Pith. sign in

REVIEW 4 major objections 6 minor 69 references

Perturb-and-Revise: Flexible 3D Editing with Generative Trajectories

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Perturb-and-Revise claims that pushing an optimized NeRF partway back to random initialization lets text-driven editing change pose, appearance, and geometry in ways previous methods cannot.

desk verdict Parameter-space perturbation is a genuinely useful trick for NeRF editing; the weak theory and eval hygiene don't sink it, but the adaptive-eta rule is fitted to the same benchmark it is evaluated on. read the letter →

arxiv 2412.05279 v2 pith:PVQ3AIL4 submitted 2024-12-06 cs.CV

Susung Hong , Johanna Karras , Ricardo Martin-Brualla , Ira Kemelmacher-Shlizerman This is my paper · ORCID
classification cs.CV
keywords neuralradiancefields3Deditingscoredistillationparameterperturbationdiffusionmodelsparticle-basedvariationalinferencemulti-viewconsistencytext-guided
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that a text-guided 3D editing system can handle large geometric changes—pose changes, added objects, and shape edits—if it first pushes an optimized neural radiance field (NeRF) partway back toward random initialization. The method, Perturb-and-Revise, interpolates the source NeRF's parameters with a random NeRF's parameters, chooses the interpolation amount from a short analysis of the local loss landscape, and then runs multi-view score distillation with an identity-preserving refinement term. The paper argues that the perturbed parameters leave the source scene's local minimum and follow a natural coarse-to-fine generative trajectory toward the edit prompt, so the final object stays recognizable while undergoing changes that earlier text-based editing methods could not produce.

What carries the argument

The load-bearing object is the edited NeRF treated as a particle in parameter space, moved by the Wasserstein gradient-flow ODE that underlies score distillation. The central intervention is parameter perturbation, a linear interpolation between the optimized parameters and a random initialization, with the interpolation weight $\eta$ selected adaptively from the measured loss decrease over a few simulated distillation steps. This is what lets the particle escape the source local minimum and re-enter the coarse-to-fine generative path. The second mechanism is the identity-preserving gradient, a weighted combination of L1 and perceptual losses between renders of the current and source NeRF, added in later refinement steps as a tug-of-war term that balances the edit direction against fidelity to the source object.

What would settle it

Render the perturbed NeRF at a moderate perturbation level (around $\eta = 0.3$) before any distillation runs; if the geometry is already unrecognizable or incoherent, the claimed re-entry into a generative trajectory fails. A second check is to compare edits obtained by interpolating toward two different random initializations with the same $\eta$: if the outputs diverge sharply, the method is sensitive to which random parameters are drawn.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that a fully optimized NeRF's parameters are not a fixed point for editing: linearly interpolating them toward a random initialization, $$\theta_{\mathrm{perturbed}} = (1-\eta)\,\theta_{\mathrm{src}} + \eta\,\theta_0, \quad \theta_0 \sim P(\Theta_0),$$ puts the particle in a less-optimized basin so that the score-distillation ODE can follow a generative path toward the target distribution of the edit prompt. The paper then shows that the needed perturbation level can be inferred from the local loss landscape rather than searched for, and that identity-preserving gradients added in later refinement steps pull the result back toward the source object. This combination is what the paper claims makes color, appearance, and geometry edits, including pose changes and object additions, possible from a single text prompt.

Load-bearing premise

The method depends on the assumption that mixing the optimized network weights with random weights acts like partially undoing the optimization, so the diffusion model sees a scene that is less finished but still structurally repairable rather than a broken jumble of geometry.

Editorial extensions

If this is right

  • A single text prompt can drive pose changes and object additions in NeRFs, not just recoloring or restyling.
  • Users do not need to describe the source object, and the perturbation level is picked automatically rather than by grid search.
  • Edits complete in roughly 1.5k distillation steps plus 1k refinement steps, about 7 minutes in the paper's setup, compared with about 26 minutes for full regeneration.
  • The same perturbation idea extends to real-scene editing and, in the paper's supplementary experiments, shows preliminary gains when applied to 3D Gaussian splatting.

Reading between the lines

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

  • If interpolation is truly partial de-optimization, the same principle should transfer to other representations trained from random initialization, such as voxel grids or meshes, with the loss-decrease rule setting $\eta$.
  • The effect may depend on the parameter geometry of the chosen NeRF implementation; a controlled comparison across architectures would separate a general principle from an architecture-specific accident.
  • Sequential edits could reuse the perturb-and-revise loop to preserve identity across multiple prompts, treating each edit as a fresh generative trajectory anchored to the original object.
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

4 major / 6 minor

Summary. The paper proposes Perturb-and-Revise (PnR), a training-free method for text-guided editing of NeRF scenes. The method first perturbs the optimized NeRF parameters by linearly interpolating them with a random initialization (Eq. 3), with the perturbation amount eta chosen adaptively via a loss-landscape heuristic (Sec. 4.2). It then runs a multi-view score-distillation ODE (Eq. 4) with timestep annealing, and finally applies identity-preserving gradients (IPG) to refine the result toward the source. The authors evaluate on synthetic fashion objects and Objaverse objects, comparing against SDS, PDS, and Instruct-NeRF2NeRF, with additional qualitative comparisons to Instruct-GS2GS and MVDream regeneration. The central claims are that parameter perturbation enables geometric edits (pose changes, object addition) that prior methods cannot handle, and that the adaptive eta selection and IPG refinement improve the quality and fidelity of the edits.

