, Hao Tan, Kai Zhang, Zexiang Xu, Fujun Luan, Yinghao Xu, Yicong Hong, Kalyan Sunkavalli, Greg Shakhnarovich, and Sai Bi. Instant3d: Fast text-to-3d with sparse-view generation and large reconstruction model. arXiv preprint arXiv:2311.06214, 2023. 3

    Pith/arXiv arXiv 2023

  22. [22]

    Dreambeast: Distilling 3d fantastical animals with part-aware knowledge transfer

    Runjia Li, Junlin Han, Luke Melas-Kyriazi, Chunyi Sun, Zhaochong An, Zhongrui Gui, Shuyang Sun, Philip Torr, and Tomas Jakab. Dreambeast: Distilling 3d fantastical animals with part-aware knowledge transfer. In2025 International Conference on 3D Vision (3DV), pages 1243–1252. IEEE,

  23. [23]

    Crafts- 9 man3d: High-fidelity mesh generation with 3d native gen- eration and interactive geometry refiner.arXiv preprint arXiv:2405.14979, 2024

    Weiyu Li, Jiarui Liu, Hongyu Yan, Rui Chen, Yixun Liang, Xuelin Chen, Ping Tan, and Xiaoxiao Long. Crafts- 9 man3d: High-fidelity mesh generation with 3d native gen- eration and interactive geometry refiner.arXiv preprint arXiv:2405.14979, 2024. 3

    Pith/arXiv arXiv 2024

  24. [24]

    Connecting consistency distillation to score distillation for text-to-3d generation

    Zongrui Li, Minghui Hu, Qian Zheng, and Xudong Jiang. Connecting consistency distillation to score distillation for text-to-3d generation. InEuropean Conference on Computer Vision, pages 274–291. Springer, 2024. 3

    2024

  25. [25]

    Sparc3d: Sparse representation and construc- tion for high-resolution 3d shapes modeling.arXiv preprint arXiv:2505.14521, 2025

    Zhihao Li, Yufei Wang, Heliang Zheng, Yihao Luo, and Bihan Wen. Sparc3d: Sparse representation and construc- tion for high-resolution 3d shapes modeling.arXiv preprint arXiv:2505.14521, 2025. 3

    Pith/arXiv arXiv 2025

  26. [26]

    Luciddreamer: Towards high- fidelity text-to-3d generation via interval score matching

    Yixun Liang, Xin Yang, Jiantao Lin, Haodong Li, Xiao- gang Xu, and Yingcong Chen. Luciddreamer: Towards high- fidelity text-to-3d generation via interval score matching. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 6517–6526, 2024. 3

    2024

  27. [27]

    Partcrafter: Structured 3d mesh generation via compo- sitional latent diffusion transformers.arXiv preprint arXiv:2506.05573, 2025

    Yuchen Lin, Chenguo Lin, Panwang Pan, Honglei Yan, Yiqiang Feng, Yadong Mu, and Katerina Fragkiadaki. Partcrafter: Structured 3d mesh generation via compo- sitional latent diffusion transformers.arXiv preprint arXiv:2506.05573, 2025. 3

    Pith/arXiv arXiv 2025

  28. [28]

    Evaluating text-to-visual generation with image-to-text gen- eration

    Zhiqiu Lin, Deepak Pathak, Baiqi Li, Jiayao Li, Xide Xia, Graham Neubig, Pengchuan Zhang, and Deva Ramanan. Evaluating text-to-visual generation with image-to-text gen- eration. InEuropean Conference on Computer Vision, pages 366–384. Springer, 2024. 7

    2024

  29. [29]

    Dreamreward-x: Boosting high-quality 3d generation with human preference alignment.IEEE Transactions on Pattern Analysis and Ma- chine Intelligence, 2025

    Fangfu Liu, Junliang Ye, Yikai Wang, Hanyang Wang, Zhengyi Wang, Jun Zhu, and Yueqi Duan. Dreamreward-x: Boosting high-quality 3d generation with human preference alignment.IEEE Transactions on Pattern Analysis and Ma- chine Intelligence, 2025. 2

    2025

  30. [30]

    Syncdreamer: Gen- erating multiview-consistent images from a single-view im- age.arXiv preprint arXiv:2309.03453, 2023

    Yuan Liu, Cheng Lin, Zijiao Zeng, Xiaoxiao Long, Lingjie Liu, Taku Komura, and Wenping Wang. Syncdreamer: Gen- erating multiview-consistent images from a single-view im- age.arXiv preprint arXiv:2309.03453, 2023. 3

