pith. sign in

arxiv: 2605.24962 · v1 · pith:3C36MSSHnew · submitted 2026-05-24 · 💻 cs.CV

Tempered Self-Similarity Alignment for Physically Plausible Video Generation

Manjin Kim , Suha Kwak , Minsu Cho This is my paper

Pith reviewed 2026-06-30 11:36 UTC · model grok-4.3

classification 💻 cs.CV
keywords video generationphysical plausibilityself-similarityfoundation modelsTSA lossVideoPhy benchmarkrelational knowledgecorrespondence distribution
0
0 comments X

The pith

Aligning spatio-temporal self-similarity from foundation models makes generated videos more physically plausible.

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

Video generative models frequently produce appearance drift, implausible motion, and temporal inconsistencies. The paper identifies spatio-temporal self-similarity computed by visual foundation models as a carrier of relational knowledge about how objects interact over time. It introduces a Tempered Self-similarity Alignment loss that converts these similarities into probabilistic correspondence distributions and forces the generative model to match them on regions that change dynamically. Experiments on the VideoPhy and VideoPhy2 benchmarks report gains in physical plausibility across varied interaction scenarios. The approach relies on implicit relational structure rather than explicit physics equations.

Core claim

Transferring relational knowledge encoded in spatio-temporal self-similarity from visual foundation models into video generative models via TSA loss produces substantial improvements in physical plausibility across diverse interaction scenarios on VideoPhy and VideoPhy2.

What carries the argument

Tempered Self-similarity Alignment (TSA) loss, which transforms pairwise feature similarities into probabilistic correspondence distributions and aligns the generative model on dynamically changing regions.

If this is right

  • Substantial improvements in physical plausibility occur on VideoPhy and VideoPhy2 benchmarks.
  • The method applies across diverse interaction scenarios without explicit physics modeling.
  • Appearance drift, implausible motion, and temporal inconsistencies decrease when correspondence distributions are matched.
  • Relational knowledge transfers from foundation models suffice to enforce real-world dynamics inside generative models.

Where Pith is reading between the lines

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

  • Self-similarity patterns may implicitly encode dynamics that could be reused in other generative domains such as 3D or audio synthesis.
  • The alignment technique might be combined with existing motion or consistency losses for additive gains.
  • Testing on longer video sequences could reveal whether the relational transfer scales beyond short clips.

Load-bearing premise

Pairwise feature similarities computed by an off-the-shelf visual foundation model on real video already encode the physical interaction rules needed for plausible generation.

What would settle it

Generate videos with and without the TSA loss on the same set of prompts involving object interactions, then measure whether the rate of physical violations remains unchanged.

Figures

Figures reproduced from arXiv: 2605.24962 by Manjin Kim, Minsu Cho, Suha Kwak.

Figure 1
Figure 1. Figure 1: Aligning spatio-temporal correspondence improves realistic dynamics in video generation. Given the same video in￾put, we compare spatio-temporal correspondence probability maps obtained from (a) a visual foundation model and (b) a video gen￾erative model. The foundation model captures clear correspon￾dences, whereas the generative model produces noisy and inaccu￾rate ones. (c) We here transfer the accurate… view at source ↗
Figure 2
Figure 2. Figure 2: Framework overview. Our method aligns the noisy spatio-temporal correspondences of a video diffusion model with the accurate correspondences of a visual foundation model, guiding the generative model toward more realistic motion dynamics. By restricting this alignment to dynamic regions, it encourages motion-focused alignment for physically plausible video generation. foundation model gϕ. We then obtain in… view at source ↗
Figure 4
Figure 4. Figure 4: Motion-saliency masks at different thresholds k. In￾creasing k highlights a larger proportion of the most dynamic re￾gions in the video. 5. Experiments 5.1. Implementation Details We adopt CogVideoX-2B [56] and VideoMAEv2-B [50] as the base text-to-video diffusion model and visual founda￾tion model, respectively. We employ a lightweight projec￾tor composed of a 3-layer MLP followed by a 3D convo￾lutional l… view at source ↗
Figure 5
Figure 5. Figure 5: Ablation studies on VideoPhy. The PC scores are reported. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative results. Red rectangles highlight regions with physically implausible or temporally inconsistent motion. Our method generates videos with physically realistic dynamics. embedded in STSS to video generative models. Please refer to Appendix A for more qualitative results. 6. Conclusion We have introduced Tempered Self-similarity Alignment (TSA), a novel framework that transfers spatio-temporal co… view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative results. Red rectangles highlight regions with physically implausible or temporally inconsistent motion. Our method generates videos with physically realistic dynamics [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
read the original abstract