Significance. If the claimed results hold, the paper proposes a simple and effective mechanism for 3D editing that goes beyond texture/color changes and handles geometric edits such as pose changes and object addition. The idea of hijacking the generative ODE trajectory by parameter-space perturbation is original and conceptually appealing. The paper provides extensive qualitative results, an ablation on eta selection, and ablations for the IPG component. The main strengths are the breadth of edit types demonstrated and the efficiency of the approach (1.5k editing steps vs. 10k for regeneration). However, the quantitative evaluation is not yet fully convincing: only three quantitative baselines are compared, point estimates are reported without error bars, and the adaptive eta selection is tuned on the same evaluation benchmark, which weakens the claims of optimality and state-of-the-art performance.

major comments (4)
  1. [Sec. 4.2 / Supp. C / Table 3] The adaptive eta selection relies on two constants, Delta_min and eta_max, which are set based on observations on the same evaluation benchmark that is later used to claim that the adaptive scheme is near-optimal (Fig. 8). Specifically, Supp. C states that Delta_min is chosen from Table 3 (CLIP directional similarity on the same dataset) and eta_max from the success-rate plot in the main paper. This makes the comparison against fixed eta values circular. To support the claim, the authors should either tune on a disjoint development set and evaluate on a separate test set, or provide a sensitivity analysis showing that the final editing metrics are robust to the choice of these constants across a plausible range.
  2. [Sec. 4.1 / Supp. B] The theoretical justification for parameter perturbation only establishes that the distribution of theta_perturbed interpolates between a point mass and the initialization distribution. It does not show that a particular interpolated parameter vector corresponds to a semantically coherent, partially 'undone' version of the source scene, which is the premise of the 'generative trajectory' story. The authors should provide direct evidence, for example, renderings of the perturbed NeRF at eta values 0, 0.2, 0.4, 0.6, 0.8 before any editing steps, or a plot of the source-prompt loss along the interpolation path. Such evidence would clarify whether the perturbation produces gradual semantic degradation or an incoherent state that is then regenerated from scratch.
  3. [Sec. 5.2 / Table 1] The quantitative comparison lacks error bars, significance tests, and multiple seeds. Many of the reported differences are tiny: the average CLIP-Dir-Sim is 0.0565 for PnR versus 0.0557 for Instruct-NeRF2NeRF, and for ViT-B/16 Instruct-NeRF2NeRF is actually higher (0.0549 vs. 0.0534). Without variance estimates, these numbers do not support the claim that PnR 'achieves state-of-the-art results'. In addition, Instruct-GS2GS is included in the qualitative comparison (Fig. 2) but omitted from Table 1. The authors should report standard deviations over multiple runs/seeds and include Instruct-GS2GS in the quantitative evaluation, or moderate the strength of the claim.
  4. [Sec. 5.4 / Fig. 8] The y-axis of Fig. 8(a) is labeled 'percentage of successful experiments (without errors)', but the text never defines what constitutes a success or an error. This metric is used to justify eta_max=0.6 and to compare adaptive eta with fixed eta, so a precise operational definition (e.g., based on a CLIP threshold, manual inspection, or a convergence criterion) is necessary. Without it, the robustness claim is not fully verifiable.
minor comments (6)
  1. [Fig. 4] The figure layout is confusing: the horizontal axis appears to represent optimization steps with irregular spacing (100 to 900) and then 10000 for the regeneration row. Please label the axes clearly and specify whether the perturbation rows share the same step sequence.
  2. [Sec. 4.1] The statement 'adding more noise to the parameters is equivalent to undoing more of the optimization process' is a key intuition but is presented without support. Either provide a reference or an experiment (e.g., showing that perturbed renderings resemble earlier optimizer states) or soften the wording.
  3. [Sec. 5.1 / Figs. 2, 9] The asterisk in the figure captions says 'we use an identical update rule and schedule' for PDS and PnR, but the methods differ in parameter perturbation and IPG. Please clarify which components are identical (e.g., the score-distillation backbone and timestep schedule) to avoid confusion.
  4. [Sec. 4.3 / Supp. E] The values of lambda_L1 and lambda_p are only given in the supplementary material. Consider reporting them in the main text, since they are part of the method definition.
  5. [Table 2] The abbreviation CLIP-Dir-Con is not defined in the main text; define it when first used (it is defined in the supplement as CLIP directional consistency).
  6. [Abstract / conclusion] The phrase 'makes possible a variety of NeRF editing' is awkward; consider 'enables a variety of NeRF edit types' for clarity.

