Pith. sign in

REVIEW 3 major objections 4 minor 7 cited by

VoxHammer: Training-Free Precise and Coherent 3D Editing in Native 3D Space

T0 review · 3 major / 4 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 VoxHammer claims that a pretrained structured 3D latent diffusion model can perform precise local 3D editing without any training, by inverting the input asset to noise and reusing the inverted latents and attention key/value tokens to lock

desk verdict A useful training-free 3D editing method with credible ablations, but the headline comparison is under-specified because the FLUX-edited 2D image may not be given to all baselines. read the letter →

arxiv 2508.19247 v1 pith:YXI2Q5MG submitted 2025-08-26 cs.CV

Lin Li , Zehuan Huang , Haoran Feng , Gengxiong Zhuang , Rui Chen , Chunchao Guo , Lu Sheng This is my paper
classification cs.CV
keywords 3Deditingtraining-freediffusioninversionstructuredlatentkey-valuereplacementlocalnativespacebenchmark
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

VoxHammer claims that precise local 3D editing can be done directly in the latent space of a pretrained structured 3D generative model, with no training. The method first inverts an input 3D asset to its terminal noise, caching latents and attention key/value tokens along the way; during editing it denoises from that noise while overwriting preserved regions with the cached features. This keeps untouched geometry and texture intact while letting the edited region change freely. On a new human-annotated benchmark of 100 objects with labeled 3D edit masks, the paper reports large gains over optimization-based and multi-view baselines across preservation, overall quality, and condition-alignment metrics. The whole pipeline runs at inference time on one GPU, making it a practical tool and a potential source of paired training data for future in-context 3D generation.

What carries the argument

The load-bearing device is a cached inversion trajectory: running the input 3D asset backward through the structured latent diffusion process to terminal noise while recording two quantities at each timestep—the inverted latent state and the key/value tensors of every self-attention layer. These caches are indexed by latent time, block, positional encoding, layer, and attention type. During editing, two masked replacement operations apply: latent replacement overwrites the denoising states of preserved voxels with the inverted states, and key-value replacement overwrites the K and V tensors of preserved tokens with the cached ones, optionally blocking attention mixing between edited and pres

What would settle it

Run the same asset through VoxHammer under two conditions: a straight inversion-reconstruction with no edit, and an edit with a semantically strong prompt (e.g., change a dog's head into a cat's) while keeping the mask small. If cached key/value replacement is doing its job, preserved-region geometry and texture in the edit output should match the inversion-reconstruction within the inversion's own error—masked PSNR within about 1 dB and Chamfer distance near-identical. A measurable divergence beyond that threshold, especially far from the mask, would show the new edit context is overriding th

Watch

Extended reading notes

Core claim

The paper establishes that local 3D editing can be cast as an inversion-and-inpainting problem entirely within the latent space of a structured 3D diffusion model. Given a 3D asset and a 3D edit mask, VoxHammer first runs the asset backward through the two-stage (structure, then sparse-latent) denoiser to recover its terminal noise, caching the latents and per-layer key/value tensors at every timestep. In the editing phase, it denoises from that noise with the edit condition, but at each step overwrites the preserved-region latents with the cached inverted latents and swaps in the cached key/value tokens for the attention layers. This feature replacement acts as an explicit geometric and tex

Load-bearing premise

The method assumes that cached inverted latents and attention key/value tokens from the source asset's inversion trajectory remain valid and self-consistent when the denoising trajectory is re-run with a new edit condition; the paper validates this empirically on its benchmark rather than by proof.

Editorial extensions

If this is right

  • Edits happen in the same 3D latent space used for generation, so there is no multi-view reconstruction step that can introduce position bias or cross-view inconsistency.
  • Unedited regions are not just softly influenced by the prompt; they are deterministically returned to the source trajectory, which is why masked PSNR/SSIM and Chamfer distance improve sharply over multi-view baselines.
  • Because the method is training-free and backbone-agnostic, any asset that can be encoded into structured latents—meshes, NeRFs, Gaussian splats, parts, or whole scenes—can be edited with the same recipe.
  • The pipeline can synthesize paired (original, edited) 3D assets with known 3D masks, providing the paired data that supervised in-context 3D editing models currently lack.
  • The two-stage inversion itself delivers high-fidelity reconstruction of the source asset, which can serve as a diagnostic or as an initialization for other 3D generative tasks.

Reading between the lines

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

  • Editorial inference: the cached key-value replacement recipe should transfer to image-space diffusion editing—substituting preserved-region attention tokens from a source inversion trajectory could give similarly strong background preservation for 2D local edits, where current methods rely on softer mask-weighting.
  • Editorial inference: the method's hard replacement of preserved features likely constrains edits most cleanly when the edit is local in latent-feature space; edits changing global semantics or lighting may over-constrain the result, so a boundary-aware or learned-blend version could widen the range of editable attributes.
  • Editorial inference: because VoxHammer outputs deterministic, explicitly masked edits, it can act as a data engine—filter its outputs with masked-preservation thresholds and feed them into a feed-forward 3D editor, converting the per-asset two-minute cost into amortized inference.
  • Editorial inference: the Edit3D-Bench protocol—human-annotated 3D masks with paired prompts—could become a standard evaluation for future 3D editing, since it makes preservation metrics such as masked PSNR and Chamfer distance directly comparable across methods.
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 / 4 minor

Summary. The paper proposes VoxHammer, a training-free 3D local editing method built on the TRELLIS structured 3D latent diffusion model. The source 3D asset is inverted through both the structure (ST) and sparse-latent (SLAT) stages, caching intermediate latents and self-attention key/value tensors. At editing time, the model denoises from the inverted noise and overwrites the latents and K/V tokens of unedited regions with the cached source features, guided by a 3D mask, while the edited region is conditioned on a 2D inpainted image produced by FLUX.1 Fill. The paper introduces Edit3D-Bench, a human-annotated benchmark of 100 objects with 3 prompts each, and reports that VoxHammer outperforms five baselines on unedited-region preservation, overall quality, and condition-alignment metrics. The central claim is that editing in native 3D space with inversion-based feature replacement yields precise preservation and coherent edits without training.

