pith. machine review for the scientific record. sign in

arxiv: 2603.29092 · v2 · submitted 2026-03-31 · 💻 cs.CV

Recognition: no theorem link

TrajectoryMover: Generative Movement of Object Trajectories in Videos

Kiran Chhatre , Hyeonho Jeong , Yulia Gryaditskaya , Christopher E. Peters , Chun-Hao Paul Huang , Paul Guerrero
Authors on Pith no claims yet

Pith reviewed 2026-05-14 00:15 UTC · model grok-4.3

classification 💻 cs.CV
keywords video editinggenerative modelsobject trajectorysynthetic data3D motionpaired video dataTrajectoryMover
0
0 comments X

The pith

A new video generator uses synthetic paired data to move objects along altered 3D trajectories while keeping their original motion intact.

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

The paper introduces TrajectoryAtlas, a pipeline that produces large-scale synthetic paired videos showing the same scene with an object on two different 3D paths. It then fine-tunes a video model called TrajectoryMover on this data so the model learns to apply new trajectories to real footage. Earlier editing methods could prescribe 2D or 3D paths but could not reliably move an existing 3D motion pattern to a new path because they lacked suitable training pairs. If the synthetic data transfers well, the approach gives non-expert users a way to reposition moving objects in short clips without breaking scene consistency or motion plausibility.

Core claim

We introduce TrajectoryAtlas, a new data generation pipeline for large-scale synthetic paired video data and a video generator TrajectoryMover fine-tuned with this data. We show that this successfully enables generative movement of object trajectories.

What carries the argument

TrajectoryAtlas, a synthetic data pipeline that creates paired videos of the same scene with an object following two different 3D trajectories, used to fine-tune TrajectoryMover for trajectory editing.

Load-bearing premise

Synthetic paired videos created by TrajectoryAtlas are realistic and diverse enough that fine-tuning on them lets TrajectoryMover edit real videos without visible artifacts or implausible motion.

What would settle it

Apply TrajectoryMover to a set of real videos with known ground-truth object paths and measure whether the output motion matches the new target trajectory while preserving the original relative 3D dynamics and avoiding artifacts.

Figures

Figures reproduced from arXiv: 2603.29092 by Christopher E. Peters, Chun-Hao Paul Huang, Hyeonho Jeong, Kiran Chhatre, Paul Guerrero, Yulia Gryaditskaya.

Figure 1
Figure 1. Figure 1: TrajectoryMover enables intuitive video editing by allowing users to translate an object’s 3D motion path to a new starting location using simple bounding box controls across diverse and complex scenarios, including drop, roll, and drag motions. Our model successfully aligns the generated trajectory with the target initial location. Furthermore, the model dynamically adapts the motion to the new path to en… view at source ↗
Figure 2
Figure 2. Figure 2: TrajectoryAtlas data generation pipeline. The pipeline has five stages, Asset Cache Preparation, Preflight Validation, Collision Aware Sampling and Scaling, Task Simulation, and Canonical Rendering with Runtime Metadata. Inputs including camera, 3D scene, lights and materials, and Objaverse or primitive assets are converted to reusable collision caches, then skip render preflight selects valid frames. Pair… view at source ↗
Figure 3
Figure 3. Figure 3: TrajectoryMover architecture. We concatenate three latent streams ztrj, zsrc, and zbb before denoising. In the control image, red marks the source box and green marks the target box. Data generation. TrajectoryAtlas uses Blender (Cycles) for rendering and Py￾Bullet for physics. We use curated Ev￾ermotion [13] indoor scenes and a fore￾ground object pool of 119 assets, with 98 Objaverse objects [11] and 21 p… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison with baselines. We compare TrajectoryMover with SFM, ATI, DaS, VACE, and I2VEdit on four representative motion scenarios. Red boxes indicate the source object location in the input video, green boxes indicate the target location at frame 0, pink boxes highlight regions of failure, and cyan boxes highlight regions of success. TrajectoryMover follows the intended motion most consistent… view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative ablation analysis. We compare the full model with ablations using only primitives, only scene modification, without scene modification, and drop￾only motion training. Red boxes indicate source object location, green boxes indicate target frame-0 location, and pink boxes mark representative regions of failure while cyan boxes highlight region of success results. The full model gives the best bal… view at source ↗
read the original abstract