Circularity Check

1 steps flagged · score 4.0 of 10

The central perturbation-and-revise mechanism is independent, but the adaptive-η evaluation is partly fitted: its two constants are selected using the same benchmark curves that are then presented as evidence of near-optimality.

  1. fitted input called prediction [Supplementary Material Sec. C (Algorithms for Parameter Perturbation and Adaptive η Selection); main text Sec. 5.4 / Fig. 8]
    "Here, we have two parameters to control,∆min and ηmax. ∆min is set to 1000 based on observations that it achieves near-optimal CLIP directional similarity and CLIP directional consistency, as shown in Table 3.ηmax is set to 0.6 based on the finding that the percentage of successful experiments drops significantly when η exceeds 0.6, as shown in the main paper."

    The adaptive η-selection algorithm's free parameters are chosen by inspecting the same benchmark curves that Fig. 8 then uses to claim the adaptive method is near-optimal. ηmax = 0.6 is set from the observed success-rate drop for η > 0.6, and Δmin = 1000 is set to yield near-optimal CLIP directional similarity and consistency on the same evaluation set. Because DetermineEta caps η at ηmax and uses Δmin to place the operating point, the reported 'near-maximum' success rate and CLIP similarity of the adaptive selection partly restate the tuning criterion rather than independently predicting it. This is a fitted-input-called-prediction circularity.

full rationale

I walked the derivation chain from the generative ODE (Eq. 4) through the perturbation (Eq. 3), the adaptive η selection, and the IPG refinement. The core mechanism is not self-definitional: the score-distillation ODE is an established particle-based variational inference formulation, and the interpolation in Eq. 3 is presented as a modeling assumption with a distributional analysis in Supp. B that does not assume the editing result. The claim that parameter interpolation 'undoes optimization' is under-supported but is an unproven assumption, not a circular step. No load-bearing self-citation or uniqueness-imported-from-authors pattern appears; the cited particle/VSD works are external and the paper's own contribution is the perturbation and refinement recipe. The one substantive circularity is in the quantitative evaluation of the adaptive η method: the constants Δmin and ηmax are set by observing the same CLIP/success-rate benchmark that Fig. 8 uses to demonstrate near-optimal adaptive performance, so that particular 'prediction' is partly constructed by the hyperparameter choice. This affects a secondary claim about η selection, not the central flexible-editing result, so the overall circularity score is moderate rather than high.

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

The central claim rests on adopted background assumptions (score distillation as particle ODE, reliability of pretrained multi-view diffusion, geometric meaning of parameter interpolation) and on two heuristic free parameters in the adaptive eta rule. No new physical entities are introduced; 'generative trajectories' and 'identity-preserving gradient' are names for optimization terms, not new ontology.

free parameters (4)
  • eta_max = 0.6.
    Maximum perturbation in the adaptive eta rule; set based on observed drop in success rate for eta > 0.6 on the fashion editing benchmark (Supp. C, Fig. 8).
  • Delta_min = 1000.
    Minimum loss decrease threshold in the adaptive eta rule; chosen to achieve near-optimal CLIP directional similarity and consistency (Supp. C, Table 3).
  • IPG loss weights (lambda_L1, lambda_p) = lambda_L1 = 300.0, lambda_p = 30000.0 per Supp. E; Table 4 shows lambda_L1 = 10000-50000, lambda_p = 100-500.
    Weights for L1 and perceptual losses in the identity-preserving gradient, tuned via ablation on the same evaluation data (Supp. Table 4).
  • Timestep annealing schedule = U(0.75, 0.75) to U(0.02, 0.4) by 80% of editing steps.
    Hand-selected noise-level schedule, differing from the fixed U(0.02, 0.98) used by prior work (Supp. E).
assumptions (5)
  • domain assumption Score distillation updates can be interpreted as a Wasserstein gradient flow and generative ODE over NeRF parameters (Eq. 2).
    Adopted from VSD [62] and used throughout Sec. 3 and Sec. 4.1 to justify editing as following a generative trajectory. Not re-derived in this paper.
  • ad hoc to paper Linear interpolation between theta_src and a random initialization (Eq. 3) partially reverses optimization and creates a versatile initialization.
    Supp. B proves the interpolated distribution converges to a point mass or to P(Theta_0) at the extremes, but the claim that intermediate eta 'undoes optimization' is assumed, not derived.
  • ad hoc to paper The early-step loss decrease with the edit prompt is a proxy for the depth and volume of the basin of attraction and determines a good eta.
    Sec. 4.2 and Algorithm 2 use an inverted exponential decay of Delta_L with manually chosen Delta_min and eta_max; no theoretical justification is given.
  • domain assumption A pretrained multi-view diffusion model (MVDream, trained on Objaverse) provides multi-view consistent gradients sufficient to guide geometry edits.
    Eq. 4 relies on MVDream's multi-view consistency; the method inherits the model's compositionality and bias limitations, as acknowledged in Supp. G.
  • domain assumption InstantNGP's parameter space has a geometry in which interpolation mixes semantic levels in a controllable way.
    Sec. 4.1 and Supp. E apply interpolation to InstantNGP hash-grid parameters; the assumption that this behaves like coarse-to-fine reversion is implicit.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Perturb-and-Revise: Flexible 3D Editing with Generative Trajectories." pith.science (2026). https://pith.science/paper/PVQ3AIL4

