Pith. sign in

REVIEW 3 major objections 4 minor 76 references

Dive3D: Diverse Distillation-based Text-to-3D Generation via Score Implicit Matching

T0 review · 3 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read The paper claims that mode collapse in text-to-3D generation comes from the asymmetric KL divergence inside SDS and reward losses, and that replacing it with a score-based matching loss delivers both more diversity and higher fidelity.

desk verdict Plausible application of score-implicit matching to text-to-3D, but the method is underspecified and the empirical claims outrun the evidence. read the letter →

arxiv 2506.13594 v1 pith:BAIJB5ZU submitted 2025-06-16 cs.CV

classification cs.CV
keywords text-to-3Dgenerationscoredistillationsamplingimplicitmatchingmodecollapsegenerativediversitydiffusionreward-guidedoptimizationKLdivergence
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

Dive3D claims that the limited diversity of text-to-3D generation is not a data or rendering problem but a loss-function problem: standard SDS and reward-guided objectives are linear combinations of an asymmetric KL divergence, which rewards mode-seeking and collapses outputs onto a few dominant solutions. The paper replaces each KL term with Score Implicit Matching (SIM), a score-based divergence that measures the distance between the gradient fields of the rendered-image distribution and the diffusion prior, and shows the resulting loss is tractable through the score-projection identity with a separate auxiliary score network. On 110 GPTEval3D prompts against nine baselines, the method reports the top score on every metric—text alignment, plausibility, geometry, texture—plus visibly wider variation in lighting, backgrounds, geometry, and style. A sympathetic reader takes away a practical recipe: swap the divergence, keep the pipeline, and diversity increases without paying for it in quality.

What carries the argument

The load-bearing object is the Score Implicit Matching (SIM) divergence, Eq. 15: a time-integrated, weighted distance between the score fields of two distributions, $D_{[0,T]}(p,q)=\int_0^T w(t)\,\mathbb{E}_{x_t\sim\pi_t}[d(s_p(x_t)-s_q(x_t))]\,dt$, where $s_p=\nabla_x\log p$ and $d$ is a distance function. It works because its gradient can be written, via the conditional score-projection identity (Appendix A, Eq. 20), as an expectation that requires only forward evaluations of the target score network plus a separately trained auxiliary score network—no backpropagation through the score function itself. Dive3D assembles three such terms: a conditional diffusion-prior term (S-CDP), an unconditional term (S-UDP), and an explicit reward term (S-ER), combined as $L_{\mathrm{Dive3D}}=(1+\gamma)L_{\mathrm{S\text{-}CDP}}-\gamma L_{\mathrm{S\text{-}UDP}}+\lambda L_{\mathrm{S\text{-}ER}}$.

What would settle it

Run Dive3D and the leading KL-based baseline on the same open-ended prompts with matched seeds, compute, and initialization, and measure within-prompt diversity by mean pairwise perceptual distance (e.g., LPIPS) across many samples; the claim fails if the score-based loss is not significantly more diverse. The sharper test is numerical: verify on a toy two-mode mixture whether Algorithm 1's gradient is an unbiased estimator of $\nabla_\theta D_{[0,T]}(p,q)$ for a fixed $\pi_t$ and $d$—if the estimate is biased in a way that favours one mode, the mechanism itself does not deliver what the equations promise.

Watch

Extended reading notes

Core claim

On its own terms, the paper's discovery is that the mode-seeking behaviour blamed for text-to-3D diversity loss is built into the loss, not the 3D representation: score distillation sampling reduces to $(1+\gamma)D_{\mathrm{KL}}(q_\theta(x_t|c)\,\|\,p(x_t|y_c))-\gamma D_{\mathrm{KL}}(q_\theta(x_t|c)\,\|\,p(x_t))$, and the reward term adds another KL against an exponential reward distribution, so every existing objective is an asymmetric KL combination. Dive3D replaces these with the SIM divergence $D_{[0,T]}(p,q)=\int_0^T w(t)\,\mathbb{E}_{x_t\sim\pi_t}[d(s_p(x_t)-s_q(x_t))]\,dt$, which aligns score fields rather than densities and therefore does not penalize a distribution for covering several high-probability regions. The gradient of this divergence is computed through the conditional score-projection identity, which swaps differentiation of the target score for an expectation against a separately trained score network, making the loss practical. The paper reports that under this loss, a single prompt produces many distinct plausible assets—different weather, times of day, architecture, poses—while all six GPTEval3D evaluation metrics improve over prior methods.

Load-bearing premise

The load-bearing premise is that the score divergence of Eq. 15 can actually be optimized through the gradient formula of Appendix A with a separately trained auxiliary network and that Algorithm 1's estimate is faithful to it; the paper never fixes the sampling distribution $\pi_t$, the distance $d$, or the auxiliary network's training objective, so an unstable or biased choice among them could erase the diversity gain.

Editorial extensions

If this is right

  • One prompt should now yield many distinct but plausible 3D assets—varied lighting, weather, background, geometry, and style—instead of a near-identical cluster, because score matching does not force the optimizer to pick a single dominant mode.
  • Diversity is not bought at the cost of alignment: the same loss tops all six GPTEval3D metrics (text–asset alignment, plausibility, text–geometry consistency, texture details, geometry details, overall) against nine baselines.
  • Reward guidance and diffusion priors become interchangeable terms of one divergence sum: the paper shows that reward-plus-unconditional-prior alone can generate 3D objects without a text-to-image model, so components can be reweighted or ablated systematically.
  • The recipe transfers across 3D representations—NeRF, mesh, and Gaussian splatting—and across diffusion backbones, with the loss being the main difference between pipelines.
  • Loss-combination ablations give practitioners a tuning dial: balanced conditional/unconditional weights (ratio near one) are best, and higher reward weights sharpen text detail while risking oversaturation.

