arxiv: 2605.03337 · v1 · submitted 2026-05-05 · 💻 cs.CV · cs.AI

Recognition: unknown

FreeTimeGS++: Secrets of Dynamic Gaussian Splatting and Their Principles

Lucas Yunkyu Lee , Soonho Kim , Youngwook Kim , Sangmin Kim , Jaesik Park

Authors on Pith no claims yet

Pith reviewed 2026-05-08 01:21 UTC · model grok-4.3

classification 💻 cs.CV cs.AI

keywords 4D Gaussian Splattingdynamic reconstructiontemporal partitioninggated marginalizationneural velocity fieldsspatiotemporal consistencystability

0 comments

The pith

Dissecting 4D Gaussian Splatting uncovers temporal partitioning and consistency gaps that gated marginalization and neural velocity fields can resolve for more stable dynamic reconstructions.

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

This paper aims to explain why 4D Gaussian Splatting works well for dynamic scenes by breaking down the components that previous methods used only as heuristics. The authors build a controlled version of an existing approach to isolate the effects of Gaussian durations creating automatic time splits and the mismatch between matching pixels and maintaining smooth motion over time. They then design FreeTimeGS++ around gated marginalization, which selectively combines information, and neural velocity fields, which explicitly model motion, to fix these issues. Readers interested in reliable video reconstruction would value the shift toward methods that are both effective and consistent across different training runs.

Core claim

By formalizing and reproducing the heuristics in a baseline called FreeTimeGS_ours, the analysis reveals that Gaussian splatting in 4D exhibits emergent temporal partitioning based on how long each Gaussian lasts and a discrepancy between photometric fidelity and spatiotemporal consistency. FreeTimeGS++ addresses these through gated marginalization and neural velocity fields to produce dynamic representations with greater stability and lower variation between repeated runs.

What carries the argument

Gated marginalization and neural velocity fields that control information flow across time and model explicit motion to counteract partitioning and consistency problems in 4D Gaussian splatting.

If this is right

Dynamic scene reconstructions become more stable across different training runs.
Run-to-run variance decreases, making results more reproducible.
The approach provides a reliable base for further development in 4D Gaussian Splatting.
Performance improves by directly targeting the identified temporal and consistency factors.

Where Pith is reading between the lines

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

This analysis framework could help diagnose similar hidden issues in other dynamic rendering techniques like neural radiance fields.
Explicit velocity modeling might allow better handling of fast-moving objects or longer time sequences than implicit methods.
Releasing the implementation could accelerate community efforts to build upon these principles rather than re-deriving heuristics.

Load-bearing premise

That the temporal partitioning and photometric-spatiotemporal discrepancy are the primary reasons for instability in prior methods and that gated marginalization plus neural velocity fields can fix them directly.

What would settle it

Running multiple trainings of the baseline and the new method on identical dynamic scenes and comparing the variance in reconstruction quality and stability metrics would confirm or refute the gains in robustness.

Figures

Figures reproduced from arXiv: 2605.03337 by Jaesik Park, Lucas Yunkyu Lee, Sangmin Kim, Soonho Kim, Youngwook Kim.

**Figure 1.** Figure 1: Implicit temporal partitioning through Gaussian temporal scales. In (a), the horizontal axis is normalized by the full sequence span T. photometric fidelity (e.g., PSNR), potentially by over-parameterizing the scene’s spatial and temporal priors. Consequently, the current literature predominantly focuses on optimizing 2D rendering metrics, often overlooking whether the learned representations reflect physi… view at source ↗

**Figure 2.** Figure 2: Photometric-motion decoupling in FreeTimeGSours. We compare rendered RGB, Gaussian velocity visualization, and activated short-lived Gaussians. Motion consistency. Dynamic Gaussian methods differ in how they represent motion, ranging from canonical-deformation fields to explicit trajectories and native spacetime primitives. However, they are typically evaluated using imagespace rendering metrics such as P… view at source ↗