Despite remarkable advances in video generative models, they still struggle to generate physically realistic videos, frequently exhibiting appearance drift, implausible motion, and temporal inconsistencies. In this work, we address this limitation by transferring relational knowledge encoded in spatio-temporal self-similarity (STSS) from visual foundation models into video generative models. STSS represents pairwise similarities among features across space and time, revealing the relational structure of how objects interact with other entities throughout a video, effectively capturing real-world dynamics, including object motion and semantic transformations. To transfer this relational knowledge, we propose Tempered Self-similarity Alignment (TSA) loss, which transforms STSS into probabilistic correspondence distributions and trains the video generative model to align its correspondence distributions with those of the visual foundation model on dynamically changing regions. Evaluated on VideoPhy and VideoPhy2 benchmarks, our method demonstrates substantial improvements in physical plausibility across diverse interaction scenarios, validating the effectiveness of transferring relational knowledge for physically realistic video generation.

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.

Referee Report

2 major / 1 minor

Summary. The paper claims that transferring relational knowledge encoded in spatio-temporal self-similarity (STSS) from visual foundation models into video generative models via a Tempered Self-Similarity Alignment (TSA) loss produces substantial improvements in physical plausibility across diverse interaction scenarios on the VideoPhy and VideoPhy2 benchmarks.

Significance. If the central claim holds and the TSA loss demonstrably enforces physical interaction rules (rather than merely appearance consistency), the work would offer a practical mechanism for improving dynamics in video generation without requiring explicit physics simulators.

major comments (2)
  1. [Abstract] Abstract: the claim of 'substantial improvements' is presented without any quantitative metrics, ablation results, or description of how the TSA loss is implemented or regularized, so the data-to-claim link cannot be verified.
  2. [Method] Method (TSA loss definition): no evidence is provided that the correspondence distributions derived from off-the-shelf VFM features on dynamically changing regions differ systematically between physically valid and invalid videos, leaving open the possibility that alignment primarily suppresses appearance drift rather than correcting implausible dynamics such as collisions or conservation violations.
minor comments (1)
  1. The abstract and method would benefit from explicit equations for the transformation of STSS into probabilistic correspondence distributions and the precise form of the tempered alignment objective.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and indicate planned revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim of 'substantial improvements' is presented without any quantitative metrics, ablation results, or description of how the TSA loss is implemented or regularized, so the data-to-claim link cannot be verified.

    Authors: We agree that the abstract would be strengthened by including supporting details. In the revised version, we will incorporate key quantitative metrics from the VideoPhy and VideoPhy2 evaluations, reference main ablation findings, and add a brief description of the TSA loss formulation and regularization approach. revision: yes

  2. Referee: [Method] Method (TSA loss definition): no evidence is provided that the correspondence distributions derived from off-the-shelf VFM features on dynamically changing regions differ systematically between physically valid and invalid videos, leaving open the possibility that alignment primarily suppresses appearance drift rather than correcting implausible dynamics such as collisions or conservation violations.