Significance. If the comparative claim is substantiated, VoxHammer is a timely and useful training-free alternative to multi-view-edit-and-reconstruct pipelines. The paper has concrete strengths: the two-stage inversion is clearly motivated and evaluated (Tab. 3, Fig. 6); the ablations in Tab. 1 show that both attention K/V replacement and inversion-based noise initialization contribute to preservation; and the new Edit3D-Bench addresses a real gap in evaluating 3D local editing. The main weaknesses concern the empirical validation rather than the method's internal derivation: the evaluation protocol does not isolate the contribution of the 3D mechanism from the strong 2D FLUX inpainting prior, and the headline comparative claim lacks statistical support. The supplementary limitation section honestly acknowledges weak text alignment and resolution constraints, which further tempers the abstract's text-prompt framing.

major comments (3)
  1. [Sec. 4.1 / Fig. 2 / Tab. 1] The comparative claim in Sec. 4.2 is confounded by unequal conditioning information. VoxHammer's own pipeline uses a FLUX.1 Fill inpainted image as a strong, task-specific edit signal (Fig. 2), and the benchmark includes a FLUX-generated edited image for every sample. The paper does not state whether Vox-E, MVEdit, Tailor3D, Instant3DiT, and TRELLIS receive the same edited image and 3D mask, or only the text prompt. If the baselines are text-only or use their own edit signals, the large margins in CLIP-T, DINO-I, FID, and FVD in Tab. 1 may reflect the power of FLUX.1 Fill rather than VoxHammer's native-3D latent/KV replacement. Please report the exact conditioning given to each baseline and add a control where all image-conditionable baselines receive the same FLUX-edited image, or an ablation that replaces FLUX with a weaker inpainter.
  2. [Sec. 4.2 / Tab. 1] The phrase 'significantly outperforms all baselines across nearly all metrics' is not supported statistically. Tab. 1 reports point estimates without standard deviations, confidence intervals, or paired significance tests across the 300 edit tasks. Some differences are small (CLIP-T 0.287 vs 0.283; DINO-I 0.947 vs 0.911), and the magnitude of the preservation gains (e.g., PSNR 41.68 vs 27.70) may be inflated by a favorable conditioning protocol. Please provide per-sample distributions, error bars, and statistical tests, or soften the 'significantly' claim accordingly.
  3. [Sec. 3.2 / Sec. 3.3 / Sec. 4.1] Several load-bearing hyperparameters are fixed without a held-out protocol or sensitivity analysis: the CFG scale (5.0 for both stages), the late-time CFG interval [0.5, 1.0], the soft-mask dilation/Gaussian falloff, and the optional attention masking. Since all comparisons and ablations use the same settings, it is unclear whether the reported performance is robust or overfit to Edit3D-Bench. Please add ablations over the interval and CFG scale, and report the soft-mask parameter choices. This is needed to support the claim that the method's behavior is not tuned specifically to the presented benchmark.
minor comments (4)
  1. [Sec. 4.3] The text says 'four variants' but only two are described (w/o Attn KV and w/ Noise Re-init). Please correct the count or list the missing variants.
  2. [Tab. 1] Vox-E reports CD as '/', so the geometry-preservation comparison is incomplete for one baseline. Please explain why (e.g., no accessible mesh) and indicate whether this affects the aggregate claims.
  3. [Sec. 7 / Tab. 2] DINO-I is described as 'similarity between the editing results and the original object'; for local editing this can reward conservative edits that change little. This should be discussed as a possible confound. Also, the user study in Tab. 2 compares only against two baselines and lacks details on task counts and inter-subject agreement.
  4. [General] Implementation details needed for reproducibility: the mask-projection procedure for masked PSNR/SSIM/LPIPS, the rendering/video sequence for FVD, and the soft-mask dilation/Gaussian parameters. No code or benchmark release is mentioned; releasing Edit3D-Bench would strengthen the contribution.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: VoxHammer's preserved-region mechanism is an explicit hard latent/KV replacement built on external base models, and the paper's comparative claims are empirical rather than derived from fitted constants.

full rationale

VoxHammer's editing operations are stated directly in Eqs. (4)-(7): preserved-region latents are overwritten with cached inversion latents and preserved-region K/V tokens are overwritten with cached tokens. This is an explicitly constructive mechanism, not a hidden reduction. The preservation metrics therefore chiefly measure inversion/decoding fidelity, but the paper does not disguise a fitted parameter as a prediction or define its claimed result in terms of the metric. Load-bearing components are external: TRELLIS [90] as the base model, RF-Solver [78] for inversion, and FLUX.1 Fill [3] for the 2D edit image. Author self-citations ([23], [34]-[36], [83]) appear only in related-work or application context and are not used to justify the central claim. The self-constructed Edit3D-Bench and the possible unequal use of the FLUX-edited image across baselines are evaluation and fairness concerns, not circularity of the derivation. The Limitation section acknowledges remaining text-alignment and backbone-resolution limits without revealing any step that reduces to its own input. Overall, no circular step is exhibited; the paper's contributions are engineering constructions validated empirically.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The method contributes a feature-replacement scheme on top of existing pretrained components (TRELLIS, RF-Solver, FLUX). There are no new physical entities. The central claim rests on four hand-set hyperparameters and five domain/ad-hoc assumptions, the most fragile being that cached inverted features remain compatible with a new edit condition.

free parameters (4)
  • Classifier-free guidance scale (omega) = 5.0 (both ST and SLAT stages)
    Sec. 4.1. Chosen to balance reconstruction fidelity and edit creativity; no validation protocol reported.
  • Sampling steps = 25 for inversion and denoising
    Sec. 4.1. Fixed discretization for ODE integration; no grid search or sensitivity analysis reported.
  • Late-time CFG interval = t in [0.5, 1.0]
    Sec. 3.2. CFG is applied only in this interval to preserve invertibility of early steps; interval is a hand-set constant.
  • Soft mask parameters = Not specified (dilation + Gaussian falloff)
    Sec. 3.3. The softness of the edit mask is described qualitatively; exact kernel sizes and sigma are absent, making the implementation ambiguous.
assumptions (5)
  • domain assumption TRELLIS is a pretrained structured 3D latent diffusion model that can encode and decode meshes, NeRFs, and 3DGS assets with high fidelity.
    Sec. 3.1. The entire method runs inside TRELLIS; any failure of the base model's encoding/decoding directly limits editing quality.
  • domain assumption The rectified-flow ODE is invertible and the RF-Solver second-order Taylor scheme (Eq. 1-2) gives accurate inversion in TRELLIS's 3D latent space.
    Sec. 3.2. The inversion trajectory is the foundation for all subsequent replacement; if the ODE integration error is large, the cached features will not reconstruct the source.
  • ad hoc to paper Applying CFG only in the late interval t in [0.5, 1.0] preserves invertibility while providing sufficient semantic guidance.
    Sec. 3.2. This schedule is specific to VoxHammer and is justified only empirically; no analysis of where the interval should be.
  • domain assumption The 2D image produced by FLUX.1 Fill (or another inpaint model) accurately represents the user's intended edit.
    Sec. 3 and Fig. 2. The edited image is the main conditioning signal; if it is wrong, the 3D edit is wrong regardless of the 3D machinery.
  • ad hoc to paper Overwriting latents and K/V tokens in unedited regions at every denoising step yields an on-manifold, coherent result.
    Sec. 3.3. This is the core heuristic; no theoretical guarantee that the mixed features lie on the model's data manifold.