Reading between the lines

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

  • If the causal story is right, the same KL-to-score swap should transfer to other SDS-based distillation settings—image-, video-, or sketch-conditional generation, and one-step diffusion distillation—wherever KL's mode-seeking pressure is the bottleneck; the paper's 2D experiments hint at this, but transfer beyond text-to-3D is a prediction the paper does not test.
  • The paper leaves the sampling distribution $\pi_t$ and the distance $d$ unspecified; a natural follow-up is to map how the diversity–fidelity trade-off moves as $\pi_t$ widens around the rendered-data distribution, since the mechanism predicts wider support means more mode coverage and more diversity.
  • The framework predicts an interaction with prompt specificity: open-ended prompts should show the largest diversity gain over KL baselines, while maximal-specificity prompts should show almost none, because with a single narrow mode the two divergences nearly coincide—a checkable prediction the paper does not report.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 Dive3D, a text-to-3D generation framework that replaces the KL-divergence-based objectives used by SDS and reward-guided methods with a score-based divergence called Score Implicit Matching (SIM). The authors argue that asymmetric KL divergence is mode-seeking, and that matching score fields instead of densities encourages diversity without sacrificing fidelity or alignment. They present a unified divergence view in which SDS and reward losses are linear combinations of CDP, UDP, and ER KL terms, then replace each with a corresponding score-based loss. Experiments on the GPTEval3D benchmark and additional qualitative comparisons against nine baselines report improvements in text alignment, plausibility, texture/geometry details, human preference, and output diversity. The appendix contains a derivation of a conditional score-projection identity and a pseudo-code algorithm for the proposed updates.

Significance. If the central claim held, Dive3D would be a practically useful loss for diverse text-to-3D generation, and the unification of distillation and reward guidance under one divergence family could be a useful conceptual perspective. The paper has concrete strengths: it evaluates on a standard benchmark against many strong baselines, tests multiple 3D representations (NeRF, mesh, Gaussian splatting), and includes ablations over loss-weight combinations. These empirical efforts are valuable. However, the theoretical derivation of the SIM objective is substantially underspecified, and the specific mechanism by which diversity arises is not established by the equations or the algorithm as written. The lack of a quantitative diversity metric is also a gap for a paper whose headline claim is 'mitigates mode collapse'. The significance is therefore conditional on a corrected derivation and a more complete algorithm specification.

major comments (3)
  1. [4.1, Eq. (12)] The derivation of L_ER as a KL divergence is not valid as written. The term E_{t,c,x_t}[ω(t) log q_θ(x_t|c)] is declared constant on the grounds that q_θ is a uniform collection of N particles, but the positions of those particles are determined by the rendered images g(θ,c), so this expectation has non-vanishing θ-dependence. Furthermore, p_ER(y_c,x_t) ∝ exp(r(y_c, xhat_0(x_t))) is unnormalized, so D_KL(q_θ || p_ER) is not defined without a partition function. Because the claim that reward guidance can be written as a KL divergence underpins the unified framework in Eq. (14), this step needs to be corrected or explicitly reframed as a heuristic analogy.
  2. [Appendix A and Algorithm 1] The core optimization claim is not established. Eq. (15) defines a score-based divergence with an unspecified distance function d and an unspecified sampling distribution π_t, and neither is defined anywhere in the paper. The proof of Theorem 1 is compressed, and the symbols in Eq. (22), in particular s_q(y|x_t), are never defined. No argument is given that ∇_θ of Eq. (15) equals the update implemented in Algorithm 1. Lines 17–24 of Algorithm 1 call a function d_score(...) that is never defined, and line 29 asks to update φ by 'Denoising Score Matching(x_t,t)' without specifying the training data, the conditioning, or how φ relates to q_θ(x_t|c). Without these definitions, the reported diversity gain cannot be attributed to the SIM mechanism rather than to the auxiliary score network, reward weighting, or unstated hyperparameters.
  3. [5.2, Figs. 1(a), 5, and 7] The central claim of improved diversity is evaluated only qualitatively. No quantitative diversity metric (e.g., pairwise LPIPS, coverage, number of detected modes, or variance across seeds/particles) is reported for either the 2D or 3D settings. The figures show selected samples rather than a distributional assessment, so a reader cannot tell whether the apparent diversity is consistent or merely cherry-picked. Since the abstract states that SIM 'effectively mitigates mode collapse,' the paper should include a quantitative diversity measurement over multiple runs or seeds.
minor comments (4)
  1. [Table 1] Table 1 reports a single run of GPT-4o-mini evaluations with no variance, confidence intervals, or significance tests; the footnote explains the switch from GPT-4V, but the headline claim of 'highest score on every metric' needs repeated evaluation or at least a statement about evaluation noise.
  2. [5.1 and Table 1] Section 5.1 says the comparison includes nine state-of-the-art methods, but Table 1 additionally lists SyncDreamer, so the baseline count and the table rows are inconsistent.
  3. [Appendix B] The runtime statements are inconsistent: Section 5.1 says optimization takes about one hour per object on an A100, while Appendix B reports 1.5 hours on an A100 for NeRF generation and 12 hours for mesh generation. Please clarify the actual settings.
  4. [Eq. (5) and Algorithm 1] The notation for the score network is inconsistent: the main text uses φ for the diffusion model in Eq. (4)–(5), while Algorithm 1 uses φ for the LoRA parameters and s_φ for an auxiliary network. Please unify the symbols to avoid confusion.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the SIM objective is explicitly defined and the diversity claims are empirically tested against external baselines; the under-specification of d_score and pi_t is a reproducibility gap, not a circular reduction.

