Pith. sign in

REVIEW 2 major objections 6 minor 35 references

GeoCache: Training-Free Acceleration of Multi-View Texture Diffusion via Geometric Delta Transport

T0 review · 2 major / 6 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read The paper claims that geometrically corresponding surface points share transferable denoising evolution in multi-view texture diffusion, and that transporting per-step clean-signal changes across views yields a stronger speed-fidelity…

desk verdict A genuinely new cross-view caching idea with a strong evaluation, but the claimed "same transferred configuration" portability is contradicted by the paper's own Table 1 and needs a fix before publication. read the letter →

arxiv 2608.13255 v1 pith:RUYIBIN2 submitted 2026-08-13 cs.CV cs.AI

Haotang Li , Zhenyu Qi , Shaohan Henry Wang , Kebin Peng , Yutong Zhao , Zi Wang , Bo Liu , Huanrui Yang
show 1 more author
Sen He
This is my paper · ORCID
classification cs.CVcs.AI
keywords training-freeaccelerationmulti-viewdiffusiontexturegenerationgeometriccorrespondencecachedeltatransportdenoising3D
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 argues that the redundancy in multi-view texture diffusion is geometric, not just temporal: the same surface point seen from different views changes in a transferable way in the predicted clean-image space, even though intermediate features stay view-specific. It introduces GeoCache, a training-free plugin that runs the denoiser on a rotating subset of anchor views at each step and transports their per-step change ($\Delta x_0$) to every other view along geometry-based correspondences, preserving each view's own state. Periodic full-view refreshes and sampler-consistent reconstruction keep the error bounded and the denoising trajectory valid. On Hunyuan3D-2.1 it reports a $2.21\times$ denoiser-loop speedup at the lowest MV-LPIPS among tested methods above $2\times$, and it transfers unchanged to SyncMVD and MVPainter. A sympathetic reader would care because painting is the dominant cost of 3D asset generation, and existing step caches degrade cross-view consistency, whereas this method preserves it.

What carries the argument

The central object is the geometric delta-transport rule, Eq. (1): $x_0^{(v)}(t) = x_0^{(v)}(t-1) + \mathcal{G}_{A\to v}[\Delta x_0^{(A)}(t)]$, where $\mathcal{G}_{A\to v}$ is a row-stochastic sparse gather that maps tokens in anchor views to corresponding tokens in target view $v$ using precomputed position maps (K nearest taps within 1% of the bounding-box diagonal, area-weighted). It operates on the first difference of the predicted clean signal, not on features or full states, which lets each target view keep its own content and noise while inheriting the shared surface's denoising evolution. Around this sit three supporting mechanisms: batch-sliced anchor forwards that preserve each retained view's positional identity in multi-view attention, periodic full-view refresh steps that re-ground every view, and sampler-consistent reconstruction that converts transported $x_0$ back into $\epsilon$, $v$, or flow velocity so the solver's history stays valid. The method requires no retraining and consumes position maps the texturing pipeline already computes.

What would settle it

Take an asset with a region seen by only one anchor view, run GeoCache, and compare the transported $x_0$ of a non-anchor view at a cached step with the genuine $x_0$ from a full forward; if the error in that region exceeds the gap between GeoCache's reported 0.029 MV-LPIPS and the value-copy failure at 0.101, the correspondence-based transport is not carrying the shared evolution.

Watch

Extended reading notes

Core claim

GeoCache's central claim is that in geometry-conditioned multi-view texture diffusion, the predicted clean signal $x_0$ at a denoising step evolves in a way that is shared across views for the same surface point, so only a few anchor views need to run the denoiser at each step. The per-step change $\Delta x_0$ of the anchors is transported through a fixed sparse linear gather operator built from position maps and added to each non-anchor view's own previous $x_0$, rather than overwriting it; unmatched tokens (disocclusions, background) simply keep their state until the next full refresh. Four full-view steps per trajectory bound the accumulated drift, and converting the transported $x_0$ back into the sampler's native parameterization keeps the multistep history consistent. The paper reports that on Hunyuan3D-2.1 this yields a $2.21\times$ denoiser-loop speedup at MV-LPIPS $0.0293$ and MV-PSNR $33.60$ dB, the best fidelity of any method tested above $2\times$, with the same schedule leading fastest and leanest on SyncMVD and best fidelity and lowest FLOPs among accelerated methods on MVPainter.