@misc{pith2026241205279,
  author       = {Pith},
  title        = {Pith review of: Perturb-and-Revise: Flexible 3D Editing with Generative Trajectories},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PVQ3AIL4}},
  note         = {Machine review of arXiv:2412.05279}
}
read the original abstract

Recent advancements in text-based diffusion models have accelerated progress in 3D reconstruction and text-based 3D editing. Although existing 3D editing methods excel at modifying color, texture, and style, they struggle with extensive geometric or appearance changes, thus limiting their applications. To this end, we propose Perturb-and-Revise, which makes possible a variety of NeRF editing. First, we perturb the NeRF parameters with random initializations to create a versatile initialization. The level of perturbation is determined automatically through analysis of the local loss landscape. Then, we revise the edited NeRF via generative trajectories. Combined with the generative process, we impose identity-preserving gradients to refine the edited NeRF. Extensive experiments demonstrate that Perturb-and-Revise facilitates flexible, effective, and consistent editing of color, appearance, and geometry in 3D. For 360{\deg} results, please visit our project page: https://susunghong.github.io/Perturb-and-Revise.

Figures

Figures reproduced from arXiv: 2412.05279 by the authors.

Figure 1
Figure 1. Perturb-and-Revise takes a source NeRF and an edit prompt as input and produces the edited result through: (1) versatile ini￾tialization via parameter perturbation, (2) multi-view consistent score distillation, and (3) refinement with the identity-preserving gradient. Abstract Recent advancements in text-based diffusion models have accelerated progress in 3D reconstruction and text￾based 3D editing. Although existin… view at source ↗
Figure 2
Figure 2. We qualitatively compare our method to Instruct-NeRF2NeRF [ [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Conceptual figure. The target distribution in the figures [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: Effect of parameter perturbation. In this example, we [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Baseline comparisons with a wide range of edits. We compare our method with Score Distillation Sampling (SDS) [ [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Baseline comparisons of editing various general 3D objects from the Objaverse dataset [ [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: Comparisons with Instruct-NeRF2NeRF (dataset update) [ [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Ablation study on the selection of η. (a) and (b) show the CLIP direction similarity and the percentage of successful experiments (without errors) for different η values, respectively. When averaged across all types of edits, our adaptive method achieves near-maximum p…
Figure 9
Figure 9. Figure 9: Real scene editing results. An asterisk (*) indicates we [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: Original scene, edited scene, and image-level gradients are shown at 0, 2500, and 5000 optimization steps. We can see that the [PITH_FULL_IMAGE:figures/full_fig_p015_10.png]
Figure 11
Figure 11. Figure 11: Effects of IPG refinement steps. IPG refinement steps restore changed attributes that were not explicitly mentioned in the edit [PITH_FULL_IMAGE:figures/full_fig_p016_11.png]
Figure 12
Figure 12. Figure 12: 3DGS experiment. The perturbation ap￾proach improves the depth map and overall geometry of the resulting object. Extension to 3DGS. Adapting our method to 3DGS [22] presents unique challenges, particularly in addressing adaptive densification and the ill-defined inter…
Figure 13
Figure 13. Figure 13: Additional comparisons of fashion object editing with Score Distillation [ [PITH_FULL_IMAGE:figures/full_fig_p017_13.png]
Figure 14
Figure 14. Figure 14: Additional comparisons of general object editing with Score Distillation [ [PITH_FULL_IMAGE:figures/full_fig_p018_14.png]
Figure 15
Figure 15. Figure 15: Additional visualizations for the selection of [PITH_FULL_IMAGE:figures/full_fig_p019_15.png]
Figure 16
Figure 16. Figure 16: Failure cases. F. Computational efficiency Our approach requires approximately 7 minutes with IPG and 4 minutes without, to produce meaningful results. This is faster than the 13 minutes needed for Instruct-NeRF2NeRF. We attribute this to Instruct-NeRF2NeRF requiring …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

69 extracted references · 29 canonical work pages

  1. [1]

    Align your latents: High-resolution video synthesis with la- tent diffusion models

    Andreas Blattmann, Robin Rombach, Huan Ling, Tim Dock- horn, Seung Wook Kim, Sanja Fidler, and Karsten Kreis. Align your latents: High-resolution video synthesis with la- tent diffusion models. CVPR, 2023. 2

    work page 2023

  2. [2]