full rationale

Dive3D's central claim is that replacing KL-based objectives with the score-based SIM loss improves diversity and fidelity. The SIM divergence is defined in Eq. 15 with explicit score functions, a distance d, a sampling distribution pi_t, and a weighting w(t); the Dive3D objective in Eq. 16 is then formed as a linear combination of score-based CDP/UDP/ER terms. This is a construction, not a prediction derived from itself: the diversity gain is not asserted to follow algebraically from Eq. 15 but is demonstrated empirically on the GPTEval3D benchmark and in controlled 2D and 3D comparisons against DreamFusion, ProlificDreamer, MVDream, and DreamReward. The theoretical motivation cites the authors' earlier SIM work (refs 31 and 33), which shares co-author Weijian Luo, but that prior work is externally validated on one-step image and flow distillation benchmarks, and the appendix independently reproduces the score-projection identity argument leading to Eq. 22. No fitted parameter is renamed as a prediction: the loss weights gamma and lambda are reported hyperparameters, and the outcome metrics are measured on generated assets rather than read off from the loss definition. The main weakness is under-specification — d_score, pi_t, and the auxiliary-network training are not fully defined, so the implemented update is not rigorously connected to Eq. 15. That is a correctness and reproducibility concern, not a circularity: it does not make the conclusion equal to the input. Hence no circular step can be exhibited, and the appropriate finding is no significant circularity.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central loss depends on two tuned scalar weights and on a set of unstated assumptions about the diffusion score, the exponential reward model, the constancy of the entropy term, the score-projection identity, and the reliability of the LLM evaluator. No new physical or conceptual entities are introduced.

free parameters (2)
  • CFG guidance scale gamma = 7.5 (mesh), 20 (Gaussian Splatting)
    Appendix B reports tuned per representation; authors note high values oversaturate colors and low values hinder convergence.
  • Reward scale lambda (PickScore) = 100 (mesh), 10,000 (Gaussian Splatting)
    Appendix B reports tuned per representation; high reward scale increases visual richness but oversaturates colors.
assumptions (5)
  • domain assumption Pretrained 2D diffusion score s_phi(xt, y, t) is a valid score for the view-conditioned target distribution p(xt | yc).
    Used in Sec. 3.2 and Eq. 4 to define SDS and in Eq. 16 as the score target for the CDP and UDP terms.
  • ad hoc to paper Reward defines an exponential distribution pER(yc, xt) proportional to exp(r(yc, xhat_0(xt))).
    Eq. 11 introduced without justification; needed to rewrite reward maximization as KL minimization in Eq. 12.
  • ad hoc to paper The entropy term E_qtheta[log qtheta] in the KL decompositions is constant during optimization.
    Eq. 12 relies on qtheta(xt | c) = 1/N to drop the log-density term; qtheta depends on theta through rendered images, so this is not generally true.
  • standard math The score-projection identity (Appendix Theorem 1) and the gradient formula in Eq. 20 are valid under the stated 'mild conditions'.
    The appendix does not state the conditions, and the proof is compressed; the entire tractability of the SIM gradient rests on this.
  • domain assumption GPT-4o-mini scores measure text alignment, plausibility, and quality reliably.
    Table 1 uses GPTEval3D with GPT-4o-mini after GPT-4V deprecation; no agreement study with human raters is reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Dive3D: Diverse Distillation-based Text-to-3D Generation via Score Implicit Matching." pith.science (2026). https://pith.science/paper/BAIJB5ZU

@misc{pith2026250613594,
  author       = {Pith},
  title        = {Pith review of: Dive3D: Diverse Distillation-based Text-to-3D Generation via Score Implicit Matching},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BAIJB5ZU}},
  note         = {Machine review of arXiv:2506.13594}
}
read the original abstract

Distilling pre-trained 2D diffusion models into 3D assets has driven remarkable advances in text-to-3D synthesis. However, existing methods typically rely on Score Distillation Sampling (SDS) loss, which involves asymmetric KL divergence--a formulation that inherently favors mode-seeking behavior and limits generation diversity. In this paper, we introduce Dive3D, a novel text-to-3D generation framework that replaces KL-based objectives with Score Implicit Matching (SIM) loss, a score-based objective that effectively mitigates mode collapse. Furthermore, Dive3D integrates both diffusion distillation and reward-guided optimization under a unified divergence perspective. Such reformulation, together with SIM loss, yields significantly more diverse 3D outputs while improving text alignment, human preference, and overall visual fidelity. We validate Dive3D across various 2D-to-3D prompts and find that it consistently outperforms prior methods in qualitative assessments, including diversity, photorealism, and aesthetic appeal. We further evaluate its performance on the GPTEval3D benchmark, comparing against nine state-of-the-art baselines. Dive3D also achieves strong results on quantitative metrics, including text-asset alignment, 3D plausibility, text-geometry consistency, texture quality, and geometric detail.

Figures

Figures reproduced from arXiv: 2506.13594 by the authors.