**Figure 3.** Figure 3: Space-time slices for different relocation policies. The vertical axis represents time, and the horizontal axis corresponds to a spatial slice of the background view at source ↗

**Figure 5.** Figure 5: Qualitative comparison of velocity map consistency on the view at source ↗

**Figure 6.** Figure 6: Visualization of persistent and transient components in NVF view at source ↗

read the original abstract

The recent surge in 4D Gaussian Splatting (4DGS) has achieved impressive dynamic scene reconstruction. While these methods demonstrate remarkable performance, the specific drivers behind such gains remain less explored, making a systematic understanding of the underlying principles challenging. In this paper, we perform a comprehensive analysis of these hidden factors to provide a clearer perspective on the 4DGS framework. We first establish a controlled baseline, FreeTimeGS_ours, by formalizing and reproducing the heuristics of the state-of-the-art FreeTimeGS. Using this framework, we dissect 4DGS along its fundamental axes and uncover key secrets, including the emergent temporal partitioning driven by Gaussian durations and the discrepancy between photometric fidelity and spatiotemporal consistency. Based on these insights, we propose FreeTimeGS++, a principled method that employs gated marginalization and neural velocity fields to achieve superior stability and robust dynamic representations. Our approach yields reproducible results with reduced run-to-run variance. We will release our implementation to provide a reliable foundation for future 4DGS research.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper reproduces a 4DGS baseline, spots some emergent temporal behaviors, and adds gated marginalization plus neural velocity fields for lower variance, but the causal connection between observations and gains is not isolated.

read the letter

The main takeaway is that this work gives a useful look at what actually happens inside dynamic Gaussian splatting by reproducing FreeTimeGS as a controlled baseline and then naming two patterns: Gaussians self-partitioning time through their duration parameters and a gap between photometric accuracy and spatiotemporal consistency. From those observations they introduce gated marginalization and neural velocity fields in FreeTimeGS++ to cut run-to-run variance and improve stability. The reproduction step and the identification of those behaviors are the parts that feel genuinely helpful for the subfield. Anyone already tuning 4DGS models will probably find the baseline and the listed heuristics worth checking against their own setups, especially once the code drops. The soft spot is the missing isolation. The paper shows the new version performs better than the reproduced baseline, yet it does not run the clean on/off experiments that would tie the stability gains directly to the two new components rather than to other implementation choices or hyperparameter shifts. That leaves the attribution observational. The circularity concern also lands: the insights come from dissecting the same framework they then improve, so it is not obvious whether the fixes address root causes or just compensate for the baseline's particular quirks. This paper is for researchers already working on 4D Gaussian methods who want better reproducibility and some intuition about design levers. It is not broad enough or rigorous enough to interest people outside that niche. I would still send it for peer review because the baseline reproduction and the code release could serve as a practical reference point even if the causal claims need more scrutiny in revision.

Referee Report

2 major / 1 minor

Summary. The paper reproduces FreeTimeGS heuristics to form a controlled baseline (FreeTimeGS_ours), uses it to identify emergent temporal partitioning driven by Gaussian durations and a photometric-spatiotemporal discrepancy, and introduces FreeTimeGS++ with gated marginalization and neural velocity fields, claiming these yield superior stability, robust dynamic representations, and reduced run-to-run variance. The authors commit to releasing the implementation.

Significance. If the empirical claims hold under controlled conditions, the work could clarify principles underlying 4DGS and supply a reproducible baseline for the community; the planned code release is a clear strength for reproducibility.

major comments (2)

[§4 (Experiments and Ablations)] The central claim that gated marginalization and neural velocity fields directly address the identified temporal partitioning and photometric-spatiotemporal discrepancy to produce superior stability rests on observational analysis of the baseline rather than causal evidence. No controlled ablations are described that activate these components independently while holding all other factors fixed (see the evaluation of FreeTimeGS++ and comparison to FreeTimeGS_ours).
[§3 (Analysis of 4DGS) and §4] The reproduction of FreeTimeGS_ours and subsequent derivation of 'secrets' from observations within that same framework creates a circularity risk for the attribution of gains; external validation against independent 4DGS methods or quantitative isolation of run-to-run variance reduction is needed to support the stability claims.