how reviews work

0 comments
Cite this review

Pith. "Pith review of VoxHammer: Training-Free Precise and Coherent 3D Editing in Native 3D Space." pith.science (2026). https://pith.science/paper/YXI2Q5MG

@misc{pith2026250819247,
  author       = {Pith},
  title        = {Pith review of: VoxHammer: Training-Free Precise and Coherent 3D Editing in Native 3D Space},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YXI2Q5MG}},
  note         = {Machine review of arXiv:2508.19247}
}
read the original abstract

3D local editing of specified regions is crucial for game industry and robot interaction. Recent methods typically edit rendered multi-view images and then reconstruct 3D models, but they face challenges in precisely preserving unedited regions and overall coherence. Inspired by structured 3D generative models, we propose VoxHammer, a novel training-free approach that performs precise and coherent editing in 3D latent space. Given a 3D model, VoxHammer first predicts its inversion trajectory and obtains its inverted latents and key-value tokens at each timestep. Subsequently, in the denoising and editing phase, we replace the denoising features of preserved regions with the corresponding inverted latents and cached key-value tokens. By retaining these contextual features, this approach ensures consistent reconstruction of preserved areas and coherent integration of edited parts. To evaluate the consistency of preserved regions, we constructed Edit3D-Bench, a human-annotated dataset comprising hundreds of samples, each with carefully labeled 3D editing regions. Experiments demonstrate that VoxHammer significantly outperforms existing methods in terms of both 3D consistency of preserved regions and overall quality. Our method holds promise for synthesizing high-quality edited paired data, thereby laying the data foundation for in-context 3D generation. See our project page at https://huanngzh.github.io/VoxHammer-Page/.

Figures

Figures reproduced from arXiv: 2508.19247 by the authors.

Figure 1
Figure 1. High-quality 3D assets edited by our method using text prompts. Our method uses a training-free approach to perform percise [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Pipeline. Given an input 3D model, a user-specified editing region, and a text prompt, the off-the-shelf models [3, 40] are used to inpaint the rendered view from the 3D model. Subsequently, our VoxHammer, a training-free framework based on structured 3D diffusion models [90], performs native 3D editing conditioned on the input 3D and the edited image. Subsequent works [1, 2, 4, 5, 20, 26, 42, 63, 104] attempt to ed… view at source ↗
Figure 3
Figure 3. Architecture of VoxHammer. Our framework adopts TRELLIS [90] as the base model, which predicts sparse structures at the first structure (ST) stage and denoise fine-grained structured latents at the second sparse-latent (SLAT) stage. VoxHammer performs inversion prediction in both the ST and SLAT stages, which map the textured 3D asset to its terminal noise, with latents and key/value tensors cached at each timestep.… view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Qualitative comparisons on Edit3D-Bench. Our method achieves best performance on precision of editing and overall quality. edited image using DINO-I [60], and its alignment with the text prompt using CLIP-T [65]. 4.2. Main Results Quantitative comparison. As shown in T…
Figure 5
Figure 5. Figure 5: Ablation studies. Results demonstrate the effectiveness of key-value replacement in attention mechanism and latent replacement [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: The impact of inversion stages on reconstruction. ST stage inversion lacks detailed consistency, while inversion on both stages achieves fine-grained geometry and texture reconstruction. tively avoids these issues, demonstrating the robustness of our native 3D, inversi…
Figure 7
Figure 7. Figure 7: More applications. VoxHammer easily generalizes to part-aware 3D object, scene, and NeRF [58] or 3DGS [38] editing. We show the input models in the top row and the edited results in the bottom two rows. were then asked to select the result that best aligned with the te…
Figure 8
Figure 8. Figure 8: Visualization results of text-condition 3D editing [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: Pipeline of text-condition (left) and image-condition (right) 3D editing. Source Model Edited Result Source Model Edited Result Source Model Edited Result Some figures with Spider-Man. A robot with the cat head. A wooden toy with three balls red apples and a hammer. A …
Figure 10
Figure 10. Figure 10: More visualization results of image-condition 3D editing [PITH_FULL_IMAGE:figures/full_fig_p014_10.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 7 Pith papers

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

  1. Beyond Prompts: Unconditional 3D Inversion for Out-of-Distribution Shapes

    cs.CV 2026-04 unverdicted novelty 7.0 of 10

    Text-to-3D models lose prompt sensitivity for out-of-distribution shapes due to sink traps but retain geometric diversity via unconditional priors, enabling a decoupled inversion method for robust editing.

  2. VecSet-Edit: Unleashing Pre-trained LRM for Mesh Editing from Single Image

    cs.CV 2026-02 unverdicted novelty 7.0 of 10

    VecSet-Edit is the first method to perform high-fidelity mesh editing from a single image by analyzing and manipulating spatial token subsets in a pre-trained VecSet LRM.

  3. EditFlow3D: Automated Local Editing of 3D Assets with Trajectory Preservation

    cs.CV 2026-08 conditional novelty 6.0 of 10

    Mask-guided differential flow with a soft preservation loss enables training-free local 3D editing that keeps unedited regions close to the source asset.

  4. Hunyuan3D-Buffalo 1.0: A Unified Multimodal Model for Scalable 3D Generation, Understanding, and Editing

    cs.CV 2026-08 conditional novelty 6.0 of 10

    A unified 3D multimodal model combines understanding, text-to-3D generation, instruction-guided editing, and part generation in one architecture, trained on an 87M-sample corpus, with claimed state-of-the-art results.

  5. TanGO: Training-Free 3D Editing via Tangent-Space Guidance and Optimization

    cs.CV 2026-07 conditional novelty 6.0 of 10

    Per-token tangent-space steering, with strength set by velocity-direction mismatch, improves localized training-free 3D editing over global-scaling baselines.

  6. EditVerse3D: High-Quality 3D Object Editing with Region-Aware Learning

    cs.CV 2026-07 conditional novelty 6.0 of 10

    An end-to-end 3D editing framework achieves high-fidelity local edits from coarse bounding boxes and 2D image prompts using region-aware loss reweighting and a large-scale parts-derived training dataset.