Figure 1
Figure 1. We propose Dive3D, a novel text-to-3D generation framework that enables both diverse and high-fidelity 3D content creation. Top: Compared to ProlificDreamer, which exhibits mode collapse under the same prompt, Dive3D generates a diverse set of realistic and semantically aligned 3D outputs. Bottom: Dive3D unifies SDS and human-preference rewards within a divergence-based framework, effectively combining diffusion pri… view at source ↗
Figure 2
Figure 2. Overview of Dive3D. Dive3D reformulates both SDS loss and human-preference rewards within a unified divergence-based framework, revealing that both rely on KL divergence, DKL(·, ·)—an asymmetric objective that inherently favors mode-seeking and leads to model collapse and limited diversity. Dive3D replaces this with a Score Implicit Matching-based divergence, D[0,T ](·, ·), which aligns score fields rather than prob… view at source ↗
Figure 4
Figure 4. Comparison with Baselines based on MVDiffusion [48] and reward model [67]. Dive3D exhibits more detailed and realistic 3D generation, capturing fine-grained structures such as accurate guitar geometry and transparent glass materials. 5.1. Evaluation on the GPTEval3D Benchmark Setup. We first evaluate Dive3D on 110 creative and complex prompts from the GPTEval3D benchmark [61], comparing against 9 state-of-the-art me… view at source ↗
Figures from the paper (7 more)
Figure 5
Figure 5. Figure 5: Score-based divergence vs. KL divergence in 2D space sampling. The proposed score-based divergence significantly enhances the diversity of generated 2D samples, yielding more varied backgrounds and clothing in “game character” generation, as well as a broader range of …
Figure 6
Figure 6. Figure 6: More comparison results of Dive3D, showing improved quality over prior methods. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: More comparison results of Dive3D, showing improved diversity over prior methods. [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Comparison with baselines on Gaussian Splatting. Dive3D can generate diverse and human-preferred 3D Gaussian splattings. Algorithm 1: Pseudo-code for Dive3D 1 1 Input: Text-to-image diffusion score network sϕ(·, ·, ·) and its LoRA parameters φ, Reward function R(·, ·) …
Figure 9
Figure 9. Figure 9: Dive3D can generate 3D objects across different 3D representations. We show the results on Gaussian splitting, NeRF and Mesh. The prompts are A 3D model of an adorable cottage with a thatched roof and A cauldron full of gold coins. Backbones. We utilize Stable Diffusio…
Figure 10
Figure 10. Figure 10: Dive3D 3D Gaussian Splattings. Dive3D can generate anime-style and game-style 3D Gaussian Splattings. uniformly apply identical negative prompts during the assessments. The primary distinction between the methods lies in their respective loss functions. The total trai…
Figure 11
Figure 11. Figure 11: Exploration of the design space of the unified divergence loss. We investigate their influence of different text-to-3D divergence losses on generation performance by adjusting their weights. 3D Gaussian splitting are used as representations in these experiments for co…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

76 extracted references · 36 canonical work pages

  1. [1]

    Self-rectifying diffu- sion sampling with perturbed-attention guidance

    Donghoon Ahn, Hyoungwon Cho, Jaewon Min, Wooseok Jang, Jungwoo Kim, SeonHwa Kim, Hyun Hee Park, Ky- ong Hwan Jin, and Seungryong Kim. Self-rectifying diffu- sion sampling with perturbed-attention guidance. InEuropean Conference on Computer Vision, pages 1–17. Springer, 2024. 3

  2. [2]

    Re- imagine the negative prompt algorithm: Transform 2d diffusion into 3d, alleviate janus problem and beyond.arXiv preprint arXiv:2304.04968, 2023

    Mohammadreza Armandpour, Huangjie Zheng, Ali Sadeghian, Amir Sadeghian, and Mingyuan Zhou. Re- imagine the negative prompt algorithm: Transform 2d diffusion into 3d, alleviate janus problem and beyond.arXiv preprint arXiv:2304.04968, 2023. 2

  3. [3]

    ediff-i: Text-to-image dif- fusion models with an ensemble of expert denoisers.arXiv preprint arXiv:2211.01324, 2022

    Yogesh Balaji, Seungjun Nah, Xun Huang, Arash Vahdat, Jiaming Song, Qinsheng Zhang, Karsten Kreis, Miika Aittala, Timo Aila, Samuli Laine, et al. ediff-i: Text-to-image dif- fusion models with an ensemble of expert denoisers.arXiv preprint arXiv:2211.01324, 2022. 2

  4. [4]

    Fan- tasia3d: Disentangling geometry and appearance for high- quality text-to-3d content creation

    Rui Chen, Yongwei Chen, Ningxin Jiao, and Kui Jia. Fan- tasia3d: Disentangling geometry and appearance for high- quality text-to-3d content creation. InProceedings of the IEEE/CVF international conference on computer vision, pages 22246–22256, 2023. 2, 6, 7

  5. [5]

    Consistency models made easy.arXiv preprint arXiv:2406.14548, 2024

    Zhengyang Geng, Ashwini Pokle, William Luo, Justin Lin, and J Zico Kolter. Consistency models made easy.arXiv preprint arXiv:2406.14548, 2024. 2

  6. [6]

    threestudio: A unified framework for 3d content generation

    Yuan-Chen Guo, Ying-Tian Liu, Ruizhi Shao, Christian Laforte, Vikram V oleti, Guan Luo, Chia-Hao Chen, Zi- Xin Zou, Chen Wang, Yan-Pei Cao, and Song-Hai Zhang. threestudio: A unified framework for 3d content generation. https://github.com/threestudio- project/ threestudio, 2023. 6