    Pith/arXiv arXiv 2023

  31. [31]

    Wonder3d: Sin- gle 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, et al. Wonder3d: Sin- gle image to 3d using cross-domain diffusion. InProceed- ings of the IEEE/CVF conference on computer vision and pattern recognition, pages 9970–9980, 2024. 3

    2024

  32. [32]

    Shading meets motion: Self-supervised indoor 3d reconstruction via simultaneous shape-from-shading and structure-from-motion

    Guoyu Lu. Shading meets motion: Self-supervised indoor 3d reconstruction via simultaneous shape-from-shading and structure-from-motion. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 16508– 16519, 2025. 2

    2025

  33. [33]

    Data synthesis with diverse styles for face recognition via 3dmm-guided diffusion

    Yuxi Mi, Zhizhou Zhong, Yuge Huang, Qiuyang Yuan, Xuan Zhao, Jianqing Xu, Shouhong Ding, Shaoming Wang, Rizen Guo, and Shuigeng Zhou. Data synthesis with diverse styles for face recognition via 3dmm-guided diffusion. InProceed- ings of the Computer Vision and Pattern Recognition Con- ference, pages 21203–21214, 2025. 2

    2025

  34. [34]

    Nerf: Representing scenes as neural radiance fields for view syn- thesis.Communications of the ACM, 65(1):99–106, 2021

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view syn- thesis.Communications of the ACM, 65(1):99–106, 2021. 2

    2021

  35. [35]

    Partcraft: Crafting creative objects by parts

    Kam Woh Ng, Xiatian Zhu, Yi-Zhe Song, and Tao Xiang. Partcraft: Crafting creative objects by parts. InEuropean Conference on Computer Vision, pages 420–437. Springer,

  36. [36]

    Object-level visual prompts for compositional image generation.arXiv preprint arXiv:2501.01424, 2025

    Gaurav Parmar, Or Patashnik, Kuan-Chieh Wang, Daniil Os- tashev, Srinivasa Narasimhan, Jun-Yan Zhu, Daniel Cohen- Or, and Kfir Aberman. Object-level visual prompts for compositional image generation.arXiv preprint arXiv:2501.01424, 2025. 3

    Pith/arXiv arXiv 2025

  37. [37]

    Nested attention: Semantic-aware attention values for concept personalization

    Or Patashnik, Rinon Gal, Daniil Ostashev, Sergey Tulyakov, Kfir Aberman, and Daniel Cohen-Or. Nested attention: Semantic-aware attention values for concept personalization. InProceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers. Association for Computing Machinery, 2025. 3

    2025

  38. [38]

    Dreamfusion: Text-to-3d using 2d diffusion.arXiv preprint arXiv:2209.14988, 2022

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

    Pith/arXiv arXiv 2022

  39. [39]

    Apply hierarchical- chain-of-generation to complex attributes text-to-3d gener- ation

    Yiming Qin, Zhu Xu, and Yang Liu. Apply hierarchical- chain-of-generation to complex attributes text-to-3d gener- ation. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 18521–18530, 2025. 2, 3

    2025

  40. [40]

    Learning transferable visual models from natural language supervi- sion

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervi- sion. InInternational conference on machine learning, pages 8748–8763. PmLR, 2021. 7

    2021

  41. [41]

    pops: Photo-inspired diffusion operators

    Elad Richardson, Yuval Alaluf, Ali Mahdavi-Amiri, and Daniel Cohen-Or. pops: Photo-inspired diffusion operators. InProceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers, pages 1–12, 2025. 3

    2025

  42. [42]

    Piece it together: Part-based concepting with ip- priors.arXiv preprint arXiv:2503.10365, 2025

    Elad Richardson, Kfir Goldberg, Yuval Alaluf, and Daniel Cohen-Or. Piece it together: Part-based concepting with ip- priors.arXiv preprint arXiv:2503.10365, 2025. 2, 3

    Pith/arXiv arXiv 2025

  43. [43]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 2

    2022

  44. [44]

    Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation

    Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 22500– 22510, 2023. 3

    2023

  45. [45]

    In- stantbooth: Personalized text-to-image generation without test-time finetuning

    Jing Shi, Wei Xiong, Zhe Lin, and Hyun Joon Jung. In- stantbooth: Personalized text-to-image generation without test-time finetuning. InProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pages 8543–8552, 2024. 3