Generative video editing has enabled several intuitive editing operations for short video clips that would previously have been difficult to achieve, especially for non-expert editors. Existing methods focus on prescribing an object's 3D or 2D motion trajectory in a video, or on altering the appearance of an object or a scene, while preserving both the video's plausibility and identity. Yet a method to move an object's 3D motion trajectory in a video, i.e., moving an object while preserving its relative 3D motion, is currently still missing. The main challenge lies in obtaining paired video data for this scenario. Previous methods typically rely on clever data generation approaches to construct plausible paired data from unpaired videos, but this approach fails if one of the videos in a pair can not easily be constructed from the other. Instead, we introduce TrajectoryAtlas, a new data generation pipeline for large-scale synthetic paired video data and a video generator TrajectoryMover fine-tuned with this data. We show that this successfully enables generative movement of object trajectories. Project page: https://chhatrekiran.github.io/trajectorymover

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 introduces TrajectoryAtlas, a data generation pipeline for large-scale synthetic paired video data, along with TrajectoryMover, a video generator fine-tuned on this data. The central claim is that this enables generative editing of videos to move an object's 3D motion trajectory while preserving its relative 3D motion, identity, and overall plausibility, addressing a gap where prior methods could not easily construct such paired data from unpaired videos.

Significance. If the result holds, the work would provide a practical solution for 3D trajectory manipulation in generative video editing, which is currently missing from the literature. The strength lies in the explicit construction of paired synthetic data rather than relying on clever unpaired-to-paired conversions; however, the significance is tempered by the lack of demonstrated generalization to real videos.

major comments (2)
  1. [Abstract] Abstract: The claim that TrajectoryMover 'successfully enables generative movement of object trajectories' is asserted without any quantitative support (e.g., real-video FID, optical flow consistency, or user-study metrics). This is load-bearing because the entire contribution rests on the unverified assumption that synthetic paired data from TrajectoryAtlas generalizes without introducing artifacts or implausible motion on real inputs.
  2. [Introduction] Introduction / Method: No details are provided on how TrajectoryAtlas ensures sufficient diversity in lighting, physics, and appearance to close the synthetic-to-real domain gap. Without such analysis or ablation, it is impossible to evaluate whether the fine-tuning step actually bridges the gap or merely overfits to synthetic statistics.
minor comments (1)
  1. [Abstract] The project page link is given but no quantitative results or failure cases are referenced in the abstract; including a brief summary of key metrics would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive report. We agree that stronger quantitative evidence on real-video generalization is needed and that the TrajectoryAtlas pipeline requires more explicit description of its diversity mechanisms. We will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The claim that TrajectoryMover 'successfully enables generative movement of object trajectories' is asserted without any quantitative support (e.g., real-video FID, optical flow consistency, or user-study metrics). This is load-bearing because the entire contribution rests on the unverified assumption that synthetic paired data from TrajectoryAtlas generalizes without introducing artifacts or implausible motion on real inputs.

    Authors: We acknowledge that the current abstract and results section rely primarily on qualitative demonstrations and synthetic quantitative metrics. In the revision we will add real-video FID scores, optical-flow consistency measurements, and a user study comparing TrajectoryMover outputs against baselines on real inputs. These experiments have already been run and confirm that the synthetic-to-real gap is bridged without introducing systematic artifacts. revision: yes

  2. Referee: [Introduction] Introduction / Method: No details are provided on how TrajectoryAtlas ensures sufficient diversity in lighting, physics, and appearance to close the synthetic-to-real domain gap. Without such analysis or ablation, it is impossible to evaluate whether the fine-tuning step actually bridges the gap or merely overfits to synthetic statistics.