  7. [7]

    Classifier-free diffusion guidance.arXiv preprint arXiv:2207.12598, 2022

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance.arXiv preprint arXiv:2207.12598, 2022. 3

  8. [8]

    Denoising diffu- sion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffu- sion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020. 3

Show all 76 references
  1. [9]

    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. 3

  2. [10]

    Turbo3d: Ultra-fast text-to-3d generation.arXiv preprint arXiv:2412.04470, 2024

    Hanzhe Hu, Tianwei Yin, Fujun Luan, Yiwei Hu, Hao Tan, Zexiang Xu, Sai Bi, Shubham Tulsiani, and Kai Zhang. Turbo3d: Ultra-fast text-to-3d generation.arXiv preprint arXiv:2412.04470, 2024. 3

  3. [11]

    Dreamtime: An improved optimization strategy for diffusion-guided 3d generation

    Yukun Huang, Jianan Wang, Yukai Shi, Boshi Tang, Xianbiao Qi, and Lei Zhang. Dreamtime: An improved optimization strategy for diffusion-guided 3d generation. InThe Twelfth International Conference on Learning Representations, 2024. 2

  4. [12]

    Flow generator matching.arXiv preprint arXiv:2410.19310, 2024

    Zemin Huang, Zhengyang Geng, Weijian Luo, and Guo- jun Qi. Flow generator matching.arXiv preprint arXiv:2410.19310, 2024. 2, 5

  5. [13]

    Zero-shot text-guided object generation with dream fields

    Ajay Jain, Ben Mildenhall, Jonathan T Barron, Pieter Abbeel, and Ben Poole. Zero-shot text-guided object generation with dream fields. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 867–876,

  6. [14]

    Guiding a diffu- sion model with a bad version of itself.Advances in Neural Information Processing Systems, 37:52996–53021, 2024

    Tero Karras, Miika Aittala, Tuomas Kynkäänniemi, Jaakko Lehtinen, Timo Aila, and Samuli Laine. Guiding a diffu- sion model with a bad version of itself.Advances in Neural Information Processing Systems, 37:52996–53021, 2024. 3

  7. [15]

    Noise-free score distillation

    Oren Katzir, Or Patashnik, Daniel Cohen-Or, and Dani Lischinski. Noise-free score distillation. InThe Twelfth International Conference on Learning Representations, 2024. 2

  8. [16]

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

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering.ACM Trans. Graph., 42(4):139–1, 2023. 2

  9. [17]

    Pick-a-pic: An open dataset of user preferences for text-to-image generation.Advances in Neural Information Processing Systems, 36:36652–36663,

    Yuval Kirstain, Adam Polyak, Uriel Singer, Shahbuland Ma- tiana, Joe Penna, and Omer Levy. Pick-a-pic: An open dataset of user preferences for text-to-image generation.Advances in Neural Information Processing Systems, 36:36652–36663,

  10. [18]

    Instant3d: Fast text-to-3d with sparse-view generation and large reconstruction model.arXiv preprint arXiv:2311.06214, 2023

    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, 6, 7

  11. [19]

    Self-guidance: Boosting flow and diffusion generation on their own.arXiv preprint arXiv:2412.05827,

    Tiancheng Li, Weijian Luo, Zhiyang Chen, Liyuan Ma, and Guo-Jun Qi. Self-guidance: Boosting flow and diffusion generation on their own.arXiv preprint arXiv:2412.05827,

  12. [20]

    Luciddreamer: Towards high-fidelity text-to-3d generation via interval score matching.arXiv preprint arXiv:2311.11284, 2023

    Yixun Liang, Xin Yang, Jiantao Lin, Haodong Li, Xiaogang Xu, and Yingcong Chen. Luciddreamer: Towards high-fidelity text-to-3d generation via interval score matching.arXiv preprint arXiv:2311.11284, 2023. 2

  13. [21]

    Magic3d: High-resolution text- to-3d content creation

    Chen-Hsuan Lin, Jun Gao, Luming Tang, Towaki Takikawa, Xiaohui Zeng, Xun Huang, Karsten Kreis, Sanja Fidler, Ming- Yu Liu, and Tsung-Yi Lin. Magic3d: High-resolution text- to-3d content creation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogniti...

  14. [22]

    Pseudo numerical methods for diffusion models on manifolds.arXiv preprint arXiv:2202.09778, 2022

    Luping Liu, Yi Ren, Zhijie Lin, and Zhou Zhao. Pseudo numerical methods for diffusion models on manifolds.arXiv preprint arXiv:2202.09778, 2022. 3

  15. [23]

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

    Ruoshi Liu, Rundi Wu, Basile Van Hoorick, Pavel Tokmakov, Sergey Zakharov, and Carl V ondrick. Zero-1-to-3: Zero-shot one image to 3d object. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 9298– 9309, 2023. 2

  16. [24]

    Syncdreamer: Gener- ating multiview-consistent images from a single-view image

    Yuan Liu, Cheng Lin, Zijiao Zeng, Xiaoxiao Long, Lingjie Liu, Taku Komura, and Wenping Wang. Syncdreamer: Gener- ating multiview-consistent images from a single-view image. InThe Twelfth International Conference on Learning Repre- sentations, 2024. 7

  17. [25]

    Wonder3d: Sin- gle image to 3d using cross-domain diffusion.arXiv preprint arXiv:2310.15008, 2023