minor comments (1)

[Abstract] The abstract states that the implementation will be released but provides no details on repository location, license, or exact components included.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our experimental design and validation strategy. We agree that strengthening causal evidence through additional ablations and broadening external comparisons will improve the manuscript. We address each major comment below.

read point-by-point responses

Referee: [§4 (Experiments and Ablations)] The central claim that gated marginalization and neural velocity fields directly address the identified temporal partitioning and photometric-spatiotemporal discrepancy to produce superior stability rests on observational analysis of the baseline rather than causal evidence. No controlled ablations are described that activate these components independently while holding all other factors fixed (see the evaluation of FreeTimeGS++ and comparison to FreeTimeGS_ours).

Authors: We acknowledge that the present evaluation compares the complete FreeTimeGS++ system against FreeTimeGS_ours without isolating the individual effects of gated marginalization and neural velocity fields. This limits the strength of causal claims. In the revised version we will add controlled ablation experiments that enable each component independently while freezing all other factors, thereby directly measuring their contributions to stability, temporal partitioning, and photometric-spatiotemporal consistency. revision: yes
Referee: [§3 (Analysis of 4DGS) and §4] The reproduction of FreeTimeGS_ours and subsequent derivation of 'secrets' from observations within that same framework creates a circularity risk for the attribution of gains; external validation against independent 4DGS methods or quantitative isolation of run-to-run variance reduction is needed to support the stability claims.

Authors: We recognize the risk of circular attribution when insights are drawn solely from the reproduced baseline. While FreeTimeGS_ours was constructed as a faithful, controlled reproduction of the original heuristics, we will expand the experimental section to include direct quantitative comparisons against independent 4DGS methods from the literature. In addition, we will report run-to-run variance statistics (means and standard deviations across multiple random seeds) for both FreeTimeGS_ours and FreeTimeGS++ to provide objective evidence of the claimed stability improvement. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper reproduces heuristics from prior FreeTimeGS work to form a controlled baseline (FreeTimeGS_ours), performs empirical dissection to identify patterns such as temporal partitioning and photometric-spatiotemporal discrepancy, then proposes gated marginalization and neural velocity fields as improvements. This follows a standard analysis-then-extend workflow without any self-definitional loops, fitted parameters renamed as predictions, or load-bearing self-citations that reduce the central claims to inputs by construction. The approach remains self-contained because the new components are introduced as direct responses to observed behaviors in the reproduced framework, with code release promised for independent verification against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Insufficient detail in abstract to identify specific free parameters, axioms, or invented entities; the work appears to rely on standard assumptions from Gaussian Splatting literature.

pith-pipeline@v0.9.0 · 5491 in / 1213 out tokens · 38621 ms · 2026-05-08T01:21:21.559941+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Reference graph

Works this paper leans on

32 extracted references · 12 canonical work pages · 1 internal anchor

[1]

In: Computer Vision – ECCV 2024: 18th European Conference, Milan, Italy, September 29–October 4, 2024, Proceedings, Part XV

Bae, J., Kim, S., Yun, Y., Lee, H., Bang, G., Uh, Y.: Per-gaussian embedding- based deformation for deformable 3d gaussian splatting. In: Computer Vision – ECCV 2024: 18th European Conference, Milan, Italy, September 29–October 4, 2024, Proceedings, Part XV. p. 321–335. Springer-Verlag, Berlin, Heidelberg (2024). https://doi.org/10.1007/978-3-031-72633-0_...

work page doi:10.1007/978-3-031-72633-0_18 2024
[2]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

Edstedt, J., Sun, Q., Bökman, G., Wadenbäck, M., Felsberg, M.: Roma: Robust dense feature matching. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 19790–19800 (June 2024)

2024
[3]

arXiv preprint arXiv:2403.12365 (2024)