    Authors: We will expand the TrajectoryAtlas section with a dedicated paragraph and table enumerating the diversity parameters: randomized HDR lighting maps, varied physical material properties and gravity scales, 500+ distinct 3D object models with texture randomization, and procedural scene layouts. We will also include an ablation that trains on progressively less diverse subsets and measures the resulting drop in real-video metrics, demonstrating that the full diversity set is required to close the domain gap. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained via new synthetic data construction

full rationale

The paper introduces TrajectoryAtlas as a new pipeline to generate large-scale synthetic paired video data and fine-tunes TrajectoryMover on it to enable generative object trajectory movement. No equations, parameter fittings, or derivations are described that reduce any prediction or result to the inputs by construction. The central approach relies on explicit synthetic data synthesis rather than self-citation chains, ansatzes smuggled via prior work, or renaming known results. Claims of success are presented as empirical outcomes from this pipeline, with no load-bearing steps that collapse to tautology. This matches the default expectation of a non-circular data-driven method.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 2 invented entities

The claim depends on the assumption that synthetic paired data can substitute for real paired data in fine-tuning a video generator, plus the new entities TrajectoryAtlas and TrajectoryMover.

axioms (1)
  • domain assumption Synthetic video pairs generated from unpaired sources can train models that generalize to real videos for trajectory editing.
    Invoked to justify the data pipeline and fine-tuning strategy.
invented entities (2)
  • TrajectoryAtlas no independent evidence
    purpose: Pipeline to generate large-scale synthetic paired video data for trajectory movement
    Newly proposed data generation method
  • TrajectoryMover no independent evidence
    purpose: Fine-tuned video generator enabling trajectory relocation
    Newly proposed model

pith-pipeline@v0.9.0 · 5513 in / 1115 out tokens · 56544 ms · 2026-05-14T00:15:50.530988+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

42 extracted references · 42 canonical work pages · 2 internal anchors

  1. [1]

    AI, D.: Open-weight text-guided video editing (2025), https://platform.decart.ai/

    work page 2025

  2. [2]

    ICCV (2025)

    Bai, J., Xia, M., Fu, X., Wang, X., Mu, L., Cao, J., Liu, Z., Hu, H., Bai, X., Wan, P., et al.: Recammaster: Camera-controlled generative rendering from a single video. ICCV (2025)

    work page 2025

  3. [3]

    the method of paired comparisons

    Bradley, R.A., Terry, M.E.: Rank analysis of incomplete block designs i. the method of paired comparisons. Biometrika39(3/4), 324–345 (1952)

    work page 1952

  4. [4]

    arXiv preprint arXiv:2511.20640 (2025)

    Burgert, R., Herrmann, C., Cole, F., Ryoo, M.S., Wadhwa, N., Voynov, A., Ruiz, N.: Motionv2v: Editing motion in a video. arXiv preprint arXiv:2511.20640 (2025)

    work page arXiv 2025

  5. [5]

    Carion, N., Gustafson, L., Hu, Y.T., Debnath, S., Hu, R., Suris, D., Ryali, C., Alwala, K.V., Khedr, H., Huang, A., Lei, J., Ma, T., Guo, B., Kalla, A., Marks, M., Greer, J., Wang, M., Sun, P., Rädle, R., Afouras, T., Mavroudi, E., Xu, K., Wu, T.H., Zhou, Y., Momeni, L., Hazra, R., Ding, S., Vaze, S., Porcher, F., Li, F., Li, S., Kamath, A., Cheng, H.K., ...

    work page 2026

  6. [6]

    CVPR (2025)

    Chen, S., Guo, H., Zhu, S., Zhang, F., Huang, Z., Feng, J., Kang, B.: Video depth anything: Consistent depth estimation for super-long videos. CVPR (2025)

    work page 2025

  7. [7]

    arXiv preprint arXiv:2305.13840 (2023)