Load-bearing premise

GeoCache assumes that running the denoiser on only the anchor views at a cached step, which removes the other views from multi-view attention, leaves the anchors' predicted per-step change close enough to the full-context value that transporting it to all views does not corrupt the result; the schedule that bounds the accumulated error is chosen empirically rather than derived.

Editorial extensions

If this is right

  • If GeoCache's claim holds, multi-view texture diffusion can be accelerated by omitting whole-view forwards rather than skipping denoising steps, so every step of the trajectory is still computed for at least the anchors and cross-view harmonization is never entirely absent.
  • The same transported-increment rule should transfer to any batched multi-view pipeline with geometric correspondence, since the method only changes the definition of an anchor (batch row, tile row, or token-frame slice), which the paper demonstrates by carrying the Hunyuan-tuned schedule unchanged to SyncMVD and MVPainter.
  • Because GeoCache composes geometric caching with mild step reduction, it can be stacked with faster samplers or distillation rather than competing with them, extending the speed-fidelity frontier beyond what either axis alone reaches.
  • The reported slope advantage (each additional $0.1\times$ of speed costs GeoCache about $+3.1\%$ MV-LPIPS, versus $+12.4$ to $+33.5\%$ for temporal caches) implies that the gap in fidelity grows as targets move past $2\times$, making the geometric axis the cheaper way to push speed.
  • On cost centers where the denoiser loop dominates (production-resolution assets), a $2.21\times$ loop speedup translates into a larger end-to-end saving, though at the default $6\times512^2$ resolution the measured end-to-end gain is $1.07\times$.

Reading between the lines

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

  • Inference: because GeoCache's saving is along the view axis, the same plugin should combine with temporal caches such as TeaCache and MagCache; the paper does not test this stacking, but the two reuse axes are orthogonal in the sense that one omits views and the other omits steps.
  • Inference: the delta-transport update resembles an explicit Euler step on a per-surface-point increment carried across views; this suggests a stability analysis could be formalized by bounding the spectral norm of $\mathcal{G}_{A\to v}$ and the deviation between anchor-only and full multi-view attention, turning the empirical refresh schedule into a derived one.
  • Inference: adaptive anchor selection based on visibility would likely extend the method to sparsely observed regions; the paper explicitly leaves this as future work, but the correspondence operator already carries visibility information through the tolerance test.
  • Inference: the same geometric axis could apply to video diffusion with a fixed multi-view rig or a moving camera, where scene points are observed from multiple frames; the method's 'views' need not be static renders of one object.
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

2 major / 6 minor

Summary. The paper proposes GeoCache, a training-free plugin that accelerates multi-view texture diffusion by running the denoiser on a rotating subset of anchor views at each step and transporting the anchors' per-step change in the predicted clean signal (Delta x0) to non-anchor views through a fixed correspondence gather built from rendered position maps. Periodic full-view refreshes bound drift, and the transported x0 is converted to the sampler's native parameterization to preserve trajectory consistency. The authors evaluate on Hunyuan3D-2.1, SyncMVD, and MVPainter against temporal caches (TeaCache, MagCache, FORA, TaylorSeer) and step reduction, reporting that GeoCache achieves the best fidelity among methods above 2x denoiser speedup on Hunyuan (2.21x, MV-LPIPS 0.0293, MV-PSNR 33.60 dB), the highest speedup on SyncMVD, and the best fidelity/cost on MVPainter. Ablations identify delta transport, refresh placement, and correspondence as load-bearing components.

Significance. If the reported results hold, GeoCache identifies and exploits a genuinely complementary redundancy axis, cross-view geometric evolution of x0, that step caches do not use, with a simple and parameter-free core operation (Eq. 1) whose only adjusted knobs are schedule hyperparameters. The evaluation is unusually thorough: three backbones with different samplers and view counts, matched seeds and protocols, per-asset standard deviations, component ablations, interaction cells, and honest negatives (K=1 taps, confidence threshold as a no-op). The main correction needed is the unsupported 'same transferred configuration' claim for SyncMVD, which Table 1 contradicts; once the portability evidence is either supplied or the claim is softened, the core speed-fidelity contribution is defensible and relevant to the multi-view texturing community.