    Authors: The TSA loss is deliberately restricted to dynamically changing regions to emphasize relational structures tied to motion and interactions. The consistent gains on VideoPhy and VideoPhy2 benchmarks—which explicitly measure physical plausibility aspects such as collisions and conservation—indicate that the alignment improves dynamics rather than only suppressing appearance changes. We will add a supporting analysis of correspondence distribution differences between valid and invalid videos in the revision. revision: partial

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The provided abstract and description present TSA as an alignment loss that transforms STSS into correspondence distributions and matches them between a frozen VFM and the generative model on dynamic regions. No equations, fitting procedures, or self-citations appear in the text that would reduce any claimed result to an input by construction. The method is defined directly as the proposed transfer mechanism and evaluated on external benchmarks (VideoPhy, VideoPhy2), with no self-definitional loops, renamed known results, or load-bearing self-citations. The derivation chain remains independent of its own outputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 1 invented entities

The approach rests on the untested premise that foundation-model STSS already encodes transferable physical dynamics; no free parameters, axioms, or new entities are quantified in the abstract.

invented entities (1)
  • Tempered Self-similarity Alignment (TSA) loss no independent evidence
    purpose: Transforms STSS into probabilistic correspondence distributions and enforces alignment on dynamic regions
    Introduced in the abstract as the core training objective; no independent evidence supplied.

pith-pipeline@v0.9.1-grok · 5697 in / 1118 out tokens · 32241 ms · 2026-06-30T11:36:31.884924+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

62 extracted references · 28 canonical work pages · 16 internal anchors

  1. [1]

    Cosmos World Foundation Model Platform for Physical AI

    Niket Agarwal, Arslan Ali, Maciej Bala, Yogesh Balaji, Erik Barker, Tiffany Cai, Prithvijit Chattopadhyay, Yongxin Chen, Yin Cui, Yifan Ding, et al. Cosmos world foun- dation model platform for physical ai.arXiv preprint arXiv:2501.03575, 2025. 6

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  2. [2]

    VideoPhy: Evaluating Physical Commonsense for Video Generation

    Hritik Bansal, Zongyu Lin, Tianyi Xie, Zeshun Zong, Michal Yarom, Yonatan Bitton, Chenfanfu Jiang, Yizhou Sun, Kai- Wei Chang, and Aditya Grover. Videophy: Evaluating physical commonsense for video generation.arXiv preprint arXiv:2406.03520, 2024. 1, 2, 5

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  3. [3]

    Videophy-2: A challenging action-centric physical commonsense evaluation in video generation.arXiv preprint arXiv:2503.06800, 2025

    Hritik Bansal, Clark Peng, Yonatan Bitton, Roman Golden- berg, Aditya Grover, and Kai-Wei Chang. Videophy-2: A challenging action-centric physical commonsense evaluation in video generation.arXiv preprint arXiv:2503.06800, 2025. 1, 2, 5

    work page arXiv 2025

  4. [4]

    Moalign: Motion-centric representation alignment for video diffusion models.arXiv preprint arXiv:2510.19022, 2025

    Aritra Bhowmik, Denis Korzhenkov, Cees GM Snoek, Amirhossein Habibian, and Mohsen Ghafoorian. Moalign: Motion-centric representation alignment for video diffusion models.arXiv preprint arXiv:2510.19022, 2025. 1

    work page arXiv 2025

  5. [5]

    Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets

    Andreas Blattmann, Tim Dockhorn, Sumith Kulal, Daniel Mendelevitch, Maciej Kilian, Dominik Lorenz, Yam Levi, Zion English, Vikram V oleti, Adam Letts, et al. Stable video diffusion: Scaling latent video diffusion models to large datasets.arXiv preprint arXiv:2311.15127, 2023. 1, 2, 3

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  6. [6]

    Video generation models as world simulators.OpenAI Blog, 1(8):1, 2024

    Tim Brooks, Bill Peebles, Connor Holmes, Will DePue, Yufei Guo, Li Jing, David Schnurr, Joe Taylor, Troy Luh- man, Eric Luhman, et al. Video generation models as world simulators.OpenAI Blog, 1(8):1, 2024. 1, 2