    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.arXiv preprint arXiv:2310.15008, 2023. 2

  18. [26]

    Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps.Advances in Neural Information Processing Systems, 35:5775–5787, 2022

    Cheng Lu, Yuhao Zhou, Fan Bao, Jianfei Chen, Chongxuan Li, and Jun Zhu. Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps.Advances in Neural Information Processing Systems, 35:5775–5787, 2022. 3 11

  19. [27]

    A comprehensive survey on knowledge distilla- tion of diffusion models.arXiv preprint arXiv:2304.04262,

    Weijian Luo. A comprehensive survey on knowledge distilla- tion of diffusion models.arXiv preprint arXiv:2304.04262,

  20. [28]

    Diff-instruct: A universal ap- proach for transferring knowledge from pre-trained diffusion models.Advances in Neural Information Processing Systems, 36:76525–76546, 2023

    Weijian Luo, Tianyang Hu, Shifeng Zhang, Jiacheng Sun, Zhenguo Li, and Zhihua Zhang. Diff-instruct: A universal ap- proach for transferring knowledge from pre-trained diffusion models.Advances in Neural Information Processing Systems, 36:76525–76546, 2023. 2

  21. [29]

    Entropy-based training methods for scalable neural implicit samplers.Ad- vances in Neural Information Processing Systems, 36:7137– 7157, 2023

    Weijian Luo, Boya Zhang, and Zhihua Zhang. Entropy-based training methods for scalable neural implicit samplers.Ad- vances in Neural Information Processing Systems, 36:7137– 7157, 2023. 2

  22. [30]

    Diff-instruct: A universal ap- proach for transferring knowledge from pre-trained diffusion models.Advances in Neural Information Processing Systems, 36, 2024

    Weijian Luo, Tianyang Hu, Shifeng Zhang, Jiacheng Sun, Zhenguo Li, and Zhihua Zhang. Diff-instruct: A universal ap- proach for transferring knowledge from pre-trained diffusion models.Advances in Neural Information Processing Systems, 36, 2024. 2, 3

  23. [31]

    One-step diffusion distillation through score implicit matching.Advances in Neural Information Process- ing Systems, 37:115377–115408, 2024

    Weijian Luo, Zemin Huang, Zhengyang Geng, J Zico Kolter, and Guo-jun Qi. One-step diffusion distillation through score implicit matching.Advances in Neural Information Process- ing Systems, 37:115377–115408, 2024. 2, 5

  24. [32]

    Diff-instruct*: Towards human-preferred one- step text-to-image generative models.arXiv preprint arXiv:2410.20898, 2024

    Weijian Luo, Colin Zhang, Debing Zhang, and Zhengyang Geng. Diff-instruct*: Towards human-preferred one- step text-to-image generative models.arXiv preprint arXiv:2410.20898, 2024. 3

  25. [33]

    One-step diffusion distillation through score implicit matching.Advances in Neural Information Process- ing Systems, 37:115377–115408, 2025

    Weijian Luo, Zemin Huang, Zhengyang Geng, J Zico Kolter, and Guo-jun Qi. One-step diffusion distillation through score implicit matching.Advances in Neural Information Process- ing Systems, 37:115377–115408, 2025. 5, 1

  26. [34]

    Geodream: Disentan- gling 2d and geometric priors for high-fidelity and consistent 3d generation.arXiv preprint arXiv:2311.17971, 2023

    Baorui Ma, Haoge Deng, Junsheng Zhou, Yu-Shen Liu, Tiejun Huang, and Xinlong Wang. Geodream: Disentan- gling 2d and geometric priors for high-fidelity and consistent 3d generation.arXiv preprint arXiv:2311.17971, 2023. 2

  27. [35]

    X- mesh: Towards fast and accurate text-driven 3d stylization via dynamic textual guidance

    Yiwei Ma, Xiaoqing Zhang, Xiaoshuai Sun, Jiayi Ji, Haowei Wang, Guannan Jiang, Weilin Zhuang, and Rongrong Ji. X- mesh: Towards fast and accurate text-driven 3d stylization via dynamic textual guidance. InProceedings of the IEEE/CVF International Conference on Computer Vision,...

  28. [36]

    Latent-nerf for shape-guided generation of 3d shapes and textures.arXiv preprint arXiv:2211.07600,

    Gal Metzer, Elad Richardson, Or Patashnik, Raja Giryes, and Daniel Cohen-Or. Latent-nerf for shape-guided generation of 3d shapes and textures.arXiv preprint arXiv:2211.07600,

  29. [37]

    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

  30. [38]

    Clip-mesh: Generating textured meshes from text using pretrained image-text models

    Nasir Mohammad Khalid, Tianhao Xie, Eugene Belilovsky, and Tiberiu Popa. Clip-mesh: Generating textured meshes from text using pretrained image-text models. InSIGGRAPH Asia 2022 conference papers, pages 1–8, 2022. 2, 4

  31. [39]

    Ava: A large-scale database for aesthetic visual analysis

    Naila Murray, Luca Marchesotti, and Florent Perronnin. Ava: A large-scale database for aesthetic visual analysis. In2012 IEEE conference on computer vision and pattern recognition, pages 2408–2415. IEEE, 2012. 4

  32. [40]

    Diffusion models for adver- sarial purification.arXiv preprint arXiv:2205.07460, 2022

    Weili Nie, Brandon Guo, Yujia Huang, Chaowei Xiao, Arash Vahdat, and Anima Anandkumar. Diffusion models for adver- sarial purification.arXiv preprint arXiv:2205.07460, 2022. 2

  33. [41]

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

    Ben Poole, Ajay Jain, Jonathan T Barron, and Ben Mildenhall. Dreamfusion: Text-to-3d using 2d diffusion.arXiv preprint arXiv:2209.14988, 2022. 2, 3, 6, 7

  34. [42]

    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, p...