    2024

  46. [46]

    Mvdream: Multi-view diffusion for 3d gen- eration.arXiv preprint arXiv:2308.16512, 2023

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

    Pith/arXiv arXiv 2023

  47. [47]

    Chimera: 10 Compositional image generation using part-based concept- ing.arXiv preprint arXiv:2510.18083, 2025

    Shivam Singh, Yiming Chen, Agneet Chatterjee, Amit Raj, James Hays, Yezhou Yang, and Chitra Baral. Chimera: 10 Compositional image generation using part-based concept- ing.arXiv preprint arXiv:2510.18083, 2025. 2, 3

    arXiv 2025

  48. [48]

    Puppeteer: Rig and animate your 3d models.arXiv preprint arXiv:2508.10898, 2025

    Chaoyue Song, Xiu Li, Fan Yang, Zhongcong Xu, Jiacheng Wei, Fayao Liu, Jiashi Feng, Guosheng Lin, and Jianfeng Zhang. Puppeteer: Rig and animate your 3d models.arXiv preprint arXiv:2508.10898, 2025. 6

    Pith/arXiv arXiv 2025

  49. [49]

    Dreamgaussian: Generative gaussian splatting for effi- cient 3d content creation

    Jiaxiang Tang, Jiawei Ren, Hang Zhou, Ziwei Liu, and Gang Zeng. Dreamgaussian: Generative gaussian splatting for effi- cient 3d content creation. InThe Twelfth International Con- ference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024, 2024. 6, 7

    2024

  50. [50]

    Efficient part-level 3d object generation via dual volume packing.arXiv preprint arXiv:2506.09980,

    Jiaxiang Tang, Ruijie Lu, Zhaoshuo Li, Zekun Hao, Xuan Li, Fangyin Wei, Shuran Song, Gang Zeng, Ming-Yu Liu, and Tsung-Yi Lin. Efficient part-level 3d object generation via dual volume packing.arXiv preprint arXiv:2506.09980,

    Pith/arXiv arXiv

  51. [51]

    Score jacobian chaining: Lifting pretrained 2d diffusion models for 3d generation

    Haochen Wang, Xiaodan Du, Jiahao Li, Raymond A Yeh, and Greg Shakhnarovich. Score jacobian chaining: Lifting pretrained 2d diffusion models for 3d generation. InPro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12619–12629, 2023. 3

    2023

  52. [52]

    Neus: Learning neural implicit surfaces by volume rendering for multi-view reconstruction

    Peng Wang, Lingjie Liu, Yuan Liu, Christian Theobalt, Taku Komura, and Wenping Wang. Neus: Learning neural implicit surfaces by volume rendering for multi-view reconstruction. arXiv preprint arXiv:2106.10689, 2021. 2

    Pith/arXiv arXiv 2021

  53. [53]

    Prolificdreamer: High-fidelity and diverse text-to-3d generation with variational score distilla- tion.Advances in neural information processing systems, 36: 8406–8441, 2023

    Zhengyi Wang, Cheng Lu, Yikai Wang, Fan Bao, Chongxuan Li, Hang Su, and Jun Zhu. Prolificdreamer: High-fidelity and diverse text-to-3d generation with variational score distilla- tion.Advances in neural information processing systems, 36: 8406–8441, 2023. 3

    2023

  54. [54]

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

    Hao Wen, Zehuan Huang, Yaohui Wang, Xinyuan Chen, and Lu Sheng. Ouroboros3d: Image-to-3d generation via 3d- aware recursive diffusion. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 21631– 21641, 2025. 2

    2025

  55. [55]

    Direct3d: Scal- able image-to-3d generation via 3d latent diffusion trans- former.Advances in Neural Information Processing Systems, 37:121859–121881, 2024

    Shuang Wu, Youtian Lin, Feihu Zhang, Yifei Zeng, Jingxi Xu, Philip Torr, Xun Cao, and Yao Yao. Direct3d: Scal- able image-to-3d generation via 3d latent diffusion trans- former.Advances in Neural Information Processing Systems, 37:121859–121881, 2024. 2

    2024

  56. [56]

    Less-to-more generalization: Unlocking more controllability by in-context generation.arXiv preprint arXiv:2504.02160, 2025