    2024

  7. [7]

    Videojam: Joint appearance-motion representations for en- hanced motion generation in video models.arXiv preprint arXiv:2502.02492, 2025

    Hila Chefer, Uriel Singer, Amit Zohar, Yuval Kirstain, Adam Polyak, Yaniv Taigman, Lior Wolf, and Shelly Sheynin. Videojam: Joint appearance-motion representations for en- hanced motion generation in video models.arXiv preprint arXiv:2502.02492, 2025. 1, 2

    work page arXiv 2025

  8. [8]

    Videocrafter2: Overcoming data limitations for high-quality video diffusion models

    Haoxin Chen, Yong Zhang, Xiaodong Cun, Menghan Xia, Xintao Wang, Chao Weng, and Ying Shan. Videocrafter2: Overcoming data limitations for high-quality video diffusion models. InCVPR, pages 7310–7320, 2024. 6

    2024

  9. [9]

    A semi- implicit material point method for the continuum simula- tion of granular materials.ACM Transactions on Graphics (TOG), 35(4):1–13, 2016

    Gilles Daviet and Florence Bertails-Descoubes. A semi- implicit material point method for the continuum simula- tion of granular materials.ACM Transactions on Graphics (TOG), 35(4):1–13, 2016. 2

    2016

  10. [10]

    Autoregressive Video Generation without Vector Quantization

    Haoge Deng, Ting Pan, Haiwen Diao, Zhengxiong Luo, Yufeng Cui, Huchuan Lu, Shiguang Shan, Yonggang Qi, and Xinlong Wang. Autoregressive video generation with- out vector quantization.arXiv preprint arXiv:2412.14169,

    work page internal anchor Pith review Pith/arXiv arXiv

  11. [11]

    Scaling recti- fied flow transformers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas M ¨uller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling recti- fied flow transformers for high-resolution image synthesis. InForty-first international conference on machine learning,

  12. [12]

    Relational rep- resentation distillation.arXiv preprint arXiv:2407.12073,

    Nikolaos Giakoumoglou and Tania Stathaki. Relational rep- resentation distillation.arXiv preprint arXiv:2407.12073,

    work page arXiv

  13. [13]

    Distilling the Knowledge in a Neural Network

    Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. Distill- ing the knowledge in a neural network.arXiv preprint arXiv:1503.02531, 2015. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  14. [14]

    Vbench: Comprehensive bench- mark suite for video generative models

    Ziqi Huang, Yinan He, Jiashuo Yu, Fan Zhang, Chenyang Si, Yuming Jiang, Yuanhan Zhang, Tianxing Wu, Qingyang Jin, Nattapol Chanpaisit, et al. Vbench: Comprehensive bench- mark suite for video generative models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21807–21818, 2024. 6

    2024

  15. [15]

    Cross-frame representation alignment for fine-tuning video diffusion models.arXiv preprint arXiv:2506.09229, 2025

    Sungwon Hwang, Hyojin Jang, Kinam Kim, Minho Park, and Jaegul Choo. Cross-frame representation alignment for fine-tuning video diffusion models.arXiv preprint arXiv:2506.09229, 2025. 2, 3

    work page arXiv 2025

  16. [16]

    Track4gen: Teaching video dif- fusion models to track points improves video generation

    Hyeonho Jeong, Chun-Hao P Huang, Jong Chul Ye, Niloy J Mitra, and Duygu Ceylan. Track4gen: Teaching video dif- fusion models to track points improves video generation. In CVPR, pages 7276–7287, 2025. 1, 2

    2025

  17. [17]

    Pyramidal flow matching for efficient video generative modeling

    Yang Jin, Zhicheng Sun, Ningyuan Li, Kun Xu, Hao Jiang, Nan Zhuang, Quzhe Huang, Yang Song, Yadong Mu, and Zhouchen Lin. Pyramidal flow matching for efficient video generative modeling.arXiv preprint arXiv:2410.05954,

    work page arXiv

  18. [18]