    In- structpix2pix: Learning to follow image editing instructions

    Tim Brooks, Aleksander Holynski, and Alexei A Efros. In- structpix2pix: Learning to follow image editing instructions. CVPR, 2023. 1, 2, 3

    work page 2023

  3. [3]

    A unified particle-optimization frame- work for scalable bayesian sampling

    Changyou Chen, Ruiyi Zhang, Wenlin Wang, Bai Li, and Liqun Chen. A unified particle-optimization frame- work for scalable bayesian sampling. arXiv preprint arXiv:1805.11659, 2018. 3, 12

    arXiv 2018

  4. [4]

    Objaverse: A universe of annotated 3d objects

    Matt Deitke, Dustin Schwenk, Jordi Salvador, Luca Weihs, Oscar Michel, Eli VanderBilt, Ludwig Schmidt, Kiana Ehsani, Aniruddha Kembhavi, and Ali Farhadi. Objaverse: A universe of annotated 3d objects. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13142–13153, 2023. 2, 3, 6, 7

    work page 2023

  5. [5]

    Particle- based variational inference with preconditioned functional gradient flow

    Hanze Dong, Xi Wang, Yong Lin, and Tong Zhang. Particle- based variational inference with preconditioned functional gradient flow. arXiv preprint arXiv:2211.13954 , 2022. 3, 12

    arXiv 2022

  6. [6]

    From data to functa: Your data point is a function and you should treat it like one.ICML, 2022

    Emilien Dupont, Hyunjik Kim, SM Eslami, Danilo Rezende, and Dan Rosenbaum. From data to functa: Your data point is a function and you should treat it like one.ICML, 2022. 2, 4

    work page 2022

  7. [7]

    Taming transformers for high-resolution image synthesis

    Patrick Esser, Robin Rombach, and Bjorn Ommer. Taming transformers for high-resolution image synthesis. In CVPR, pages 12873–12883, 2021. 2

    2021

  8. [8]

    Instruct-nerf2nerf: Editing 3d scenes with instructions

    Ayaan Haque, Matthew Tancik, Alexei A Efros, Alek- sander Holynski, and Angjoo Kanazawa. Instruct-nerf2nerf: Editing 3d scenes with instructions. arXiv preprint arXiv:2303.12789, 2023. 1, 2, 3, 5, 6, 7, 8, 15, 17, 18

    arXiv 2023

Show all 69 references
  1. [9]

    Denoising dif- fusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models. NeurIPS, 33:6840–6851, 2020. 2

    2020

  2. [10]

    Imagen video: High definition video generation with diffusion mod- els

    Jonathan Ho, William Chan, Chitwan Saharia, Jay Whang, Ruiqi Gao, Alexey Gritsenko, Diederik P Kingma, Ben Poole, Mohammad Norouzi, David J Fleet, et al. Imagen video: High definition video generation with diffusion mod- els. arXiv preprint arXiv:2210.02303, 2022. 2

    arXiv

  3. [11]

    Smoothed energy guidance: Guiding dif- fusion models with reduced energy curvature of attention

    Susung Hong. Smoothed energy guidance: Guiding dif- fusion models with reduced energy curvature of attention. arXiv preprint arXiv:2408.00760, 2024. 2

    2024 arXiv

  4. [12]

    Debi- asing scores and prompts of 2d diffusion for view-consistent text-to-3d generation

    Susung Hong, Donghoon Ahn, and Seungryong Kim. Debi- asing scores and prompts of 2d diffusion for view-consistent text-to-3d generation. Advances in Neural Information Pro- cessing Systems, 36:11970–11987, 2023. 2, 3, 15

    2023

  5. [13]

    Improving sample quality of diffusion models us- ing self-attention guidance

    Susung Hong, Gyuseong Lee, Wooseok Jang, and Seungry- ong Kim. Improving sample quality of diffusion models us- ing self-attention guidance. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 7462– 7471, 2023. 2

    2023

  6. [14]

    Direct2v: Large language mod- els are frame-level directors for zero-shot text-to-video gen- eration

    Susung Hong, Junyoung Seo, Heeseong Shin, Sunghwan Hong, and Seungryong Kim. Direct2v: Large language mod- els are frame-level directors for zero-shot text-to-video gen- eration. arXiv preprint arXiv:2305.14330, 2023. 2

    2023 arXiv

  7. [15]

    Free-bloom: Zero-shot text-to-video generator with llm director and ldm animator

    Hanzhuo Huang, Yufan Feng, Cheng Shi, Lan Xu, Jingyi Yu, and Sibei Yang. Free-bloom: Zero-shot text-to-video generator with llm director and ldm animator. arXiv preprint arXiv:2309.14494, 2023. 2

    2023 arXiv

  8. [16]

    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 Conference on Computer Vision and Pattern Recognition, pages 12469– 12478, 2024. 3

    2024

  9. [17]