  7. Edit in 2D, Verify in 3D: Reinforcement Learning for Multi-view Consistent Scene Editing

    cs.CV 2026-03 conditional novelty 6.0 of 10

    RL3DEdit fine-tunes FLUX-Kontext with GRPO using VGGT confidence and pose rewards to produce multi-view consistent 3D scene edits in a single pass.

Reference graph

Works this paper leans on

110 extracted references · 45 canonical work pages · cited by 7 Pith papers

  1. [1]

    Ed- itp23: 3d editing via propagation of image prompts to multi-view, 2025

    Roi Bar-On, Dana Cohen-Bar, and Daniel Cohen-Or. Ed- itp23: 3d editing via propagation of image prompts to multi-view, 2025. 2, 3

    2025

  2. [2]

    Kim, Noam Aigerman, Amit H

    Amir Barda, Matheus Gadelha, Vladimir G. Kim, Noam Aigerman, Amit H. Bermano, and Thibault Groueix. In- stant3dit: Multiview inpainting for fast editing of 3d ob- jects, 2024. 2, 3, 5, 6, 7, 1

    2024

  3. [3]

    FLUX.1 Tools: Introducing Fill, Depth, Canny, and Redux

    Black Forest Labs. FLUX.1 Tools: Introducing Fill, Depth, Canny, and Redux. https://bfl.ai/blog/24-11- 21-tools, 2024. Accessed: 2025-08-15. 3, 5

    2024

  4. [4]

    Mvinpainter: Learning multi-view consistent inpainting to bridge 2d and 3d editing, 2024

    Chenjie Cao, Chaohui Yu, Fan Wang, Xiangyang Xue, and Yanwei Fu. Mvinpainter: Learning multi-view consistent inpainting to bridge 2d and 3d editing, 2024. 2, 3

    2024

  5. [5]

    Generic 3d diffusion adapter using controlled multi-view editing, 2024

    Hansheng Chen, Ruoxi Shi, Yulin Liu, Bokui Shen, Jiayuan Gu, Gordon Wetzstein, Hao Su, and Leonidas Guibas. Generic 3d diffusion adapter using controlled multi-view editing, 2024. 2, 3, 5, 6, 7, 1

    2024

  6. [6]

    Shap-editor: Instruction-guided latent 3d editing in sec- onds, 2023

    Minghao Chen, Junyu Xie, Iro Laina, and Andrea Vedaldi. Shap-editor: Instruction-guided latent 3d editing in sec- onds, 2023. 2

    2023

  7. [7]

    Autopartgen: Autogres- sive 3d part generation and discovery

    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. 8

    arXiv 2025

  8. [8]

    Meshxl: Neural coordinate field for generative 3d founda- tion models, 2024

    Sijin Chen, Xin Chen, Anqi Pang, Xianfang Zeng, Wei Cheng, Yijun Fu, Fukun Yin, Yanru Wang, Zhibin Wang, Chi Zhang, Jingyi Yu, Gang Yu, Bin Fu, and Tao Chen. Meshxl: Neural coordinate field for generative 3d founda- tion models, 2024. 2

    2024

Show all 110 references
  1. [9]

    Meshanything: Artist- created mesh generation with autoregressive transformers,

    Yiwen Chen, Tong He, Di Huang, Weicai Ye, Sijin Chen, Jiaxiang Tang, Xin Chen, Zhongang Cai, Lei Yang, Gang Yu, Guosheng Lin, and Chi Zhang. Meshanything: Artist- created mesh generation with autoregressive transformers,

  2. [10]

    Ultra3d: Efficient and high- fidelity 3d generation with part attention, 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, 2025. 2

    2025

  3. [11]

    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 Rec...

    2023

  4. [12]

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

    Matt Deitke, Ruoshi Liu, Matthew Wallingford, Huong Ngo, Oscar Michel, Aditya Kusupati, Alan Fan, Chris- tian Laforte, Vikram V oleti, Samir Yitzhak Gadre, et al. Objaverse-xl: A universe of 10m+ 3d objects. Advances in Neural Information Processing Systems, 36, 2024. 2

    2024

  5. [13]

    Geometry in style: 3d stylization via sur- face normal deformation, 2025

    Nam Anh Dinh, Itai Lang, Hyunwoo Kim, Oded Stein, and Rana Hanocka. Geometry in style: 3d stylization via sur- face normal deformation, 2025. 2

    2025

  6. [14]

    Tela: Text to layer-wise 3d clothed human generation

    Junting Dong, Qi Fang, Zehuan Huang, Xudong Xu, Jingbo Wang, Sida Peng, and Bo Dai. Tela: Text to layer-wise 3d clothed human generation. In European Conference on Computer Vision, pages 19–36. Springer, 2025. 2

    2025

  7. [15]

    Interactive3d: Create what you want by interactive 3d generation, 2024

    Shaocong Dong, Lihe Ding, Zhanpeng Huang, Zibin Wang, Tianfan Xue, and Dan Xu. Interactive3d: Create what you want by interactive 3d generation, 2024. 2

    2024

  8. [16]

    From one to more: Contextual part latents for 3d gen- eration, 2025

    Shaocong Dong, Lihe Ding, Xiao Chen, Yaokun Li, Yuxin Wang, Yucheng Wang, Qi Wang, Jaehyeok Kim, Chenjian Gao, Zhanpeng Huang, Zibin Wang, Tianfan Xue, and Dan Xu. From one to more: Contextual part latents for 3d gen- eration, 2025. 2

    2025

  9. [17]

    From one to more: Contextual part latents for 3d generation

    Shaocong Dong, Lihe Ding, Xiao Chen, Yaokun Li, Yuxin Wang, Yucheng Wang, Qi Wang, Jaehyeok Kim, Chen- jian Gao, Zhanpeng Huang, et al. From one to more: Contextual part latents for 3d generation. arXiv preprint arXiv:2507.08772, 2025. 8

    2025

  10. [18]

    Prompt tuning inversion for text-driven image editing using diffusion models, 2023

    Wenkai Dong, Song Xue, Xiaoyue Duan, and Shumin Han. Prompt tuning inversion for text-driven image editing using diffusion models, 2023. 3

    2023

  11. [19]

    McHugh, and Vincent Vanhoucke

    Laura Downs, Anthony Francis, Nate Koenig, Brandon Kinman, Ryan Hickman, Krista Reymann, Thomas B. McHugh, and Vincent Vanhoucke. Google scanned ob- jects: A high-quality dataset of 3d scanned household items, 2022. 5

    2022

  12. [20]

    Preditor3d: Fast and precise 3d shape edit- ing, 2024

    Ziya Erkoc ¸, Can G ¨umeli, Chaoyang Wang, Matthias Nießner, Angela Dai, Peter Wonka, Hsin-Ying Lee, and Peiye Zhuang. Preditor3d: Fast and precise 3d shape edit- ing, 2024. 2, 3

    2024

  13. [21]

    Scaling rectified flow transformers for high-resolution image synthesis, 2024

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas M ¨uller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, Dustin Podell, Tim Dockhorn, Zion English, Kyle Lacey, Alex Goodwin, Yan- nik Marek, and Robin Rombach. Scaling rectified flow tr...