    Cross-view action recognition from temporal self- similarities

    Imran N Junejo, Emilie Dexter, Ivan Laptev, and Patrick P ´Urez. Cross-view action recognition from temporal self- similarities. InECCV, 2008. 2

    2008

  19. [19]

    View-independent action recognition from temporal self-similarities.IEEE TPAMI, 2010

    Imran N Junejo, Emilie Dexter, Ivan Laptev, and Patrick Perez. View-independent action recognition from temporal self-similarities.IEEE TPAMI, 2010. 2

    2010

  20. [20]

    Relational embedding for few-shot classification

    Dahyun Kang, Heeseung Kwon, Juhong Min, and Minsu Cho. Relational embedding for few-shot classification. In ICCV, pages 8822–8833, 2021. 2

    2021

  21. [21]

    Relational self-attention: What’s missing in attention for video understanding.NeurIPS, 34:8046–8059,

    Manjin Kim, Heeseung Kwon, Chunyu Wang, Suha Kwak, and Minsu Cho. Relational self-attention: What’s missing in attention for video understanding.NeurIPS, 34:8046–8059,

  22. [22]

    Learning correlation structures for vision trans- formers

    Manjin Kim, Paul Hongsuck Seo, Cordelia Schmid, and Minsu Cho. Learning correlation structures for vision trans- formers. InCVPR, pages 18941–18951, 2024. 2

    2024

  23. [23]

    Fcss: Fully con- volutional self-similarity for dense semantic correspondence

    Seungryong Kim, Dongbo Min, Bumsub Ham, Sangryul Jeon, Stephen Lin, and Kwanghoon Sohn. Fcss: Fully con- volutional self-similarity for dense semantic correspondence. InCVPR, 2017. 2

    2017

  24. [24]

    HunyuanVideo: A Systematic Framework For Large Video Generative Models

    Weijie Kong, Qi Tian, Zijian Zhang, Rox Min, Zuozhuo Dai, Jin Zhou, Jiangfeng Xiong, Xin Li, Bo Wu, Jianwei Zhang, et al. Hunyuanvideo: A systematic framework for large video generative models.arXiv preprint arXiv:2412.03603, 2024. 1, 2, 3, 6, 12

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  25. [25]

    Motionsqueeze: Neural motion feature learning for video understanding.arXiv preprint arXiv:2007.09933, 2020

    Heeseung Kwon, Manjin Kim, Suha Kwak, and Minsu Cho. Motionsqueeze: Neural motion feature learning for video understanding.arXiv preprint arXiv:2007.09933, 2020. 2

    work page arXiv 2007

  26. [26]

    Learning self-similarity in space and time as gen- eralized motion for action recognition.arXiv preprint arXiv:2102.07092, 2021

    Heeseung Kwon, Manjin Kim, Suha Kwak, and Minsu Cho. Learning self-similarity in space and time as gen- eralized motion for action recognition.arXiv preprint arXiv:2102.07092, 2021. 2

    work page arXiv 2021

  27. [27]

    REPA-E: Unlocking V AE for end-to-end tuning with latent diffusion transformers,

    Xingjian Leng, Jaskirat Singh, Yunzhong Hou, Zhenchang Xing, Saining Xie, and Liang Zheng. Repa-e: Unlocking vae for end-to-end tuning with latent diffusion transformers. arXiv preprint arXiv:2504.10483, 2025. 3

    work page arXiv 2025

  28. [28]

    Phy124: Fast physics-driven 4d con- tent generation from a single image.arXiv preprint arXiv:2409.07179, 2024

    Jiajing Lin, Zhenzhong Wang, Yongjie Hou, Yuzhou Tang, and Min Jiang. Phy124: Fast physics-driven 4d con- tent generation from a single image.arXiv preprint arXiv:2409.07179, 2024. 1, 2

    work page arXiv 2024

  29. [29]