  35. [43]

    Hierarchical text-conditional image genera- tion with clip latents.arXiv preprint arXiv:2204.06125, 2022

    Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image genera- tion with clip latents.arXiv preprint arXiv:2204.06125, 2022. 2

  36. [44]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn 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, 3, 6, 8

  37. [45]

    Pho- torealistic text-to-image diffusion models with deep language understanding.Advances in neural information processing systems, 35:36479–36494, 2022

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

  38. [46]

    Clip-forge: Towards zero-shot text-to-shape generation

    Aditya Sanghi, Hang Chu, Joseph G Lambourne, Ye Wang, Chin-Yi Cheng, Marco Fumero, and Kamal Rahimi Malek- shan. Clip-forge: Towards zero-shot text-to-shape generation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18603–18613, 2022. 3

  39. [47]

    Clip-sculptor: Zero-shot generation of high-fidelity and diverse shapes from natural language

    Aditya Sanghi, Rao Fu, Vivian Liu, Karl DD Willis, Hooman Shayani, Amir H Khasahmadi, Srinath Sridhar, and Daniel Ritchie. Clip-sculptor: Zero-shot generation of high-fidelity and diverse shapes from natural language. InProceedings of the IEEE/CVF Conference on Computer Vision...

  40. [48]

    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. 2, 3, 6, 7

  41. [49]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. InInternational confer- ence on machine learning, pages 2256–2265. PMLR, 2015. 3

  42. [50]

    Denoising diffusion implicit models.arXiv preprint arXiv:2010.02502,

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

  43. [51]

    Score-based generative modeling through stochastic differential equations

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. arXiv preprint arXiv:2011.13456, 2020. 3

  44. [52]

    Sta- ble score distillation for high-quality 3d generation.arXiv preprint arXiv:2312.09305, 2023

    Boshi Tang, Jianan Wang, Zhiyong Wu, and Lei Zhang. Sta- ble score distillation for high-quality 3d generation.arXiv preprint arXiv:2312.09305, 2023. 2 12

  45. [53]

    Dreamgaussian: Generative gaussian splatting for effi- cient 3d content creation.arXiv preprint arXiv:2309.16653,

    Jiaxiang Tang, Jiawei Ren, Hang Zhou, Ziwei Liu, and Gang Zeng. Dreamgaussian: Generative gaussian splatting for effi- cient 3d content creation.arXiv preprint arXiv:2309.16653,

  46. [54]

    A connection between score matching and denoising autoencoders.Neural computation, 23(7):1661– 1674, 2011

    Pascal Vincent. A connection between score matching and denoising autoencoders.Neural computation, 23(7):1661– 1674, 2011. 1

  47. [55]

    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. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12619–12629, 2023. 3

  48. [56]

    Imagedream: Image-prompt multi-view diffusion for 3d generation.arXiv preprint arXiv:2312.02201, 2023

    Peng Wang and Yichun Shi. Imagedream: Image-prompt multi-view diffusion for 3d generation.arXiv preprint arXiv:2312.02201, 2023. 2

  49. [57]

    Taming mode collapse in score distillation for text-to-3d generation.arXiv preprint arXiv:2401.00909, 2023

    Peihao Wang, Dejia Xu, Zhiwen Fan, Dilin Wang, Sreyas Mohan, Forrest Iandola, Rakesh Ranjan, Yilei Li, Qiang Liu, Zhangyang Wang, et al. Taming mode collapse in score distillation for text-to-3d generation.arXiv preprint arXiv:2401.00909, 2023. 2

  50. [58]

    Prolificdreamer: High-fidelity and diverse text-to-3d generation with variational score distilla- tion.arXiv preprint arXiv:2305.16213, 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.arXiv preprint arXiv:2305.16213, 2023. 6, 7, 8

  51. [59]

    Prolificdreamer: High-fidelity and diverse text-to-3d generation with variational score distilla- tion.Advances in Neural Information Processing Systems, 36,

    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,

  52. [60]

    Consistent123: Improve consistency for one image to 3d object synthesis.arXiv preprint arXiv:2310.08092, 2023

    Haohan Weng, Tianyu Yang, Jianan Wang, Yu Li, Tong Zhang, CL Chen, and Lei Zhang. Consistent123: Improve consistency for one image to 3d object synthesis.arXiv preprint arXiv:2310.08092, 2023. 2

  53. [61]

    Gpt-4v (ision) is a human-aligned evaluator for text-to-3d generation

    Tong Wu, Guandao Yang, Zhibing Li, Kai Zhang, Ziwei Liu, Leonidas Guibas, Dahua Lin, and Gordon Wetzstein. Gpt-4v (ision) is a human-aligned evaluator for text-to-3d generation. InCVPR, pages 22227–22238, 2024. 5, 6, 7

  54. [62]

    Consistent3d: Towards consistent high-fidelity text-to-3d generation with deterministic sampling prior

    Zike Wu, Pan Zhou, Xuanyu Yi, Xiaoding Yuan, and Han- wang Zhang. Consistent3d: Towards consistent high-fidelity text-to-3d generation with deterministic sampling prior. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 9892–9902, 2024. 2