    2024

  14. [22]

    A point set generation network for 3d object reconstruction from a sin- gle image, 2016

    Haoqiang Fan, Hao Su, and Leonidas Guibas. A point set generation network for 3d object reconstruction from a sin- gle image, 2016. 5

    2016

  15. [23]

    Personalize anything for free with diffusion trans- former, 2025

    Haoran Feng, Zehuan Huang, Lin Li, Hairong Lv, and Lu Sheng. Personalize anything for free with diffusion trans- former, 2025. 3

    2025

  16. [24]

    Bermano, Gal Chechik, and Daniel Cohen-Or

    Rinon Gal, Yuval Alaluf, Yuval Atzmon, Or Patashnik, Amit H. Bermano, Gal Chechik, and Daniel Cohen-Or. An image is worth one word: Personalizing text-to-image gen- eration using textual inversion, 2022. 3

    2022

  17. [25]

    Me- shart: Generating articulated meshes with structure-guided transformers, 2025

    Daoyi Gao, Yawar Siddiqui, Lei Li, and Angela Dai. Me- shart: Generating articulated meshes with structure-guided transformers, 2025. 2

    2025

  18. [26]

    3d mesh editing using masked lrms,

    Will Gao, Dilin Wang, Yuchen Fan, Aljaz Bozic, Tuur Stuyck, Zhengqin Li, Zhao Dong, Rakesh Ranjan, and Nikolaos Sarafianos. 3d mesh editing using masked lrms,

  19. [27]

    Romero, Tsung-Yi Lin, and Ming-Yu Liu

    Zekun Hao, David W. Romero, Tsung-Yi Lin, and Ming-Yu Liu. Meshtron: High-fidelity, artist-like 3d mesh generation at scale, 2024. 2

    2024

  20. [28]

    Neural lightrig: Unlocking accurate object normal and material estimation with multi-light diffusion, 2024

    Zexin He, Tengfei Wang, Xin Huang, Xingang Pan, and Zi- wei Liu. Neural lightrig: Unlocking accurate object normal and material estimation with multi-light diffusion, 2024. 2

    2024

  21. [29]

    Gans trained by a two time-scale update rule converge to a local nash equi- librium, 2018

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equi- librium, 2018. 5

    2018

  22. [30]

    Classifier-free diffusion guidance

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. arXiv preprint arXiv:2207.12598, 2022. 4, 5

    2022 arXiv

  23. [31]

    Denoising dif- fusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. Advances in neural informa- tion processing systems, 33:6840–6851, 2020. 2, 3

    2020

  24. [32]

    Lrm: Large reconstruction model for single image to 3d

    Yicong Hong, Kai Zhang, Jiuxiang Gu, Sai Bi, Yang Zhou, Difan Liu, Feng Liu, Kalyan Sunkavalli, Trung Bui, and Hao Tan. Lrm: Large reconstruction model for single image to 3d. arXiv preprint arXiv:2311.04400, 2023. 2

    2023 arXiv

  25. [33]

    Stereo-gs: Multi-view stereo vision model for generalizable 3d gaussian splatting reconstruc- tion, 2025

    Xiufeng Huang, Ka Chun Cheung, Runmin Cong, Simon See, and Renjie Wan. Stereo-gs: Multi-view stereo vision model for generalizable 3d gaussian splatting reconstruc- tion, 2025. 2

    2025

  26. [34]

    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. arXiv preprint arXiv:2412.03632, 2024. 2

    2024 arXiv

  27. [35]

    Epidiff: Enhancing multi-view syn- thesis via localized epipolar-constrained diffusion

    Zehuan Huang, Hao Wen, Junting Dong, Yaohui Wang, Yangguang Li, Xinyuan Chen, Yan-Pei Cao, Ding Liang, Yu Qiao, Bo Dai, et al. Epidiff: Enhancing multi-view syn- thesis via localized epipolar-constrained diffusion. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision...

    2024

  28. [36]

    Midi: Multi-instance diffusion for single image to 3d scene generation

    Zehuan Huang, Yuan-Chen Guo, Xingqiao An, Yunhan Yang, Yangguang Li, Zi-Xin Zou, Ding Liang, Xihui Liu, Yan-Pei Cao, and Lu Sheng. Midi: Multi-instance diffusion for single image to 3d scene generation. In Proceedings of the Computer Vision and Pattern Recognition Conference ,...

    2025

  29. [37]

    An edit friendly ddpm noise space: Inversion and manipulations

    Inbar Huberman-Spiegelglas, Vladimir Kulikov, and Tomer Michaeli. An edit friendly ddpm noise space: Inversion and manipulations. In Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition , pages 12469–12478, 2024. 3

    2024

  30. [38]

    3d gaussian splatting for real-time radiance field rendering

    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,

  31. [39]

    Auto-encoding variational bayes

    Diederik P Kingma. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114, 2013. 2

    2013 arXiv

  32. [40]

    Black Forest Labs. Flux. https://github.com/ black-forest-labs/flux, 2024. 3

    2024

  33. [41]

    Mesh- pad: Interactive sketch-conditioned artist-reminiscent mesh generation and editing, 2025

    Haoxuan Li, Ziya Erkoc, Lei Li, Daniele Sirigatti, Vla- dyslav Rozov, Angela Dai, and Matthias Nießner. Mesh- pad: Interactive sketch-conditioned artist-reminiscent mesh generation and editing, 2025. 2

    2025

  34. [42]

    Cmd: Controllable multiview dif- fusion for 3d editing and progressive generation, 2025

    Peng Li, Suizhi Ma, Jialiang Chen, Yuan Liu, Congyi Zhang, Wei Xue, Wenhan Luo, Alla Sheffer, Wenping Wang, and Yike Guo. Cmd: Controllable multiview dif- fusion for 3d editing and progressive generation, 2025. 2, 3

    2025

  35. [43]

    Craftsman: High-fidelity mesh generation with 3d native generation and interactive geometry refiner

    Weiyu Li, Jiarui Liu, Rui Chen, Yixun Liang, Xuelin Chen, Ping Tan, and Xiaoxiao Long. Craftsman: High-fidelity mesh generation with 3d native generation and interactive geometry refiner. arXiv preprint arXiv:2405.14979, 2024. 2