    Phys4dgen: A physics-driven framework for con- trollable and efficient 4d content generation from a single image.arXiv e-prints, pages arXiv–2411, 2024

    Jiajing Lin, Zhenzhong Wang, Shu Jiang, Yongjie Hou, and Min Jiang. Phys4dgen: A physics-driven framework for con- trollable and efficient 4d content generation from a single image.arXiv e-prints, pages arXiv–2411, 2024. 1, 2

    2024

  30. [30]

    Physgen: Rigid-body physics-grounded image- to-video generation

    Shaowei Liu, Zhongzheng Ren, Saurabh Gupta, and Shen- long Wang. Physgen: Rigid-body physics-grounded image- to-video generation. InECCV, pages 360–378. Springer, 2024

    2024

  31. [31]

    Unleashing the potential of multi-modal foun- dation models and video diffusion for 4d dynamic physical scene simulation

    Zhuoman Liu, Weicai Ye, Yan Luximon, Pengfei Wan, and Di Zhang. Unleashing the potential of multi-modal foun- dation models and video diffusion for 4d dynamic physical scene simulation. InCVPR, pages 11016–11025, 2025. 1, 2

    2025

  32. [32]

    Decoupled Weight Decay Regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017. 5

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  33. [33]

    Dream machine — ai video generator, 2024

    Luma AI. Dream machine — ai video generator, 2024. 6

    2024

  34. [34]

    Towards World Simulator: Crafting Physical Commonsense-Based Benchmark for Video Generation

    Fanqing Meng, Jiaqi Liao, Xinyu Tan, Wenqi Shao, Quan- feng Lu, Kaipeng Zhang, Yu Cheng, Dianqi Li, Yu Qiao, and Ping Luo. Towards world simulator: Crafting physical commonsense-based benchmark for video generation.arXiv preprint arXiv:2410.05363, 2024. 1, 2

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  35. [35]

    Motioncraft: Physics- based zero-shot video generation.Advances in Neural In- formation Processing Systems, 37:123155–123181, 2024

    Antonio Montanaro, Luca Savant Aira, Emanuele Aiello, Diego Valsesia, and Enrico Magli. Motioncraft: Physics- based zero-shot video generation.Advances in Neural In- formation Processing Systems, 37:123155–123181, 2024. 1, 2

    2024

  36. [36]

    Do generative video models understand physical principles?

    Saman Motamed, Laura Culp, Kevin Swersky, Priyank Jaini, and Robert Geirhos. Do generative video models learn physical principles from watching videos?, 2025.URL https://arxiv. org/abs/2501.09038, 2025. 1, 2

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  37. [37]

    OpenVid-1M: A Large-Scale High-Quality Dataset for Text-to-video Generation

    Kepan Nan, Rui Xie, Penghao Zhou, Tiehan Fan, Zhen- heng Yang, Zhijie Chen, Xiang Li, Jian Yang, and Ying Tai. Openvid-1m: A large-scale high-quality dataset for text-to- video generation.arXiv preprint arXiv:2407.02371, 2024. 5

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  38. [38]

    DINOv2: Learning Robust Visual Features without Supervision

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023. 3

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  39. [39]

    Relational knowledge distillation

    Wonpyo Park, Dongju Kim, Yan Lu, and Minsu Cho. Relational knowledge distillation. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3967–3976, 2019. 2

    2019

  40. [40]

    A material point method for viscoelastic fluids, foams and sponges

    Daniel Ram, Theodore Gast, Chenfanfu Jiang, Craig Schroeder, Alexey Stomakhin, Joseph Teran, and Pirouz Kavehpour. A material point method for viscoelastic fluids, foams and sponges. InProceedings of the 14th ACM SIG- GRAPH/Eurographics Symposium on Computer Animation, pages 157–163, 2015. 2