major comments (2)
  1. [Abstract and Section 4.2 vs Table 1] The abstract and Section 4.2 claim that the Hunyuan-tuned GeoCache schedule transfers unchanged to SyncMVD and leads every metric above 2x, but Table 1 lists SyncMVD configurations a=2, E=3, S=20 and a=2, E=2, S=20, whereas the Hunyuan headline is a=2, E=5, S=10. No SyncMVD row with E=5 and S=10 is reported, so the 'transfers unchanged' claim is not supported by the data in the manuscript. Please report the exact transferred configuration on SyncMVD or revise the portability claim; as written, the schedule-sensitivity question remains open.
  2. [Section 3.2 vs Section 4.1 and Table 1] Section 3.2 states that the Hunyuan operating point runs 10 UniPC steps, 'four of them full', and describes a two-step head, one mid-trajectory refresh, and a tail step, while Table 1 labels the same operating point as a=2, E=5, S=10, with E defined in Section 4.1 as the number of full-view refresh steps placed at the end of the trajectory. These statements are inconsistent about both the number and placement of full steps. Please specify the complete schedule, including which steps are full, so that the headline result is reproducible.
minor comments (6)
  1. [Section 4.2 and Figure 4] The text and Figure 4 refer to 'the best temporal cache' per backbone without naming which baseline appears in each panel; since Table 1 shows the best temporal method differs across backbones (MagCache on Hunyuan, FORA on SyncMVD, and step reduction on MVPainter), please label each panel's baseline in the caption or the text.
  2. [Section 3.2] The reconstruction formulas use sqrt(alpha_bar_t) without defining alpha_bar_t; please add the missing notation definitions or a pointer to the sampler's convention.
  3. [Table 2] The note 'bSeam below the stock reference of 0.061 signals blur' is confusing because some cells with below-stock seam values are discussed positively in the text; please clarify the intended interpretation of the superscript b.
  4. [Affiliation] There is a typo in the first affiliation: 'Eletrical' should be 'Electrical'.
  5. [Abstract and Section 4.1] The abstract uses the phrase 'the same transferred configuration' before the configuration notation (a, E, S) is introduced in Section 4.1; consider defining the tuple earlier or rephrasing the abstract to avoid ambiguity.
  6. [Figure 3] The schedule labels in Figure 3b, such as '0,31,4 2,5', are difficult to read because of missing spacing; please add explicit separators and clarify which steps are full refreshes.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: GeoCache's delta transport is an affine combination of the target view's own state with a geometry-gathered anchor increment, and its headline fidelity values are measured against stock runs and external baselines rather than derived from fitted inputs.

full rationale

The paper's central operation is Eq. 1, x_0^(v)(t) = x_0^(v)(t-1) + G_{A->v}[Delta x_0^(A)(t)], which is an affine combination of the target view's own previous state and a fixed linear gather of the anchor's per-step increment. No parameter in this equation is fitted to the reported metrics; the correspondence operator G is precomputed from position maps, depends only on geometry, and is explicitly field-independent. The schedule (a, E, S) is a hand-set, ablation-tested configuration rather than a fitted quantity used to force the headline result. Speedups are measured as denoiser-loop ratios against stock runs and temporal baselines on the same assets and seeds, so the MV-LPIPS and MV-PSNR values are externally anchored rather than constructed by definition. The motivating study is an empirical characterization, not a self-citation, and the paper contains no load-bearing citation to prior work by the same authors. The only significant discrepancy is the SyncMVD 'same transferred configuration' wording versus Table 1's E=3/S=20 rows; that is an internal-consistency and correctness concern, not a circular reduction, and therefore does not affect the circularity score.

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

The method introduces no fitted constants in Eq. 1; the core transport is a fixed row-stochastic gather from position maps. The free parameters are schedule choices (anchor count, refresh count, step count, taps, tolerance) that are hand-set and ablation-tested, not fit to the benchmark. The main unproved assumptions concern whether anchor-only attention and sampler-consistent reconstruction preserve a valid denoising trajectory; the paper explicitly labels stability as an empirical check.