    2024 arXiv

  36. [44]

    Craftsman3d: High-fidelity mesh generation with 3d native generation and interactive geometry refiner, 2025

    Weiyu Li, Jiarui Liu, Hongyu Yan, Rui Chen, Yixun Liang, Xuelin Chen, Ping Tan, and Xiaoxiao Long. Craftsman3d: High-fidelity mesh generation with 3d native generation and interactive geometry refiner, 2025. 2

    2025

  37. [45]

    Step1x- 3d: Towards high-fidelity and controllable generation of textured 3d assets, 2025

    Weiyu Li, Xuanyang Zhang, Zheng Sun, Di Qi, Hao Li, Wei Cheng, Weiwei Cai, Shihao Wu, Jiarui Liu, Zihao Wang, Xiao Chen, Feipeng Tian, Jianxiong Pan, Zeming Li, Gang Yu, Xiangyu Zhang, Daxin Jiang, and Ping Tan. Step1x- 3d: Towards high-fidelity and controllable generation of ...

    2025

  38. [46]

    Triposg: High- fidelity 3d shape synthesis using large-scale rectified flow models, 2025

    Yangguang Li, Zi-Xin Zou, Zexiang Liu, Dehu Wang, Yuan Liang, Zhipeng Yu, Xingchao Liu, Yuan-Chen Guo, Ding Liang, Wanli Ouyang, and Yan-Pei Cao. Triposg: High- fidelity 3d shape synthesis using large-scale rectified flow models, 2025. 2

    2025

  39. [47]

    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. 2

    2025 arXiv

  40. [48]

    Sparc3d: Sparse representation and construc- tion for high-resolution 3d shapes modeling

    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. 2

    2025 arXiv

  41. [49]

    Partcrafter: Structured 3d mesh generation via composi- tional latent diffusion transformers, 2025

    Yuchen Lin, Chenguo Lin, Panwang Pan, Honglei Yan, Yiqiang Feng, Yadong Mu, and Katerina Fragkiadaki. Partcrafter: Structured 3d mesh generation via composi- tional latent diffusion transformers, 2025. 2, 8

    2025

  42. [50]

    Part123: part-aware 3d reconstruction from a single-view image

    Anran Liu, Cheng Lin, Yuan Liu, Xiaoxiao Long, Zhiyang Dou, Hao-Xiang Guo, Ping Luo, and Wenping Wang. Part123: part-aware 3d reconstruction from a single-view image. In ACM SIGGRAPH 2024 Conference Papers , pages 1–12, 2024. 2, 8

    2024

  43. [51]

    Make-your-3d: Fast and consistent subject-driven 3d content generation, 2024

    Fangfu Liu, Hanyang Wang, Weiliang Chen, Haowen Sun, and Yueqi Duan. Make-your-3d: Fast and consistent subject-driven 3d content generation, 2024. 2

    2024

  44. [52]

    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, Ji- ayuan Gu, and Hao Su. One-2-3-45++: Fast single image to 3d objects with consistent multi-view generation and 3d diffusion. In Proceedings of the IEEE/CVF Conference on Co...

    2024

  45. [53]

    One-2-3-45: Any sin- gle image to 3d mesh in 45 seconds without per-shape opti- mization

    Minghua Liu, Chao Xu, Haian Jin, Linghao Chen, Mukund Varma T, Zexiang Xu, and Hao Su. One-2-3-45: Any sin- gle image to 3d mesh in 45 seconds without per-shape opti- mization. Advances in Neural Information Processing Sys- tems, 36, 2024. 2

    2024

  46. [54]

    Syncdreamer: Generating multiview-consistent images from a single-view image

    Yuan Liu, Cheng Lin, Zijiao Zeng, Xiaoxiao Long, Lingjie Liu, Taku Komura, and Wenping Wang. Syncdreamer: Generating multiview-consistent images from a single-view image. arXiv preprint arXiv:2309.03453, 2023. 2

    2023 arXiv

  47. [55]

    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, et al. Wonder3d: Single image to 3d using cross-domain diffusion. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Patt...

    2024

  48. [56]

    Repaint: Inpainting using denoising diffusion probabilistic models,

    Andreas Lugmayr, Martin Danelljan, Andres Romero, Fisher Yu, Radu Timofte, and Luc Van Gool. Repaint: Inpainting using denoising diffusion probabilistic models,

  49. [57]

    Lt3sd: Latent trees for 3d scene diffusion

    Quan Meng, Lei Li, Matthias Nießner, and Angela Dai. Lt3sd: Latent trees for 3d scene diffusion. arXiv preprint arXiv:2409.08215, 2024. 2

    2024 arXiv

  50. [58]

    Nerf: Representing scenes as neural radiance fields for view syn- thesis

    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, 3, 8

    2021

  51. [59]

    Null-text inversion for editing real im- ages using guided diffusion models

    Ron Mokady, Amir Hertz, Kfir Aberman, Yael Pritch, and Daniel Cohen-Or. Null-text inversion for editing real im- ages using guided diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6038–6047, 2023. 3

    2023

  52. [60]

    Dinov2: Learning robust vi- sual features without supervision, 2024

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mahmoud Assran, Nicolas Ballas, Wojciech Galuba, Rus- sell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michae...

    2024

  53. [61]

    Scalable diffusion mod- els with transformers

    William Peebles and Saining Xie. Scalable diffusion mod- els with transformers. In Proceedings of the IEEE/CVF In- ternational Conference on Computer Vision , pages 4195– 4205, 2023. 2

    2023

  54. [62]

    Barron, and Ben Milden- hall

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

    2022

  55. [63]

    Tailor3d: Customized 3d as- sets editing and generation with dual-side images, 2024

    Zhangyang Qi, Yunhan Yang, Mengchen Zhang, Long Xing, Xiaoyang Wu, Tong Wu, Dahua Lin, Xihui Liu, Jiaqi Wang, and Hengshuang Zhao. Tailor3d: Customized 3d as- sets editing and generation with dual-side images, 2024. 2, 3, 5, 7, 1

    2024

  56. [64]

    Deocc-1-to-3: 3d de-occlusion from a single image via self-supervised multi- view diffusion, 2025

    Yansong Qu, Shaohui Dai, Xinyang Li, Yuze Wang, You Shen, Liujuan Cao, and Rongrong Ji. Deocc-1-to-3: 3d de-occlusion from a single image via self-supervised multi- view diffusion, 2025. 2

    2025

  57. [65]

    Learning transferable visual models from natural language supervision, 2021

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision, 2021. 6

    2021

  58. [66]