    Estimation of non- normalized statistical models by score matching

    Aapo Hyv ¨arinen and Peter Dayan. Estimation of non- normalized statistical models by score matching. Journal of Machine Learning Research, 6(4), 2005. 3

    2005

  10. [18]

    Zero-shot text-guided object gen- eration with dream fields

    Ajay Jain, Ben Mildenhall, Jonathan T Barron, Pieter Abbeel, and Ben Poole. Zero-shot text-guided object gen- eration with dream fields. In CVPR, pages 867–876, 2022. 1

    2022

  11. [19]

    Percep- tual losses for real-time style transfer and super-resolution

    Justin Johnson, Alexandre Alahi, and Li Fei-Fei. Percep- tual losses for real-time style transfer and super-resolution. In Computer Vision–ECCV 2016: 14th European Confer- ence, Amsterdam, The Netherlands, October 11-14, 2016, Proceedings, Part II 14, pages 694–711. Springer, 2016. 5

    2016

  12. [20]

    Dreampose: Fashion image-to-video synthesis via stable diffusion

    Johanna Karras, Aleksander Holynski, Ting-Chun Wang, and Ira Kemelmacher-Shlizerman. Dreampose: Fashion image-to-video synthesis via stable diffusion. arXiv preprint arXiv:2304.06025, 2023. 2

    2023 arXiv

  13. [21]

    Elucidating the design space of diffusion-based generative models

    Tero Karras, Miika Aittala, Timo Aila, and Samuli Laine. Elucidating the design space of diffusion-based generative models. NeurIPS, 2022. 2, 3

    2022

  14. [22]

    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 Transactions on Graphics (ToG), 42(4):1–14, 2023. 3, 16

    2023

  15. [23]

    Text2video-zero: Text-to- image diffusion models are zero-shot video generators.arXiv preprint arXiv:2303.13439, 2023

    Levon Khachatryan, Andranik Movsisyan, Vahram Tade- vosyan, Roberto Henschel, Zhangyang Wang, Shant Navasardyan, and Humphrey Shi. Text2video-zero: Text-to- image diffusion models are zero-shot video generators.arXiv preprint arXiv:2303.13439, 2023. 2

    2023 arXiv

  16. [24]

    Decomposing nerf for editing via feature field distil- lation

    Sosuke Kobayashi, Eiichi Matsumoto, and Vincent Sitz- mann. Decomposing nerf for editing via feature field distil- lation. Advances in Neural Information Processing Systems, 35:23311–23330, 2022. 3

    2022

  17. [25]

    Posterior dis- tillation sampling

    Juil Koo, Chanho Park, and Minhyuk Sung. Posterior dis- tillation sampling. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 13352–13361, 2024. 1, 2, 3, 5, 6, 7, 8, 15, 17, 18

    2024

  18. [26]

    Imagenet classification with deep convolutional neural net- works

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural net- works. Communications of the ACM, 60(6):84–90, 2017. 8

    2017

  19. [27]

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

    Yixun Liang, Xin Yang, Jiantao Lin, Haodong Li, Xiao- gang Xu, and Yingcong Chen. Luciddreamer: Towards high- fidelity text-to-3d generation via interval score matching. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 6517–6526, 2024. 16

    2024

  20. [28]

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

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

    2023

  21. [29]

    Stein variational gradient de- scent: A general purpose bayesian inference algorithm

    Qiang Liu and Dilin Wang. Stein variational gradient de- scent: A general purpose bayesian inference algorithm. Ad- vances in neural information processing systems , 29, 2016. 3

    2016

  22. [30]

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

    Ruoshi Liu, Rundi Wu, Basile Van Hoorick, Pavel Tok- makov, Sergey Zakharov, and Carl V ondrick. Zero-1-to- 3: Zero-shot one image to 3d object. arXiv preprint arXiv:2303.11328, 2023. 2, 3

    2023 arXiv

  23. [31]

    Editing condi- tional radiance fields

    Steven Liu, Xiuming Zhang, Zhoutong Zhang, Richard Zhang, Jun-Yan Zhu, and Bryan Russell. Editing condi- tional radiance fields. In Proceedings of the IEEE/CVF inter- national conference on computer vision , pages 5773–5783,

  24. [32]

    Geometry in sampling methods: A review on manifold mcmc and particle-based variational inference methods

    Chang Liua and Jun Zhub. Geometry in sampling methods: A review on manifold mcmc and particle-based variational inference methods. Advancements in Bayesian Methods and Implementations, 47:239, 2022. 3, 12

    2022

  25. [33]

    Nerf in the wild: Neural radiance fields for uncon- strained photo collections

    Ricardo Martin-Brualla, Noha Radwan, Mehdi SM Sajjadi, Jonathan T Barron, Alexey Dosovitskiy, and Daniel Duck- worth. Nerf in the wild: Neural radiance fields for uncon- strained photo collections. In Proceedings of the IEEE/CVF conference on computer vision and pattern recogn...