    Chen, W., Ji, Y., Wu, J., Wu, H., Xie, P., Li, J., Xia, X., Xiao, X., Lin, L.: Control- a-video: Controllable text-to-video diffusion models with motion prior and reward feedback learning. arXiv preprint arXiv:2305.13840 (2023)

    work page arXiv 2023

  8. [8]

    In: CVPR

    Chen, Z., Wu, J., Wang, W., Su, W., Chen, G., Xing, S., Zhong, M., Zhang, Q., Zhu, X., Lu, L., et al.: Internvl: Scaling up vision foundation models and aligning for generic visual-linguistic tasks. In: CVPR. pp. 24185–24198 (2024)

    work page 2024

  9. [9]

    ICLR (2023)

    Cong, Y., Xu, M., Simon, C., Chen, S., Ren, J., Xie, Y., Perez-Rua, J.M., Rosenhahn, B., Xiang, T., He, S.: Flatten: optical flow-guided attention for consistent text-to- video editing. ICLR (2023)

    work page 2023

  10. [10]

    Coumans, E., Bai, Y.: Pybullet, a python module for physics simulation for games, robotics and machine learning.http://pybullet.org (2016–2019)

    work page 2016

  11. [11]

    CVPR (2023)

    Deitke, M., Schwenk, D., Salvador, J., Weihs, L., Michel, O., VanderBilt, E., Schmidt, L., Ehsani, K., Kembhavi, A., Farhadi, A.: Objaverse: A universe of annotated 3d objects. CVPR (2023)

    work page 2023

  12. [12]

    In: ECCV

    Deng, Y., Wang, R., Zhang, Y., Tai, Y.W., Tang, C.K.: Dragvideo: Interactive drag-style video editing. In: ECCV. pp. 183–199. Springer (2024)

    work page 2024

  13. [13]

    https://evermotion.org/, accessed: 2026-03-05 16 K

    Evermotion: Evermotion. https://evermotion.org/, accessed: 2026-03-05 16 K. Chhatre et al

    work page 2026

  14. [14]

    ACM SIGGRAPH 2025 Conference Papers (2025)

    Gu, Z., Yan, R., Lu, J., Li, P., Dou, Z., Si, C., Dong, Z., Liu, Q., Lin, C., Liu, Z., Wang, W., Liu, Y.: Diffusion as shader: 3d-aware video diffusion for versatile video generation control. ACM SIGGRAPH 2025 Conference Papers (2025)

    work page 2025

  15. [15]

    In: ACM SIGGRAPH Asia 2025 Conference Papers

    Gu, Z., Yan, R., Lu, J., Li, P., Dou, Z., Si, C., Dong, Z., Liu, Q., Lin, C., Liu, Z., et al.: Diffusion as shader: 3d-aware video diffusion for versatile video generation control. In: ACM SIGGRAPH Asia 2025 Conference Papers. pp. 1–12 (2025)

    work page 2025

  16. [16]

    arXiv preprint arXiv:2512.25075 (2025)

    Huang, Z., Jeong, H., Chen, X., Gryaditskaya, Y., Wang, T.Y., Lasenby, J., Huang, C.H.: Spacetimepilot: Generative rendering of dynamic scenes across space and time. arXiv preprint arXiv:2512.25075 (2025)

    work page arXiv 2025

  17. [17]

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

    Jeong, H., Lee, S., Ye, J.C.: Reangle-a-video: 4d video generation as video-to- video translation. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 11164–11175 (2025)

    work page 2025

  18. [18]

    ICLR (2023)

    Jeong, H., Ye, J.C.: Ground-a-video: Zero-shot grounded video editing using text- to-image diffusion models. ICLR (2023)

    work page 2023

  19. [19]

    ICCV (2025)

    Jiang, Z., Han, Z., Mao, C., Zhang, J., Pan, Y., Liu, Y.: Vace: All-in-one video creation and editing. ICCV (2025)

    work page 2025

  20. [20]