    Shaojin Wu, Mengqi Huang, Wenxu Wu, Yufeng Cheng, Fei Ding, and Qian He. Less-to-more generalization: Unlocking more controllability by in-context generation.arXiv preprint arXiv:2504.02160, 2025. 2, 3, 6, 7

    Pith/arXiv arXiv 2025

  57. [57]

    Direct3d-s2: Gigascale 3d generation made easy with spatial sparse attention.arXiv preprint arXiv:2505.17412, 2025

    Shuang Wu, Youtian Lin, Feihu Zhang, Yifei Zeng, Yikang Yang, Yajie Bao, Jiachen Qian, Siyu Zhu, Xun Cao, Philip Torr, et al. Direct3d-s2: Gigascale 3d generation made easy with spatial sparse attention.arXiv preprint arXiv:2505.17412, 2025. 3

    Pith/arXiv arXiv 2025

  58. [58]

    Structured 3d latents for scalable and versatile 3d gen- eration

    Jianfeng Xiang, Zelong Lv, Sicheng Xu, Yu Deng, Ruicheng Wang, Bowen Zhang, Dong Chen, Xin Tong, and Jiaolong Yang. Structured 3d latents for scalable and versatile 3d gen- eration. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 21469–21480, 2025. 2, 3, 6, 7

    2025

  59. [59]

    Frankenstein: Generating semantic- compositional 3d scenes in one tri-plane

    Han Yan, Yang Li, Zhennan Wu, Shenzhou Chen, Weix- uan Sun, Taizhang Shang, Weizhe Liu, Tian Chen, Xiaqiang Dai, Chao Ma, et al. Frankenstein: Generating semantic- compositional 3d scenes in one tri-plane. InSIGGRAPH Asia 2024 Conference Papers, pages 1–11, 2024. 3

    2024

  60. [60]

    Phycage: Physically plausible compositional 3d asset gener- ation from a single image.arXiv preprint arXiv:2411.18548,

    Han Yan, Mingrui Zhang, Yang Li, Chao Ma, and Pan Ji. Phycage: Physically plausible compositional 3d asset gener- ation from a single image.arXiv preprint arXiv:2411.18548,

    Pith/arXiv arXiv

  61. [61]

    X-part: high fidelity and structure coherent shape decomposition, 2025

    Xinhao Yan, Jiachen Xu, Yang Li, Changfeng Ma, Yunhan Yang, Chunshi Wang, Zibo Zhao, Zeqiang Lai, Yunfei Zhao, Zhuo Chen, and Chunchao Guo. X-part: high fidelity and structure coherent shape decomposition, 2025. 5

    2025

  62. [62]

    X-part: high fidelity and structure coher- ent shape decomposition.arXiv preprint arXiv:2509.08643,

    Xinhao Yan, Jiachen Xu, Yang Li, Changfeng Ma, Yunhan Yang, Chunshi Wang, Zibo Zhao, Zeqiang Lai, Yunfei Zhao, Zhuo Chen, et al. X-part: high fidelity and structure coher- ent shape decomposition.arXiv preprint arXiv:2509.08643,

    arXiv

  63. [63]

    Qwen3 technical report.arXiv preprint arXiv:2505.09388, 2025

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. Qwen3 technical report.arXiv preprint arXiv:2505.09388, 2025. 5, 6

    Pith/arXiv arXiv 2025

  64. [64]

    Holopart: Generative 3d part amodal segmentation.arXiv preprint arXiv:2504.07943, 2025

    Yunhan Yang, Yuan-Chen Guo, Yukun Huang, Zi-Xin Zou, Zhipeng Yu, Yangguang Li, Yan-Pei Cao, and Xihui Liu. Holopart: Generative 3d part amodal segmentation.arXiv preprint arXiv:2504.07943, 2025. 3

    Pith/arXiv arXiv 2025

  65. [65]

    Wonder3d++: Cross-domain diffusion for high-fidelity 3d generation from a single im- age.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

    Yuxiao Yang, Xiaoxiao Long, Zhiyang Dou, Cheng Lin, Yuan Liu, Qingsong Yan, Yuexin Ma, Haoqian Wang, Zhiqiang Wu, and Wei Yin. Wonder3d++: Cross-domain diffusion for high-fidelity 3d generation from a single im- age.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025. 2

    2025

  66. [66]

    Omnipart: Part-aware 3d genera- tion with semantic decoupling and structural cohesion.arXiv preprint arXiv:2507.06165, 2025