    2021

  26. [34]

    Latent-nerf for shape-guided generation of 3d shapes and textures

    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,

    arXiv

  27. [35]

    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. 1, 3, 5

    2021

  28. [36]

    T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models

    Chong Mou, Xintao Wang, Liangbin Xie, Jian Zhang, Zhon- gang Qi, Ying Shan, and Xiaohu Qie. T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models. arXiv preprint arXiv:2302.08453, 2023. 2

    2023 arXiv

  29. [37]

    Instant neural graphics primitives with a mul- tiresolution hash encoding

    Thomas M ¨uller, Alex Evans, Christoph Schied, and Alexan- der Keller. Instant neural graphics primitives with a mul- tiresolution hash encoding. ACM Transactions on Graphics (ToG), 41(4):1–15, 2022. 3, 7, 17

    2022

  30. [38]

    Extracting triangular 3d models, materials, and lighting from images

    Jacob Munkberg, Jon Hasselgren, Tianchang Shen, Jun Gao, Wenzheng Chen, Alex Evans, Thomas M¨uller, and Sanja Fi- dler. Extracting triangular 3d models, materials, and lighting from images. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , ...

    2022

  31. [39]

    Neural scene graphs for dynamic scenes

    Julian Ost, Fahim Mannan, Nils Thuerey, Julian Knodt, and Felix Heide. Neural scene graphs for dynamic scenes. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 2856–2865, 2021. 3

    2021

  32. [40]

    Nerfies: Deformable neural radiance fields

    Keunhong Park, Utkarsh Sinha, Jonathan T Barron, Sofien Bouaziz, Dan B Goldman, Steven M Seitz, and Ricardo Martin-Brualla. Nerfies: Deformable neural radiance fields. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 5865–5874, 2021. 3

    2021

  33. [41]

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

    Ben Poole, Ajay Jain, Jonathan T Barron, and Ben Milden- hall. Dreamfusion: Text-to-3d using 2d diffusion. arXiv preprint arXiv:2209.14988, 2022. 1, 2, 3, 5, 6, 12, 17, 18

    2022 arXiv

  34. [42]

    Fatezero: Fus- ing attentions for zero-shot text-based video editing

    Chenyang Qi, Xiaodong Cun, Yong Zhang, Chenyang Lei, Xintao Wang, Ying Shan, and Qifeng Chen. Fatezero: Fus- ing attentions for zero-shot text-based video editing. arXiv preprint arXiv:2303.09535, 2023. 2

    2023 arXiv

  35. [43]

    Learn- ing transferable visual models from natural language super- vision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learn- ing transferable visual models from natural language super- vision. In ICML, pages 8748–8763. PMLR, 2021. 3, 8

    2021

  36. [44]

    High-resolution image syn- thesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj¨orn Ommer. High-resolution image syn- thesis with latent diffusion models. In CVPR, pages 10684– 10695, 2022. 1, 2, 3, 17, 19

    2022

  37. [45]

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

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily Denton, Seyed Kamyar Seyed Ghasemipour, Burcu Karagol Ayan, S Sara Mahdavi, Rapha Gontijo Lopes, et al. Photorealistic text-to-image diffusion models with deep language understanding. arXiv preprint arXiv...

    2022 arXiv

  38. [46]

    Laion-5b: An open large-scale dataset for training next generation image-text models

    Christoph Schuhmann, Romain Beaumont, Richard Vencu, Cade Gordon, Ross Wightman, Mehdi Cherti, Theo Coombes, Aarush Katta, Clayton Mullis, Mitchell Worts- man, et al. Laion-5b: An open large-scale dataset for training next generation image-text models. Advances in Neural In- f...

    2022

  39. [47]

    Retrieval-augmented score distillation for text-to-3d gener- ation

    Junyoung Seo, Susung Hong, Wooseok Jang, In `es Hyeonsu Kim, Minseop Kwak, Doyup Lee, and Seungryong Kim. Retrieval-augmented score distillation for text-to-3d gener- ation. arXiv preprint arXiv:2402.02972, 2024. 2

    2024 arXiv

  40. [48]

    Mvdream: Multi-view diffusion for 3d gen- eration

    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, 5, 6, 7, 16, 19

    2023 arXiv

  41. [49]

    Very deep convo- lutional networks for large-scale image recognition

    Karen Simonyan and Andrew Zisserman. Very deep convo- lutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556, 2014. 8

    2014 arXiv

  42. [50]

    Denois- ing diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denois- ing diffusion implicit models. In ICLR, 2021. 2

    2021

  43. [51]

    Generative modeling by esti- mating gradients of the data distribution

    Yang Song and Stefano Ermon. Generative modeling by esti- mating gradients of the data distribution. NeurIPS, 32, 2019. 2

    2019

  44. [52]

    Score-based generative modeling through stochastic differential equa- tions

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equa- tions. In ICLR, 2020. 2

    2020

  45. [53]

    Nerv: Neural reflectance and visibility fields for relighting and view synthesis

    Pratul P Srinivasan, Boyang Deng, Xiuming Zhang, Matthew Tancik, Ben Mildenhall, and Jonathan T Barron. Nerv: Neural reflectance and visibility fields for relighting and view synthesis. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pag...