    Jose, C., Moutakanni, T., Kang, D., Baldassarre, F., Darcet, T., Xu, H., Li, D., Szafraniec, M., Ramamonjisoa, M., Oquab, M., Siméoni, O., Vo, H.V., Labatut, P., Bojanowski, P.: Dinov2 meets text: A unified framework for image- and pixel-level vision-language alignment (2024)

    work page 2024

  21. [21]

    EditVerse: Unifying Image and Video Editing and Generation with In-Context Learning

    Ju, X., Wang, T., Zhou, Y., Zhang, H., Liu, Q., Zhao, N., Zhang, Z., Li, Y., Cai, Y., Liu, S., et al.: Editverse: Unifying image and video editing and generation with in-context learning. arXiv preprint arXiv:2509.20360 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  22. [22]

    In: CVPR (2025)

    Koo, J., Guerrero, P., Huang, C.H.P., Ceylan, D., Sung, M.: Videohandles: Editing 3d object compositions in videos using video generative priors. In: CVPR (2025)

    work page 2025

  23. [23]

    In: CVPR

    Koroglu, M., Caselles-Dupré, H., Jeanneret, G., Cord, M.: Onlyflow: Optical flow based motion conditioning for video diffusion models. In: CVPR. pp. 6226–6236 (2025)

    work page 2025

  24. [24]

    arXiv preprint arXiv:2512.02015 (2025) 2

    Lee, Y.C., Zhang, Z., Huang, J., Wang, J.H., Lee, J.Y., Huang, J.B., Shechtman, E., Li, Z.: Generative video motion editing with 3d point tracks. arXiv preprint arXiv:2512.02015 (2025)

    work page arXiv 2025

  25. [25]

    arXiv preprint arXiv:2601.02785 (2026)

    Li, M., Chen, J., Zhao, S., Feng, W., Tu, P., He, Q.: Dreamstyle: A unified framework for video stylization. arXiv preprint arXiv:2601.02785 (2026)

    work page arXiv 2026

  26. [26]

    CVPR (2025)

    Liu, S., Wang, T., Wang, J.H., Liu, Q., Zhang, Z., Lee, J.Y., Li, Y., Yu, B., Lin, Z., Kim, S.Y., Jia, J.: Generative video propagation. CVPR (2025)

    work page 2025

  27. [27]

    ACM SIGGRAPH Asia 2025 Conference Papers (2025)

    Liu, Y., Wang, T., Liu, F., Wang, Z., Lau, R.W.: Shape-for-motion: Precise and consistent video editing with 3d proxy. ACM SIGGRAPH Asia 2025 Conference Papers (2025)

    work page 2025

  28. [28]

    John Wiley & Sons, New York (1959)

    Luce, R.D.: Individual Choice Behavior: A Theoretical Analysis. John Wiley & Sons, New York (1959)

    work page 1959

  29. [29]

    In: Advances in Neural Information Processing Systems 28 (2015)

    Maystre, L., Grossglauser, M.: Fast and accurate inference of plackett–luce models. In: Advances in Neural Information Processing Systems 28 (2015)

    work page 2015

  30. [30]

    Oquab, M., Darcet, T., Moutakanni, T., Vo, H.V., Szafraniec, M., Khalidov, V., Fernandez, P., Haziza, D., Massa, F., El-Nouby, A., Howes, R., Huang, P.Y., Xu, H., Sharma, V., Li, S.W., Galuba, W., Rabbat, M., Assran, M., Ballas, N., Synnaeve, G., Misra, I., Jegou, H., Mairal, J., Labatut, P., Joulin, A., Bojanowski, P.: Dinov2: Learning robust visual feat...

    work page 2023

  31. [31]

    ACM SIGGRAPH Asia 2024 Conference Papers (2024) TrajectoryMover: Generative Object Trajectory Movement in Videos 17

    Ouyang, W., Dong, Y., Yang, L., Si, J., Pan, X.: I2vedit: First-frame-guided video editing via image-to-video diffusion models. ACM SIGGRAPH Asia 2024 Conference Papers (2024) TrajectoryMover: Generative Object Trajectory Movement in Videos 17

    work page 2024

  32. [32]

    arXiv preprint arXiv:2511.19827 (2025)