    Hierarchical text-conditional image generation with clip latents

    Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image generation with clip latents. arXiv preprint arXiv:2204.06125, 1(2):3, 2022. 3

    2022 arXiv

  59. [67]

    L3dg: Latent 3d gaussian diffusion

    Barbara Roessle, Norman M ¨uller, Lorenzo Porzi, Samuel Rota Bulø, Peter Kontschieder, Angela Dai, and Matthias Nießner. L3dg: Latent 3d gaussian diffusion. arXiv preprint arXiv:2410.13530, 2024. 2

    2024 arXiv

  60. [68]

    High-resolution image synthesis with latent diffusion models, 2022

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models, 2022. 3

    2022

  61. [69]

    Semantic image inversion and editing using rectified stochastic differ- ential equations

    Litu Rout, Yujia Chen, Nataniel Ruiz, Constantine Cara- manis, Sanjay Shakkottai, and Wen-Sheng Chu. Semantic image inversion and editing using rectified stochastic differ- ential equations. arXiv preprint arXiv:2410.10792, 2024. 3

    2024 arXiv

  62. [70]

    Photorealistic text-to-image diffusion models with deep language understanding

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Sali- mans, et al. Photorealistic text-to-image diffusion models with deep language understanding. Advances in neural in- forma...

    2022

  63. [71]

    V ox-e: Text-guided voxel editing of 3d ob- jects, 2023

    Etai Sella, Gal Fiebelman, Peter Hedman, and Hadar Averbuch-Elor. V ox-e: Text-guided voxel editing of 3d ob- jects, 2023. 2, 5, 7, 1

    2023

  64. [72]

    Blended point cloud diffusion for localized text- guided shape editing

    Etai Sella, Noam Atia, Ron Mokady, and Hadar Averbuch- Elor. Blended point cloud diffusion for localized text- guided shape editing. arXiv preprint arXiv:2507.15399 ,

    arXiv

  65. [73]

    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. 2

    2023 arXiv

  66. [74]

    Denoising diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502, 2020. 2, 3

    2010 arXiv

  67. [75]

    Lgm: Large multi- view gaussian model for high-resolution 3d content cre- ation

    Jiaxiang Tang, Zhaoxi Chen, Xiaokang Chen, Tengfei Wang, Gang Zeng, and Ziwei Liu. Lgm: Large multi- view gaussian model for high-resolution 3d content cre- ation. In European Conference on Computer Vision, pages 1–18. Springer, 2025. 2

    2025

  68. [76]

    Efficient part-level 3d object generation via dual volume packing, 2025

    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, 2025. 2, 8

    2025

  69. [77]

    Sv3d: Novel multi-view synthesis and 3d generation from a single image using la- tent video diffusion

    Vikram V oleti, Chun-Han Yao, Mark Boss, Adam Letts, David Pankratz, Dmitry Tochilkin, Christian Laforte, Robin Rombach, and Varun Jampani. Sv3d: Novel multi-view synthesis and 3d generation from a single image using la- tent video diffusion. In European Conference on Computer...

    2025

  70. [78]

    Taming rectified flow for inversion and editing

    Jiangshan Wang, Junfu Pu, Zhongang Qi, Jiayi Guo, Yue Ma, Nisha Huang, Yuxin Chen, Xiu Li, and Ying Shan. Taming rectified flow for inversion and editing. arXiv preprint arXiv:2411.04746, 2024. 3

    2024 arXiv

  71. [79]

    Bovik, H.R

    Zhou Wang, A.C. Bovik, H.R. Sheikh, and E.P. Simoncelli. Image quality assessment: from error visibility to structural similarity. IEEE Transactions on Image Processing, 13(4): 600–612, 2004. 5

    2004

  72. [80]

    Llama-mesh: Unifying 3d mesh generation with language models, 2024

    Zhengyi Wang, Jonathan Lorraine, Yikai Wang, Hang Su, Jun Zhu, Sanja Fidler, and Xiaohui Zeng. Llama-mesh: Unifying 3d mesh generation with language models, 2024. 2

    2024

  73. [81]

    Crm: Single image to 3d textured mesh with convolutional reconstruction model

    Zhengyi Wang, Yikai Wang, Yifei Chen, Chendong Xi- ang, Shuo Chen, Dajiang Yu, Chongxuan Li, Hang Su, and Jun Zhu. Crm: Single image to 3d textured mesh with convolutional reconstruction model. arXiv preprint arXiv:2403.05034, 2024. 2

    2024 arXiv

  74. [82]

    Octgpt: Octree-based multi- scale autoregressive models for 3d shape generation, 2025

    Si-Tong Wei, Rui-Huan Wang, Chuan-Zhi Zhou, Baoquan Chen, and Peng-Shuai Wang. Octgpt: Octree-based multi- scale autoregressive models for 3d shape generation, 2025

    2025

  75. [83]

    Ouroboros3d: Image-to-3d gen- eration via 3d-aware recursive diffusion

    Hao Wen, Zehuan Huang, Yaohui Wang, Xinyuan Chen, Yu Qiao, and Lu Sheng. Ouroboros3d: Image-to-3d gen- eration via 3d-aware recursive diffusion. arXiv preprint arXiv:2406.03184, 2024. 2

    2024 arXiv

  76. [84]

    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. arXiv preprint arXiv:2405.20343 ,

    arXiv

  77. [85]

    Dipo: Dual-state images controlled articu- lated object generation powered by diverse data, 2025

    Ruiqi Wu, Xinjie Wang, Liu Liu, Chunle Guo, Jiaxiong Qiu, Chongyi Li, Lichao Huang, Zhizhong Su, and Ming- Ming Cheng. Dipo: Dual-state images controlled articu- lated object generation powered by diverse data, 2025. 2

    2025

  78. [86]

    Direct3d: Scalable image-to-3d generation via 3d latent diffusion transformer

    Shuang Wu, Youtian Lin, Feihu Zhang, Yifei Zeng, Jingxi Xu, Philip Torr, Xun Cao, and Yao Yao. Direct3d: Scalable image-to-3d generation via 3d latent diffusion transformer. arXiv preprint arXiv:2405.14832, 2024. 2

    2024 arXiv

  79. [87]

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

    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, 2025. 2

    2025

  80. [88]

    Amodal3r: Amodal 3d recon- struction from occluded 2d images, 2025

    Tianhao Wu, Chuanxia Zheng, Frank Guan, Andrea Vedaldi, and Tat-Jen Cham. Amodal3r: Amodal 3d recon- struction from occluded 2d images, 2025. 3

    2025

  81. [89]