    2015

  41. [41]

    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. InCVPR, pages 10684– 10695, 2022. 3

    2022

  42. [42]

    Space-time behavior based correlation

    Eli Shechtman and Michal Irani. Space-time behavior based correlation. InCVPR, pages 405–412. IEEE, 2005. 2

    2005

  43. [43]

    Matching local self- similarities across images and videos

    Eli Shechtman and Michal Irani. Matching local self- similarities across images and videos. InCVPR, 2007. 2

    2007

  44. [44]

    A material point method for snow simulation.ACM Transactions on Graphics (TOG), 32(4): 1–10, 2013

    Alexey Stomakhin, Craig Schroeder, Lawrence Chai, Joseph Teran, and Andrew Selle. A material point method for snow simulation.ACM Transactions on Graphics (TOG), 32(4): 1–10, 2013. 2

    2013

  45. [45]

    Local self-similarity-based registration of human rois in pairs of stereo thermal-visible videos.Pattern Recognition, 46(2): 578–589, 2013

    Atousa Torabi and Guillaume-Alexandre Bilodeau. Local self-similarity-based registration of human rois in pairs of stereo thermal-visible videos.Pattern Recognition, 46(2): 578–589, 2013. 2

    2013

  46. [46]

    Similarity-preserving knowl- edge distillation

    Frederick Tung and Greg Mori. Similarity-preserving knowl- edge distillation. InProceedings of the IEEE/CVF inter- national conference on computer vision, pages 1365–1374,

  47. [47]

    Wan: Open and Advanced Large-Scale Video Generative Models

    Team Wan, Ang Wang, Baole Ai, Bin Wen, Chaojie Mao, Chen-Wei Xie, Di Chen, Feiwu Yu, Haiming Zhao, Jianxiao Yang, et al. Wan: Open and advanced large-scale video gen- erative models.arXiv preprint arXiv:2503.20314, 2025. 1, 2, 3

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  48. [48]

    Video modeling with correlation networks

    Heng Wang, Du Tran, Lorenzo Torresani, and Matt Feiszli. Video modeling with correlation networks. InCVPR, 2020. 2

    2020

  49. [49]

    Wisa: World simulator assistant for physics-aware text-to-video generation.arXiv preprint arXiv:2503.08153, 2025

    Jing Wang, Ao Ma, Ke Cao, Jun Zheng, Zhanjie Zhang, Jiasong Feng, Shanyuan Liu, Yuhang Ma, Bo Cheng, Dawei Leng, et al. Wisa: World simulator assistant for physics-aware text-to-video generation.arXiv preprint arXiv:2503.08153, 2025. 1, 2

    work page arXiv 2025

  50. [50]

    Videomae v2: Scaling video masked autoencoders with dual masking

    Limin Wang, Bingkun Huang, Zhiyu Zhao, Zhan Tong, Yi- nan He, Yi Wang, Yali Wang, and Yu Qiao. Videomae v2: Scaling video masked autoencoders with dual masking. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 14549–14560, 2023. 5

    2023

  51. [51]

    Geometry Forcing: Marrying Video Diffusion and 3D Representation for Consistent World Modeling

    Haoyu Wu, Diankun Wu, Tianyu He, Junliang Guo, Yang Ye, Yueqi Duan, and Jiang Bian. Geometry forcing: Marrying video diffusion and 3d representation for consistent world modeling.arXiv preprint arXiv:2507.07982, 2025. 2, 3

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  52. [52]

    What can simple arithmetic opera- tions do for temporal modeling? InICCV, pages 13712– 13722, 2023

    Wenhao Wu, Yuxin Song, Zhun Sun, Jingdong Wang, Chang Xu, and Wanli Ouyang. What can simple arithmetic opera- tions do for temporal modeling? InICCV, pages 13712– 13722, 2023. 2

    2023

  53. [53]