free parameters (5)
  • anchor_view_count (a) = 2 (headline), 1 and 3 in ablations
    Number of views that run the denoiser at each cached step; it is the main speed-fidelity knob and is hand-set per backbone.
  • full_refresh_count (E) = 5 on Hunyuan, 2 or 3 on SyncMVD/MVPainter
    Number of full-view refresh steps that bound accumulated error; placement is shown to matter more than count.
  • sampler_steps (S) = 10 on Hunyuan, 20 on SyncMVD, 25 on MVPainter
    Total denoising steps used by the accelerated run; combined with E and a to define the operating point.
  • correspondence_taps (K) = 4
    Number of nearest source taps per target token in the geometric gather; ablation reports K=1 matches K=4.
  • tolerance_fraction = 0.01 (1% of bounding-box diagonal)
    Maximum 3D distance for a source tap to count as a valid match; not systematically swept in the main text.
assumptions (5)
  • domain assumption Anchor-only batch forwards preserve each retained row's positional interpretation, and the approximation is bounded by periodic full refreshes.
    Section 3.2 'Batch-sliced anchor forward' states this and relies on the schedule to bound approximation; no formal error bound is derived.
  • domain assumption Position maps give accurate enough surface correspondence at 1% tolerance, and the row-stochastic gather handles disocclusion.
    Section 3.2 'Correspondence operator' defines the gather; if correspondence is wrong, transported deltas are wrong.
  • ad hoc to paper Converting transported x0 to the sampler's native parameterization maintains a valid denoising trajectory.
    Section 3.2 'Loop order' calls this an interface guarantee rather than a proof and claims only empirical stability.
  • domain assumption UniPC sampler trajectories remain stable under the transported-delta update.
    Section 3.2 'Loop order': the paper states stability was checked at every reported operating point, but no proof is offered.
  • domain assumption MV-LPIPS and MV-PSNR against the stock run at the same seed are meaningful fidelity measures for the speed-fidelity trade-off.
    Evaluation compares accelerated outputs to stock outputs, which isolates approximation error but does not measure absolute texture quality beyond stock.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GeoCache: Training-Free Acceleration of Multi-View Texture Diffusion via Geometric Delta Transport." pith.science (2026). https://pith.science/paper/RUYIBIN2

@misc{pith2026260813255,
  author       = {Pith},
  title        = {Pith review of: GeoCache: Training-Free Acceleration of Multi-View Texture Diffusion via Geometric Delta Transport},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RUYIBIN2}},
  note         = {Machine review of arXiv:2608.13255}
}
abstract

Geometry-conditioned multi-view diffusion enables high-quality 3D texture generation, but its repeated per-view denoiser evaluations introduce substantial computational cost. Existing training-free accelerators primarily exploit temporal redundancy by reusing computation across denoising steps. In multi-view texturing, however, skipping a step also removes the cross-view interaction that continually aligns different observations of the same surface, leading to rapidly degraded consistency and fidelity. Our analysis identifies a complementary source of redundancy: although intermediate features remain view-specific, geometrically corresponding surface points exhibit transferable evolution in their predicted clean signals. Based on this observation, we introduce \gc{}, a training-free plugin that evaluates a rotating subset of anchor views and transports their geometry-aligned per-step $\xz$ updates to the remaining views. Periodic full-view computation controls accumulated error, while sampler-consistent reconstruction preserves the denoising trajectory. \gc{} requires neither retraining nor architectural modification and uses the position maps already available in geometry-conditioned texturing pipelines. Across Hunyuan3D-2.1, SyncMVD, and MVPainter, \gc{} achieves a stronger speed--fidelity trade-off than temporal caches and step reduction at operating points above $2\times$. On Hunyuan3D-2.1, it delivers a $2.21\times$ denoiser-loop speedup with an MV-LPIPS of 0.0293 and an MV-PSNR of 33.60 dB, providing the best fidelity among all tested methods above $2\times$. The same transferred configuration reaches the highest speedup and lowest FLOPs on SyncMVD, while \gc{} achieves the lowest FLOPs and best fidelity among the accelerated methods on MVPainter. These results establish cross-view geometry as an effective acceleration axis for multi-view texture diffusion.

Figures

Figures reproduced from arXiv: 2608.13255 by the authors.

Figure 1
Figure 1. Speed–fidelity ladders on the three backbones Geo [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Step-cache failure, and the exploitable redundancy. (a) A step cache repaints the same surface differently across [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The attachment point of GeoCache, and one cached step. The denoising loop is 14.7% of the paint stage and the only [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Nine assets under stock, GeoCache, and the best temporal cache for that backbone, at matched [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 27 canonical work pages

  1. [1]

    Hunyuan3D 2.1: From Images to High-Fidelity 3D Assets with Production-Ready PBR Material , journal =

    work page

  2. [2]

    Hunyuan3D 2.0: Scaling Diffusion Models for High Resolution Textured 3D Assets Generation , journal =

    work page

  3. [3]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , year =

    Huang, Zehuan and Guo, Yuan-Chen and Wang, Haoran and Yi, Ran and Ma, Lizhuang and Cao, Yan-Pei and Sheng, Lu , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , year =

    work page

  4. [4]

    arXiv preprint arXiv:2505.12635 , year =

    Shao, Mingqi and Xiong, Feng and Sun, Zhaoxu and Xu, Mu , title =. arXiv preprint arXiv:2505.12635 , year =

    arXiv

  5. [5]