    Yunhan Yang, Yufan Zhou, Yuan-Chen Guo, Zi-Xin Zou, Yukun Huang, Ying-Tian Liu, Hao Xu, Ding Liang, Yan- Pei Cao, and Xihui Liu. Omnipart: Part-aware 3d genera- tion with semantic decoupling and structural cohesion.arXiv preprint arXiv:2507.06165, 2025. 2, 3, 6, 7

    Pith/arXiv arXiv 2025

  67. [67]

    Mvsnet: Depth inference for unstructured multi-view stereo

    Yao Yao, Zixin Luo, Shiwei Li, Tian Fang, and Long Quan. Mvsnet: Depth inference for unstructured multi-view stereo. InProceedings of the European conference on computer vi- sion (ECCV), pages 767–783, 2018. 2

    2018

  68. [68]

    Hi3dgen: High-fidelity 3d geometry generation from images via normal bridging.arXiv preprint arXiv:2503.22236, 3:2,

    Chongjie Ye, Yushuang Wu, Ziteng Lu, Jiahao Chang, Xi- aoyang Guo, Jiaqing Zhou, Hao Zhao, and Xiaoguang Han. Hi3dgen: High-fidelity 3d geometry generation from images via normal bridging.arXiv preprint arXiv:2503.22236, 3:2,

    Pith/arXiv arXiv

  69. [69]

    Ip- adapter: Text compatible image prompt adapter for text-to- image diffusion models.arXiv preprint arXiv:2308.06721,

    Hu Ye, Jun Zhang, Sibo Liu, Xiao Han, and Wei Yang. Ip- adapter: Text compatible image prompt adapter for text-to- image diffusion models.arXiv preprint arXiv:2308.06721,

    Pith/arXiv arXiv

  70. [70]

    3dshape2vecset: A 3d shape representation for neu- ral fields and generative diffusion models.ACM Transactions On Graphics (TOG), 42(4):1–16, 2023

    Biao Zhang, Jiapeng Tang, Matthias Niessner, and Peter Wonka. 3dshape2vecset: A 3d shape representation for neu- ral fields and generative diffusion models.ACM Transactions On Graphics (TOG), 42(4):1–16, 2023. 3

    2023

  71. [71]

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

    Longwen Zhang, Ziyu Wang, Qixuan Zhang, Qiwei Qiu, Anqi Pang, Haoran Jiang, Wei Yang, Lan Xu, and Jingyi Yu. Clay: A controllable large-scale generative model for creat- ing high-quality 3d assets.ACM Transactions on Graphics (TOG), 43(4):1–20, 2024. 3 11

    2024

  72. [72]

    Bang: Dividing 3d assets via generative exploded dynamics.ACM Transactions on Graphics (TOG), 44(4):1–21, 2025

    Longwen Zhang, Qixuan Zhang, Haoran Jiang, Yinuo Bai, Wei Yang, Lan Xu, and Jingyi Yu. Bang: Dividing 3d assets via generative exploded dynamics.ACM Transactions on Graphics (TOG), 44(4):1–21, 2025. 3

    2025

  73. [73]

    Shape-from-shading: a survey.IEEE trans- actions on pattern analysis and machine intelligence, 21(8): 690–706, 2002

    Ruo Zhang, Ping-Sing Tsai, James Edwin Cryer, and Mubarak Shah. Shape-from-shading: a survey.IEEE trans- actions on pattern analysis and machine intelligence, 21(8): 690–706, 2002. 2

    2002

  74. [74]

    Michelangelo: Conditional 3d shape generation based on shape-image-text aligned latent representation.Advances in neural information processing systems, 36:73969–73982,

    Zibo Zhao, Wen Liu, Xin Chen, Xianfang Zeng, Rui Wang, Pei Cheng, Bin Fu, Tao Chen, Gang Yu, and Shenghua Gao. Michelangelo: Conditional 3d shape generation based on shape-image-text aligned latent representation.Advances in neural information processing systems, 36:73969–73982,

  75. [75]

    Segmentdreamer: Towards high-fidelity text- to-3d synthesis with segmented consistency trajectory distil- lation

    Jiahao Zhu, Zixuan Chen, Guangcong Wang, Xiaohua Xie, and Yi Zhou. Segmentdreamer: Towards high-fidelity text- to-3d synthesis with segmented consistency trajectory distil- lation. InProceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 15864–15874, 2025. 2, 3 12

    2025