    Blockfusion: Expandable 3d scene generation using latent tri-plane extrapolation

    Zhennan Wu, Yang Li, Han Yan, Taizhang Shang, Weixuan Sun, Senbo Wang, Ruikai Cui, Weizhe Liu, Hiroyuki Sato, Hongdong Li, et al. Blockfusion: Expandable 3d scene generation using latent tri-plane extrapolation. ACM Trans- actions on Graphics (TOG), 43(4):1–17, 2024. 2

    2024

  82. [90]

    Structured 3d latents for scalable and versa- tile 3d generation, 2025

    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 versa- tile 3d generation, 2025. 2, 3, 4, 5, 6, 7, 1

    2025

  83. [91]

    Instantmesh: Efficient 3d mesh generation from a single image with sparse-view large reconstruction models

    Jiale Xu, Weihao Cheng, Yiming Gao, Xintao Wang, Shenghua Gao, and Ying Shan. Instantmesh: Efficient 3d mesh generation from a single image with sparse-view large reconstruction models. arXiv preprint arXiv:2404.07191 ,

    arXiv

  84. [92]

    Lam, Yan-Pei Cao, and Xi- hui Liu

    Yunhan Yang, Yukun Huang, Yuan-Chen Guo, Liangjun Lu, Xiaoyang Wu, Edmund Y . Lam, Yan-Pei Cao, and Xi- hui Liu. Sampart3d: Segment any part in 3d objects, 2024. 5, 8

    2024

  85. [93]

    Omnipart: Part-aware 3d gen- eration with semantic decoupling and structural cohesion

    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 gen- eration with semantic decoupling and structural cohesion. arXiv preprint arXiv:2507.06165, 2025. 8

    2025 arXiv

  86. [94]

    Cast: Component-aligned 3d scene reconstruction from an rgb image

    Kaixin Yao, Longwen Zhang, Xinhao Yan, Yan Zeng, Qix- uan Zhang, Lan Xu, Wei Yang, Jiayuan Gu, and Jingyi Yu. Cast: Component-aligned 3d scene reconstruction from an rgb image. ACM Transactions on Graphics (TOG), 44(4): 1–19, 2025. 8

    2025

  87. [95]

    Shapellm-omni: A native multimodal llm for 3d generation and understanding, 2025

    Junliang Ye, Zhengyi Wang, Ruowen Zhao, Shenghao Xie, and Jun Zhu. Shapellm-omni: A native multimodal llm for 3d generation and understanding, 2025. 2

    2025

  88. [96]

    3dshape2vecset: A 3d shape representation for neural fields and generative diffusion models

    Biao Zhang, Jiapeng Tang, Matthias Niessner, and Peter Wonka. 3dshape2vecset: A 3d shape representation for neural fields and generative diffusion models. ACM Trans- actions On Graphics (TOG), 42(4):1–16, 2023. 2

    2023

  89. [97]

    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 Transactions on Graphics (TOG), 43(4):1–20, 2024. 2

    2024

  90. [98]

    Bang: Dividing 3d assets via generative exploded dynamics

    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, 44(4):1–21, 2025. 8

    2025

  91. [99]

    Efros, Eli Shecht- man, and Oliver Wang

    Richard Zhang, Phillip Isola, Alexei A. Efros, Eli Shecht- man, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric, 2018. 5

    2018

  92. [100]

    The scene language: Representing scenes with programs, words, and embeddings, 2025

    Yunzhi Zhang, Zizhang Li, Matt Zhou, Shangzhe Wu, and Jiajun Wu. The scene language: Representing scenes with programs, words, and embeddings, 2025. 2

    2025

  93. [101]

    Deepmesh: Auto- regressive artist-mesh creation with reinforcement learning,

    Ruowen Zhao, Junliang Ye, Zhengyi Wang, Guangce Liu, Yiwen Chen, Yikai Wang, and Jun Zhu. Deepmesh: Auto- regressive artist-mesh creation with reinforcement learning,

  94. [102]

    Assembler: Scalable 3d part assembly via an- chor point diffusion, 2025

    Wang Zhao, Yan-Pei Cao, Jiale Xu, Yuejiang Dong, and Ying Shan. Assembler: Scalable 3d part assembly via an- chor point diffusion, 2025. 2, 8

    2025

  95. [103]

    Michelangelo: Conditional 3d shape generation based on shape-image-text aligned latent representation

    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, 2024. 2

    2024

  96. [104]

    Pro3d-editor : A progressive-views perspective for consistent and precise 3d editing, 2025

    Yang Zheng, Mengqi Huang, Nan Chen, and Zhendong Mao. Pro3d-editor : A progressive-views perspective for consistent and precise 3d editing, 2025. 2, 3

    2025

  97. [105]

    Kv-edit: Training-free image editing for precise back- ground preservation, 2025

    Tianrui Zhu, Shiyi Zhang, Jiawei Shao, and Yansong Tang. Kv-edit: Training-free image editing for precise back- ground preservation, 2025. 3

    2025

  98. [106]

    Tip-editor: An accurate 3d editor fol- lowing both text-prompts and image-prompts, 2024

    Jingyu Zhuang, Di Kang, Yan-Pei Cao, Guanbin Li, Liang Lin, and Ying Shan. Tip-editor: An accurate 3d editor fol- lowing both text-prompts and image-prompts, 2024. 2 VoxHammer: Training-Free Precise and Coherent 3D Editing in Native 3D Space Supplementary Material

    2024

  99. [107]

    Text-Condition 3D Editing Benefiting from the versatility of TRELLIS [90], our frame- work also supports text-condition 3D editing by injecting textual conditions into the inversion and denoising stages of the base model for masked assets, as illustrated in Fig. 9. Leveraging ...

  100. [108]

    In terms of evaluating editing quality, FID assesses the overall visual similarity between the edited results and the original object

    Explanation of Evaluation Metrics In terms of evaluating unedited region preservation, Cham- fer Distance assesses the geometry consistency, while masked PSNR, SSIM and LPIPS of rendered multi-view images evaluate the consistency of structures and appear- ance. In terms of eva...

  101. [109]

    10, which demonstrates the ability to achieve precise and coherent 3D editing

    More Results More results of image-condition 3D editing are shown in Fig. 10, which demonstrates the ability to achieve precise and coherent 3D editing

  102. [110]

    A dog in a yellow raincoat with boots

    Limitation Although VoxHammer preserves unedited regions and maintains overall 3D quality, several limitations remain. First, textual alignment is not yet optimal, partly due to the scarcity of large-scale captioned 3D datasets, mak- ing text condition less robust than image-b...

Pith tools

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