Gao, Q., Xu, Q., Cao, Z., Mildenhall, B., Ma, W., Chen, L., Tang, D., Neumann, U.: Gaussianflow: Splatting gaussian dynamics for 4d content creation. arXiv preprint arXiv:2403.12365 (2024)

work page arXiv 2024
[4]

IEEE Transactions on Circuits and Systems for Video Technology35(4), 3119–3133 (2024)

Guo, Z., Zhou, W., Li, L., Wang, M., Li, H.: Motion-aware 3d gaussian splatting for efficient dynamic scene reconstruction. IEEE Transactions on Circuits and Systems for Video Technology35(4), 3119–3133 (2024)

2024
[5]

Vbench: Comprehensive benchmark suite for video generative models

Huang, Y.H., Sun, Y.T., Yang, Z., Lyu, X., Cao, Y.P., Qi, X.: SC-GS: Sparse- controlled gaussian splatting for editable dynamic scenes. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 4220–4230 (June 2024).https://doi.org/10.1109/CVPR52733.2024.00404 , https://ieeexplore.ieee.org/document/10657650/

work page doi:10.1109/cvpr52733.2024.00404 2024
[6]

In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)

Jiang, D., Hou, Z., Ke, Z., Yang, X., Zhou, X., Qiu, T.: Timeformer: Capturing temporal relationships of deformable 3d gaussians for robust reconstruction. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). pp. 8721–8732 (October 2025)

2025
[7]

https://doi.org/10.1145/3592433 Xiaonan Kong and Riley G

Kerbl, B., Kopanas, G., Leimkuehler, T., Drettakis, G.: 3d gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics42(4) (July 2023). https://doi.org/10.1145/3592433, https://doi.org/10.1145/3592433, place: New York, NY, USA

work page doi:10.1145/3592433 2023
[8]

3D Gaussian Splatting as Markov Chain Monte Carlo , url =

Kheradmand, S., Rebain, D., Sharma, G., Sun, W., Tseng, Y.C., Isack, H., Kar, A., Tagliasacchi, A., Yi, K.M.: 3d gaussian splatting as markov chain monte carlo. In: Globerson, A., Mackey, L., Belgrave, D., Fan, A., Paquet, U., Tom- czak, J., Zhang, C. (eds.) Advances in Neural Information Processing Systems. vol. 37, pp. 80965–80986. Curran Associates, In...

work page doi:10.52202/079017-2573 2024
[9]

Adam: A Method for Stochastic Optimization

Kingma, D.P., Ba, J.: Adam: A method for stochastic optimization. In: Bengio, Y., LeCun, Y. (eds.) 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings (2015), http://arxiv.org/abs/1412.6980

work page internal anchor Pith review arXiv 2015
[10]

In: Leonardis, A., Ricci, E., Roth, S., Russakovsky, O., Sattler, T., Varol, G

Labe, I., Issachar, N., Lang, I., Benaim, S.: Dgd: Dynamic 3d gaussians distillation. In: Leonardis, A., Ricci, E., Roth, S., Russakovsky, O., Sattler, T., Varol, G. (eds.) Computer Vision – ECCV 2024. pp. 361–378. Springer Nature Switzerland, Cham (2025)

2024
[11]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

Lee, J.C., Rho, D., Sun, X., Ko, J.H., Park, E.: Compact 3d gaussian representation for radiance field. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 21719–21728 (2024)

2024
[12]

In: Proceedings of the Neural Information Processing Systems (2024)

Lee, J., Won, C., Jung, H., Bae, I., Jeon, H.G.: Fully explicit dynamic guassian splatting. In: Proceedings of the Neural Information Processing Systems (2024)

2024
[13]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

Li, T., Slavcheva, M., Zollhöfer, M., Green, S., Lassner, C., Kim, C., Schmidt, T., Lovegrove, S., Goesele, M., Newcombe, R., Lv, Z.: Neural 3d video synthesis from multi-view video. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 5521–5531 (June 2022)

2022
[14]