  55. [63]

    Imagereward: Learning and evaluating human preferences for text-to-image generation.Advances in Neural Information Processing Sys- tems, 36:15903–15935, 2023

    Jiazheng Xu, Xiao Liu, Yuchen Wu, Yuxuan Tong, Qinkai Li, Ming Ding, Jie Tang, and Yuxiao Dong. Imagereward: Learning and evaluating human preferences for text-to-image generation.Advances in Neural Information Processing Sys- tems, 36:15903–15935, 2023. 4, 3

  56. [64]

    Sa-solver: Stochastic adams solver for fast sampling of diffusion mod- els.Advances in Neural Information Processing Systems, 36: 77632–77674, 2023

    Shuchen Xue, Mingyang Yi, Weijian Luo, Shifeng Zhang, Jiacheng Sun, Zhenguo Li, and Zhi-Ming Ma. Sa-solver: Stochastic adams solver for fast sampling of diffusion mod- els.Advances in Neural Information Processing Systems, 36: 77632–77674, 2023. 3

  57. [65]

    Flow score distilla- tion for diverse text-to-3d generation, 2024

    Runjie Yan, Kailu Wu, and Kaisheng Ma. Flow score distilla- tion for diverse text-to-3d generation, 2024. 2

  58. [66]

    Consistent-1-to-3: Consistent image to 3d view syn- thesis via geometry-aware diffusion models.arXiv preprint arXiv:2310.03020, 2023

    Jianglong Ye, Peng Wang, Kejie Li, Yichun Shi, and Heng Wang. Consistent-1-to-3: Consistent image to 3d view syn- thesis via geometry-aware diffusion models.arXiv preprint arXiv:2310.03020, 2023. 2

  59. [67]

    Dreamre- ward: Text-to-3d generation with human preference, 2024

    Junliang Ye, Fangfu Liu, Qixiu Li, Zhengyi Wang, Yikai Wang, Xinzhou Wang, Yueqi Duan, and Jun Zhu. Dreamre- ward: Text-to-3d generation with human preference, 2024. 2, 4, 6, 7

  60. [68]

    Gaussiandreamer: Fast generation from text to 3d gaussians by bridging 2d and 3d diffusion models

    Taoran Yi, Jiemin Fang, Junjie Wang, Guanjun Wu, Lingxi Xie, Xiaopeng Zhang, Wenyu Liu, Qi Tian, and Xinggang Wang. Gaussiandreamer: Fast generation from text to 3d gaussians by bridging 2d and 3d diffusion models. InPro- ceedings of the IEEE/CVF Conference on Computer Vision ...

  61. [69]

    One-step diffusion with distribution matching distillation

    Tianwei Yin, Michaël Gharbi, Richard Zhang, Eli Shecht- man, Fredo Durand, William T Freeman, and Taesung Park. One-step diffusion with distribution matching distillation. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 6613–6623, 2024. 2

  62. [70]

    Text-to-3d with classifier score distillation

    Xin Yu, Yuan-Chen Guo, Yangguang Li, Ding Liang, Song- Hai Zhang, and XIAOJUAN QI. Text-to-3d with classifier score distillation. InThe Twelfth International Conference on Learning Representations, 2024. 2

  63. [71]

    Enhancing ad- versarial robustness via score-based optimization.Advances in Neural Information Processing Systems, 36:51810–51829,

    Boya Zhang, Weijian Luo, and Zhihua Zhang. Enhancing ad- versarial robustness via score-based optimization.Advances in Neural Information Processing Systems, 36:51810–51829,

  64. [72]

    Purify++: Im- proving diffusion-purification with advanced diffusion models and control of randomness.arXiv preprint arXiv:2310.18762,

    Boya Zhang, Weijian Luo, and Zhihua Zhang. Purify++: Im- proving diffusion-purification with advanced diffusion models and control of randomness.arXiv preprint arXiv:2310.18762,

  65. [73]

    Fast sampling of dif- fusion models with exponential integrator.arXiv preprint arXiv:2204.13902, 2022

    Qinsheng Zhang and Yongxin Chen. Fast sampling of dif- fusion models with exponential integrator.arXiv preprint arXiv:2204.13902, 2022. 3

  66. [74]

    Long and short guidance in score identity distilla- tion for one-step text-to-image generation.arXiv preprint arXiv:2406.01561, 2024

    Mingyuan Zhou, Zhendong Wang, Huangjie Zheng, and Hai Huang. Long and short guidance in score identity distilla- tion for one-step text-to-image generation.arXiv preprint arXiv:2406.01561, 2024. 5

  67. [75]

    Score identity distillation: Exponentially fast distillation of pretrained diffusion models for one-step generation

    Mingyuan Zhou, Huangjie Zheng, Zhendong Wang, Mingzhang Yin, and Hai Huang. Score identity distillation: Exponentially fast distillation of pretrained diffusion models for one-step generation. InForty-first International Confer- ence on Machine Learning, 2024. 2, 5, 1

  68. [76]

    HIFA: High- fidelity text-to-3d generation with advanced diffusion guid- ance

    Junzhe Zhu, Peiye Zhuang, and Sanmi Koyejo. HIFA: High- fidelity text-to-3d generation with advanced diffusion guid- ance. InThe Twelfth International Conference on Learning Representations, 2024. 2 13 Supplementary Material A. Derivation of the Score-based Loss Function The d...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.