    SIGGRAPH Asia 2024 Conference Papers , year =

    Liu, Yuxin and Xie, Minshan and Liu, Hanyuan and Wong, Tien-Tsin , title =. SIGGRAPH Asia 2024 Conference Papers , year =

    work page 2024

  6. [6]

    SIGGRAPH Asia 2025 Conference Papers , year =

    Yuan, Ze and Yu, Xin and Sun, Yangtian and Guo, Yuan-Chen and Cao, Yan-Pei and Liang, Ding and Qi, Xiaojuan , title =. SIGGRAPH Asia 2025 Conference Papers , year =

    work page 2025

  7. [7]

    SIGGRAPH Asia 2024 Conference Papers , year =

    Yu, Xin and Yuan, Ze and Guo, Yuan-Chen and Liu, Ying-Tian and Liu, Jianhui and Li, Yangguang and Cao, Yan-Pei and Liang, Ding and Qi, Xiaojuan , title =. SIGGRAPH Asia 2024 Conference Papers , year =

    work page 2024

  8. [8]

    Advances in Neural Information Processing Systems , volume =

    Tang, Shitao and Zhang, Fuyang and Chen, Jiacheng and Wang, Peng and Furukawa, Yasutaka , title =. Advances in Neural Information Processing Systems , volume =

    work page

Show all 35 references
  1. [9]

    arXiv preprint arXiv:2504.02762 , year =

    Yildirim, Ahmet Burak and Aydogdu, Mustafa Utku and Ceylan, Duygu and Dundar, Aysegul , title =. arXiv preprint arXiv:2504.02762 , year =

    arXiv

  2. [10]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages =

    Liu, Chenyu and Chen, Hongze and Bao, Jingzhi and Zhu, Lingting and Zhang, Runze and Chen, Weikai and Hu, Zeyu and Yin, Yingda and Luo, Keyang and Wang, Xin , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages =

  3. [11]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year =

    Liu, Feng and Zhang, Shiwei and Wang, Xiaofeng and Wei, Yujie and Qiu, Haonan and Zhao, Yuzhong and Zhang, Yingya and Ye, Qixiang and Wan, Fang , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year =

  4. [12]

    Advances in Neural Information Processing Systems (NeurIPS) , year =

    Ma, Zehong and Wei, Longhui and Wang, Feng and Zhang, Shiliang and Tian, Qi , title =. Advances in Neural Information Processing Systems (NeurIPS) , year =

  5. [13]

    arXiv preprint arXiv:2407.01425 , year =

    Selvaraju, Pratheba and Ding, Tianyu and Chen, Tianyi and Zharkov, Ilya and Liang, Luming , title =. arXiv preprint arXiv:2407.01425 , year =

    arXiv

  6. [14]

    arXiv preprint arXiv:2406.01125 , year =

    Chen, Pengtao and Shen, Mingzhu and Ye, Peng and Cao, Jianjian and Tu, Chongjun and Bouganis, Christos-Savvas and Zhao, Yiren and Chen, Tao , title =. arXiv preprint arXiv:2406.01125 , year =

    arXiv

  7. [15]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , year =

    Liu, Jiacheng and Zou, Chang and Lyu, Yuanhuiyi and Chen, Junjie and Zhang, Linfeng , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , year =

  8. [16]

    , title =

    Lv, Zhengyao and Si, Chenyang and Song, Junhao and Yang, Zhenyu and Qiao, Yu and Liu, Ziwei and Wong, Kwan-Yee K. , title =. International Conference on Learning Representations (ICLR) , year =

  9. [17]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year =

    Ma, Xinyin and Fang, Gongfan and Wang, Xinchao , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year =

  10. [18]

    International Conference on Learning Representations (ICLR) , year =

    Zou, Chang and Liu, Xuyang and Liu, Ting and Huang, Siteng and Zhang, Linfeng , title =. International Conference on Learning Representations (ICLR) , year =

  11. [19]

    arXiv preprint arXiv:2510.19755 , year =

    Liu, Jiacheng and Wang, Xinyu and Lin, Yuqi and Wang, Zhikai and Wang, Peiru and Cai, Peiliang and Zhou, Qinming and Yan, Zhengan and Yan, Zexuan and Shi, Zhengyi and Zou, Chang and Ma, Yue and Zhang, Linfeng , title =. arXiv preprint arXiv:2510.19755 , year =