Vbench: Comprehensive benchmark suite for video generative models

Li, Z., Chen, Z., Li, Z., Xu, Y.: Spacetime gaussian feature splatting for real-time dynamic view synthesis. In: 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 8508–8520. IEEE (2024).https://doi. org/10.1109/CVPR52733.2024.00813, https://ieeexplore.ieee.org/document/ 10657623/

work page doi:10.1109/cvpr52733.2024.00813 2024
[15]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2021)

Li, Z., Niklaus, S., Snavely, N., Wang, O.: Neural scene flow fields for space-time view synthesis of dynamic scenes. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2021)

2021
[16]

In: Proc

Liang, Y., Khan, N., Li, Z., Nguyen-Phuoc, T., Lanman, D., Tompkin, J., Xiao, L.: Gaufre: Gaussian deformation fields for real-time dynamic novel view synthesis. In: Proc. IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) (2025)

2025
[17]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

Lin, Y., Dai, Z., Zhu, S., Yao, Y.: Gaussian-flow: 4d reconstruction with dynamic 3d gaussian particle. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 21136–21145 (June 2024)

2024
[18]

In: Yue, Y., Garg, A., Peng, N., Sha, F., Yu, R

LIU, Q., Liu, Y., Wang, J., Lyu, X., Wang, P., Wang, W., Hou, J.: Modgs: Dynamic gaussian splatting from casually-captured monocular videos with depth priors. In: Yue, Y., Garg, A., Peng, N., Sha, F., Yu, R. (eds.) International Conference on Learn- ing Representations. vol. 2025, pp. 97048–97074 (2025),https://proceedings. iclr.cc/paper_files/paper/2025/...

2025
[19]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

Lu, Z., Guo, X., Hui, L., Chen, T., Yang, M., Tang, X., Zhu, F., Dai, Y.: 3d geometry- aware deformable gaussian splatting for dynamic view synthesis. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 8900–8910 (June 2024)

2024
[20]

Mildenhall, B., Srinivasan, P.P., Tancik, M., Barron, J.T., Ramamoorthi, R., Ng, R.: Nerf: representing scenes as neural radiance fields for view synthesis. Commun. ACM65(1), 99–106 (Dec 2021).https://doi.org/10.1145/3503250, https:// doi.org/10.1145/3503250

work page doi:10.1145/3503250 2021
[21]

In: Proceedings of the FreeTimeGS++17 IEEE/CVF International Conference on Computer Vision (ICCV)

Park, K., Sinha, U., Barron, J.T., Bouaziz, S., Goldman, D.B., Seitz, S.M., Martin- Brualla, R.: Nerfies: Deformable neural radiance fields. In: Proceedings of the FreeTimeGS++17 IEEE/CVF International Conference on Computer Vision (ICCV). pp. 5865–5874 (October 2021)

2021
[22]

ACM Trans

Park, K., Sinha, U., Hedman, P., Barron, J.T., Bouaziz, S., Goldman, D.B., Martin-Brualla, R., Seitz, S.M.: Hypernerf: a higher-dimensional representation for topologically varying neural radiance fields. ACM Trans. Graph.40(6) (Dec 2021). https://doi.org/10.1145/3478513.3480487, https://doi.org/10.1145/ 3478513.3480487

work page doi:10.1145/3478513.3480487 2021
[23]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (2021)

Pumarola, A., Corona, E., Pons-Moll, G., Moreno-Noguer, F.: D-NeRF: Neural Radiance Fields for Dynamic Scenes. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (2021)

2021
[24]

In: Leonardis, A., Ricci, E., Roth, S., Russakovsky, O., Sattler, T., Varol, G

Shaw, R., Nazarczuk, M., Song, J., Moreau, A., Catley-Chandar, S., Dhamo, H., Pérez-Pellitero, E.: Swings: Sliding windows for dynamic 3d gaussian splatting. In: Leonardis, A., Ricci, E., Roth, S., Russakovsky, O., Sattler, T., Varol, G. (eds.) Computer Vision – ECCV 2024. pp. 37–54. Springer Nature Switzerland, Cham (2025)