    Park, B., Kim, B.H., Chung, H., Ye, J.C.: Redirector: Creating any-length video retakes with rotary camera encoding. arXiv preprint arXiv:2511.19827 (2025)

    work page arXiv 2025

  33. [33]

    In: ACM SIGGRAPH 2024 Conference Papers

    Shi, X., Huang, Z., Wang, F.Y., Bian, W., Li, D., Zhang, Y., Zhang, M., Cheung, K.C., See, S., Qin, H., et al.: Motion-i2v: Consistent and controllable image-to-video generation with explicit motion modeling. In: ACM SIGGRAPH 2024 Conference Papers. pp. 1–11 (2024)

    work page 2024

  34. [34]

    arXiv preprint:2511.01266 (2025)

    Shin, J., Li, Z., Zhang, R., Zhu, J.Y., Park, J., Shechtman, E., Huang, X.: Motion- stream: Real-time video generation with interactive motion controls. arXiv preprint arXiv:2511.01266 (2025)

    work page arXiv 2025

  35. [35]

    Neurocomputing568, 127063 (2024)

    Su, J., Ahmed, M., Lu, Y., Pan, S., Bo, W., Liu, Y.: Roformer: Enhanced transformer with rotary position embedding. Neurocomputing568, 127063 (2024)

    work page 2024

  36. [36]

    WACV (2021)

    Suvorov, R., Logacheva, E., Mashikhin, A., Remizova, A., Ashukha, A., Silvestrov, A., Kong, N., Goka, H., Park, K., Lempitsky, V.: Resolution-robust large mask inpainting with fourier convolutions. WACV (2021)

    work page 2021

  37. [37]

    arXiv preprint arXiv:2312.02936 (2023)

    Teng, Y., Xie, E., Wu, Y., Han, H., Li, Z., Liu, X.: Drag-a-video: Non-rigid video editing with point-based interaction. arXiv preprint arXiv:2312.02936 (2023)

    work page arXiv 2023

  38. [38]

    Wan: Open and Advanced Large-Scale Video Generative Models

    Wan, T., Wang, A., Ai, B., Wen, B., Mao, C., Xie, C.W., Chen, D., Yu, F., Zhao, H., Yang, J., Zeng, J., Wang, J., Zhang, J., Zhou, J., Wang, J., Chen, J., Zhu, K., Zhao, K., Yan, K., Huang, L., Feng, M., Zhang, N., Li, P., Wu, P., Chu, R., Feng, R., Zhang, S., Sun, S., Fang, T., Wang, T., Gui, T., Weng, T., Shen, T., Lin, W., Wang, W., Wang, W., Zhou, W.,...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  39. [39]

    arXiv preprint arXiv:2505.22944 (2025) 4

    Wang, A., Huang, H., Fang, J.Z., Yang, Y., Ma, C.: Ati: Any trajectory instruction for controllable video generation. arXiv preprint arXiv:2505.22944 (2025)

    work page arXiv 2025

  40. [40]

    In: CVPR

    Ye, Z., Huang, H., Wang, X., Wan, P., Zhang, D., Luo, W.: Stylemaster: Stylize your video with artistic generation and translation. In: CVPR. pp. 2630–2640 (2025)

    work page 2025

  41. [41]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Yu, M., Hu, W., Xing, J., Shan, Y.: Trajectorycrafter: Redirecting camera trajectory for monocular videos via diffusion models. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 100–111 (2025)

    work page 2025

  42. [42]

    In: ICCV

    Zhang, L., Rao, A., Agrawala, M.: Adding conditional control to text-to-image diffusion models. In: ICCV. pp. 3836–3847 (2023) 18 K. Chhatre et al. Supplementary Material A Overview This supplementary material includes two parts: (i) detailed baseline repurposing procedures (Sec. B), including the shared 3D trajectory extraction pipeline, method-specific ...

    work page 2023