  12. [20]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year =

    Yang, Xingyi and Liu, Songhua and Wang, Xinchao , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year =

  13. [21]

    arXiv preprint arXiv:2511.22533 , year =

    Yang, Mengyu and Yang, Yanming and Xu, Chenyi and Song, Chenxi and Zuo, Yufan and Zhao, Tong and Li, Ruibo and Zhang, Chi , title =. arXiv preprint arXiv:2511.22533 , year =

  14. [22]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year =

    Kwon, Minkyung and Choi, Jinhyeok and Park, Jiho and Jeon, Seonghu and Jang, Jinhyuk and Seo, Junyoung and Kwak, Minseop and Kim, Jin-Hwa and Kim, Seungryong , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year =

  15. [23]

    and Lawrence, Jason and Tatarchuk, Natalya and Isidoro, John R

    Nehab, Diego and Sander, Pedro V. and Lawrence, Jason and Tatarchuk, Natalya and Isidoro, John R. , title =. Proceedings of the 22nd ACM SIGGRAPH/EUROGRAPHICS Symposium on Graphics Hardware , pages =

  16. [24]

    Advances in Neural Information Processing Systems (NeurIPS) , year =

    Zhao, Wenliang and Bai, Lujia and Rao, Yongming and Zhou, Jie and Lu, Jiwen , title =. Advances in Neural Information Processing Systems (NeurIPS) , year =

  17. [25]

    International Conference on Learning Representations (ICLR) , year =

    Song, Jiaming and Meng, Chenlin and Ermon, Stefano , title =. International Conference on Learning Representations (ICLR) , year =

  18. [26]

    arXiv preprint arXiv:2310.04378 , year =

    Luo, Simian and Tan, Yiqin and Huang, Longbo and Li, Jian and Zhao, Hang , title =. arXiv preprint arXiv:2310.04378 , year =

    arXiv

  19. [27]

    International Conference on Learning Representations (ICLR) , year =

    Salimans, Tim and Ho, Jonathan , title =. International Conference on Learning Representations (ICLR) , year =

  20. [28]

    ACM SIGGRAPH 2023 Conference Proceedings , year =

    Richardson, Elad and Metzer, Gal and Alaluf, Yuval and Giryes, Raja and Cohen-Or, Daniel , title =. ACM SIGGRAPH 2023 Conference Proceedings , year =

    2023

  21. [29]

    Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , year =

    Chen, Dave Zhenyu and Siddiqui, Yawar and Lee, Hsin-Ying and Tulyakov, Sergey and Nie. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , year =

  22. [30]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year =

    Zeng, Xianfang and Chen, Xin and Qi, Zhongqi and Liu, Wen and Zhao, Zibo and Wang, Zhibin and Fu, Bin and Liu, Yong and Yu, Gang , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year =

  23. [31]

    and Vanhoucke, Vincent , title =

    Downs, Laura and Francis, Anthony and Koenig, Nate and Kinman, Brandon and Hickman, Ryan and Reymann, Krista and McHugh, Thomas B. and Vanhoucke, Vincent , title =. IEEE International Conference on Robotics and Automation (ICRA) , year =

  24. [32]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year =

    Deitke, Matt and Schwenk, Dustin and Salvador, Jordi and Weihs, Luca and Michel, Oscar and VanderBilt, Eli and Schmidt, Ludwig and Ehsani, Kiana and Kembhavi, Aniruddha and Farhadi, Ali , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognitio...

  25. [33]

    Yago and Dideriksen, Thomas and Arora, Himanshu and Guillaumin, Matthieu and Malik, Jitendra , title =

    Collins, Jasmine and Goel, Shubham and Deng, Kenan and Luthra, Achleshwar and Xu, Leon and Gundogdu, Erhan and Zhang, Xi and Vicente, Tomas F. Yago and Dideriksen, Thomas and Arora, Himanshu and Guillaumin, Matthieu and Malik, Jitendra , title =. Proceedings of the IEEE/CVF Co...

  26. [34]

    arXiv preprint arXiv:2508.10868 , year =

    Zhang, Yibo and Zhang, Li and Ma, Rui and Cao, Nan , title =. arXiv preprint arXiv:2508.10868 , year =

    arXiv

  27. [35]

    and Shechtman, Eli and Wang, Oliver , title =

    Zhang, Richard and Isola, Phillip and Efros, Alexei A. and Shechtman, Eli and Wang, Oliver , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year =

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.