2024
[25]

In: IEEE/CVF International Conference on Computer Vision (ICCV)

Song, R., Liang, C., Xia, Y., Zimmer, W., Cao, H., Caesar, H., Festag, A., Knoll, A.: Coda-4dgs: Dynamic gaussian splatting with context and deformation awareness for autonomous driving. In: IEEE/CVF International Conference on Computer Vision (ICCV). IEEE/CVF (2025)

2025
[26]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2025)

Wang, Y., Yang, P., Xu, Z., Sun, J., Zhang, Z., Chen, Y., Bao, H., Peng, S., Zhou, X.: Freetimegs: Free gaussian primitives at anytime and anywhere for dynamic scene reconstruction. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2025). https://doi.org/10.48550/ arXiv.2506.05348

work page arXiv 2025
[27]

Vbench: Comprehensive benchmark suite for video generative models

Wu, G., Yi, T., Fang, J., Xie, L., Zhang, X., Wei, W., Liu, W., Tian, Q., Wang, X.: 4d gaussian splatting for real-time dynamic scene rendering. In: 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 20310– 20320. IEEE (2024). https://doi.org/10.1109/CVPR52733.2024.01920, https: //ieeexplore.ieee.org/document/10656774/

work page doi:10.1109/cvpr52733.2024.01920 2024
[28]

In: proceedings of Medical Image Computing and Computer Assisted Intervention – MICCAI 2024

Xie, W., Yao, J., Cao, X., Lin, Q., Tang, Z., Dong, X., Guo, X.: SurgicalGaussian: Deformable 3D Gaussians for High-Fidelity Surgical Scene Reconstruction . In: proceedings of Medical Image Computing and Computer Assisted Intervention – MICCAI 2024. vol. LNCS 15006. Springer Nature Switzerland (October 2024)

2024
[29]

ACM Transactions on Graphics 43(6) (November 2024),https://zju3dv.github.io/longvolcap

Xu, Z., Xu, Y., Yu, Z., Peng, S., Sun, J., Bao, H., Zhou, X.: Representing long volumetric video with temporal gaussian hierarchy. ACM Transactions on Graphics 43(6) (November 2024),https://zju3dv.github.io/longvolcap

2024
[30]

In: Kim, B., Yue, Y., Chaudhuri,S.,Fragkiadaki,K.,Khan,M.,Sun,Y.(eds.)InternationalConferenceon Learning Representations

Yang, Z., Yang, H., Pan, Z., Zhang, L.: Real-time photorealistic dynamic scene representation and rendering with 4d gaussian splatting. In: Kim, B., Yue, Y., Chaudhuri,S.,Fragkiadaki,K.,Khan,M.,Sun,Y.(eds.)InternationalConferenceon Learning Representations. vol. 2024, pp. 9142–9159 (2024),https://proceedings. iclr.cc/paper_files/paper/2024/file/26230ff529...

2024
[31]

In: arXiV (2025)

Zhang, Y., Keetha, N., Lyu, C., Jhamb, B., Chen, Y., Qiu, Y., Karhade, J., Jha, S., Hu, Y., Ramanan, D., Scherer, S., Wang, W.: Ufm: A simple path towards unified dense correspondence with flow. In: arXiV (2025)

2025
[32]

In: Globerson, A., Mackey, L., Belgrave, D., Fan, A., Paquet, U., Tom- czak, J., Zhang, C

Zhu, R., Liang, Y., Chang, H., Deng, J., Lu, J., Yang, W., Zhang, T., Zhang, Y.: Motiongs: Exploring explicit motion guidance for deformable 3d gaussian splatting. In: Globerson, A., Mackey, L., Belgrave, D., Fan, A., Paquet, U., Tom- czak, J., Zhang, C. (eds.) Advances in Neural Information Processing Systems. vol. 37, pp. 101790–101817. Curran Associate...

work page doi:10.52202/079017-3229 2024