    Physgaussian: Physics- integrated 3d gaussians for generative dynamics

    Tianyi Xie, Zeshun Zong, Yuxing Qiu, Xuan Li, Yutao Feng, Yin Yang, and Chenfanfu Jiang. Physgaussian: Physics- integrated 3d gaussians for generative dynamics. InCVPR, pages 4389–4398, 2024. 1, 2

    2024

  54. [54]

    Physanimator: Physics-guided generative cartoon animation

    Tianyi Xie, Yiwei Zhao, Ying Jiang, and Chenfanfu Jiang. Physanimator: Physics-guided generative cartoon animation. InCVPR, pages 10793–10804, 2025. 1, 2

    2025

  55. [55]

    Phyt2v: Llm-guided iterative self-refinement for physics- grounded text-to-video generation

    Qiyao Xue, Xiangyu Yin, Boyuan Yang, and Wei Gao. Phyt2v: Llm-guided iterative self-refinement for physics- grounded text-to-video generation. InCVPR, pages 18826– 18836, 2025. 1, 2, 6

    2025

  56. [56]

    CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer

    Zhuoyi Yang, Jiayan Teng, Wendi Zheng, Ming Ding, Shiyu Huang, Jiazheng Xu, Yuanming Yang, Wenyi Hong, Xiao- han Zhang, Guanyu Feng, et al. Cogvideox: Text-to-video diffusion models with an expert transformer.arXiv preprint arXiv:2408.06072, 2024. 1, 2, 3, 5, 6, 7, 12

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  57. [57]

    From slow bidirectional to fast causal video generators.arXiv e- prints, pages arXiv–2412, 2024

    Tianwei Yin, Qiang Zhang, Richard Zhang, William T Free- man, Fredo Durand, Eli Shechtman, and Xun Huang. From slow bidirectional to fast causal video generators.arXiv e- prints, pages arXiv–2412, 2024. 1, 2

    2024

  58. [58]

    Representation Alignment for Generation: Training Diffusion Transformers Is Easier Than You Think

    Sihyun Yu, Sangkyung Kwak, Huiwon Jang, Jongheon Jeong, Jonathan Huang, Jinwoo Shin, and Saining Xie. Representation alignment for generation: Training diffu- sion transformers is easier than you think.arXiv preprint arXiv:2410.06940, 2024. 2, 3, 6, 7, 12

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  59. [59]

    Paying More Attention to Attention: Improving the Performance of Convolutional Neural Networks via Attention Transfer

    Sergey Zagoruyko and Nikos Komodakis. Paying more at- tention to attention: Improving the performance of convolu- tional neural networks via attention transfer.arXiv preprint arXiv:1612.03928, 2016. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  60. [60]

    Physdreamer: Physics-based interac- tion with 3d objects via video generation

    Tianyuan Zhang, Hong-Xing Yu, Rundi Wu, Brandon Y Feng, Changxi Zheng, Noah Snavely, Jiajun Wu, and William T Freeman. Physdreamer: Physics-based interac- tion with 3d objects via video generation. InECCV, pages 388–406. Springer, 2024. 1, 2

    2024

  61. [61]

    Vrm: Knowledge distillation via virtual relation matching

    Weijia Zhang, Fei Xie, Weidong Cai, and Chao Ma. Vrm: Knowledge distillation via virtual relation matching. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 2707–2717, 2025. 2

    2025

  62. [62]

    arXiv preprint arXiv:2505.23656 (2025)

    Xiangdong Zhang, Jiaqi Liao, Shaofeng Zhang, Fanqing Meng, Xiangpeng Wan, Junchi Yan, and Yu Cheng. Vide- orepa: Learning physics for video generation through re- lational alignment with foundation models.arXiv preprint arXiv:2505.23656, 2025. 2, 3, 4, 5, 6, 7, 12 A. Additional Qualitative Results In Fig. 7, we provide additional qualitative comparisons o...

    work page arXiv 2025