    2021

  46. [54]

    Nerfstudio: A modular framework for neural radiance field development

    Matthew Tancik, Ethan Weber, Evonne Ng, Ruilong Li, Brent Yi, Terrance Wang, Alexander Kristoffersen, Jake Austin, Kamyar Salahi, Abhik Ahuja, et al. Nerfstudio: A modular framework for neural radiance field development. In ACM SIGGRAPH 2023 Conference Proceedings , pages 1–12...

    2023

  47. [55]

    Neural feature fusion fields: 3d distillation of self-supervised 2d image representations

    Vadim Tschernezki, Iro Laina, Diane Larlus, and Andrea Vedaldi. Neural feature fusion fields: 3d distillation of self-supervised 2d image representations. In 2022 Inter- national Conference on 3D Vision (3DV) , pages 443–453. IEEE, 2022. 3

    2022

  48. [56]

    Instruct-gs2gs: Editing 3d gaussian splats with instructions, 2024

    Cyrus Vachha and Ayaan Haque. Instruct-gs2gs: Editing 3d gaussian splats with instructions, 2024. 2

    2024

  49. [57]

    Ref-nerf: Struc- tured view-dependent appearance for neural radiance fields

    Dor Verbin, Peter Hedman, Ben Mildenhall, Todd Zickler, Jonathan T Barron, and Pratul P Srinivasan. Ref-nerf: Struc- tured view-dependent appearance for neural radiance fields. In 2022 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), pages 5481–5490. IE...

    2022

  50. [58]

    Clip-nerf: Text-and-image driven manip- ulation of neural radiance fields

    Can Wang, Menglei Chai, Mingming He, Dongdong Chen, and Jing Liao. Clip-nerf: Text-and-image driven manip- ulation of neural radiance fields. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3835–3844, 2022. 3

    2022

  51. [59]

    Nerf-art: Text-driven neural radiance fields stylization

    Can Wang, Ruixiang Jiang, Menglei Chai, Mingming He, Dongdong Chen, and Jing Liao. Nerf-art: Text-driven neural radiance fields stylization. IEEE Transactions on Visualiza- tion and Computer Graphics, 2023. 3

    2023

  52. [60]

    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. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12619–12629, 2023. 1, 2, 3

    2023

  53. [61]

    Func- tion space particle optimization for bayesian neural net- works

    Ziyu Wang, Tongzheng Ren, Jun Zhu, and Bo Zhang. Func- tion space particle optimization for bayesian neural net- works. arXiv preprint arXiv:1902.09754, 2019. 3, 12

    1902 arXiv

  54. [62]

    Prolificdreamer: High-fidelity and diverse text-to-3d generation with variational score distilla- tion

    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, 2024. 2, 3, 4, 6, 12

    2024

  55. [63]

    A latent space of stochastic diffusion models for zero-shot image editing and guidance

    Chen Henry Wu and Fernando De la Torre. A latent space of stochastic diffusion models for zero-shot image editing and guidance. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 7378–7387, 2023. 3

    2023

  56. [64]

    Tune-a-video: One-shot tuning of image diffusion models for text-to-video generation

    Jay Zhangjie Wu, Yixiao Ge, Xintao Wang, Weixian Lei, Yuchao Gu, Wynne Hsu, Ying Shan, Xiaohu Qie, and Mike Zheng Shou. Tune-a-video: One-shot tuning of image diffusion models for text-to-video generation. arXiv preprint arXiv:2212.11565, 2022. 2

    2022 arXiv

  57. [65]

    Unsu- pervised discovery of object radiance fields

    Hong-Xing Yu, Leonidas J Guibas, and Jiajun Wu. Unsu- pervised discovery of object radiance fields. arXiv preprint arXiv:2107.07905, 2021. 3

    2021 arXiv

  58. [66]

    Adding conditional control to text-to-image diffusion models

    Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3836–3847, 2023. 2

    2023

  59. [67]

    The unreasonable effectiveness of deep features as a perceptual metric

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shecht- man, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. In Proceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 586–595, 2018. 8

    2018

  60. [68]

    Tryondiffusion: A tale of two unets

    Luyang Zhu, Dawei Yang, Tyler Zhu, Fitsum Reda, William Chan, Chitwan Saharia, Mohammad Norouzi, and Ira Kemelmacher-Shlizerman. Tryondiffusion: A tale of two unets. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , pages 4606–4615,

  61. [2023]

    … raising his arms

    2 Perturb-and-Revise: Flexible 3D Editing with Generative Trajectories Supplementary Material A. Score Distillation as Particle-Based Variational Inference Our parameter perturbation and identity gradients build upon the mathematical intuition of the variational score distilla...

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.