REVIEW 4 major objections 7 minor 2 cited by
Articulate That Object Part (ATOP): 3D Part Articulation via Text and Motion Personalization
T0 review · 4 major / 7 minor · reviewed 2026-08-08 · deepseek-v4-flash
Pith's one-line read ATOP claims that text prompts and part masks can recover hinge and slider motion for unseen 3D objects using only a few reference videos.
desk verdict Novel and useful pipeline for articulating static 3D assets from text and masks, but the claimed 2D-to-3D motion transfer is not actually implemented in the loss as written — a load-bearing gap that needs fixing before the results can be trusted. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The engine is a personalized multi-view motion diffusion model built by inflating ImageDream with pseudo-temporal convolutions and spatio-temporal attention, then attaching a controllable part-motion module that maps a binary part mask through small MLPs to scale and shift parameters and applies the affine modulation $Z' = \gamma \odot Z + \beta$ to latent features. Correspondence-aware spatial attention reshapes latents as $(B N_f) \times (N_v H W) \times F$ so geometric consistency is shared across views, while cross-attention injects the rendered rest-state images through a CLIP image encoder. The second stage renders per-frame 3D gaussians with a differentiable renderer and optimizes them under the score distillation loss, after which the axis and origin are selected from candidates derived from the part's oriented bounding box by chamfer distance to the distilled point cloud.
What would settle it
Sort PartNet-Mobility and ACD test objects by the angle between the ground-truth motion axis and the four fixed view directions; if mean angular error rises sharply for objects whose motion runs mostly along a viewing direction or whose part is occluded in those views, while the generated videos still look plausible, then the fixed camera set rather than the learned motion prior is carrying the axis estimate.
Extended reading notes
Core claim
The paper's central claim is that part articulation for a static 3D mesh can be recovered from a text prompt and a part mask without per-object 3D motion annotations: first finetune a pre-trained multi-view image diffusion model on a few reference videos to learn a motion identifier for the target part, then use that personalized model as a frozen critic to optimize per-frame 3D gaussians with a score distillation sampling loss, and finally estimate the revolute or prismatic axis by an oriented-bounding-box candidate search. The authors report that this two-stage procedure yields lower mean angular error and mean position error than Shape2Motion and OPD in the few-shot setting on PartNet-Mobility, and that it transfers zero-shot to ACD objects without new finetuning.
Load-bearing premise
The load-bearing premise is that four fixed camera viewpoints, with the moving part visible and its motion mostly transverse to the viewing direction, are enough for the score-distillation optimization to pin down the motion axis.
Editorial extensions
If this is right
- Static meshes from large unannotated 3D repositories can be articulated with only a text prompt, a part mask, and a few reference videos per category, which the paper demonstrates on top-retrieved Objaverse objects.
- Part-mask conditioning gives spatial control, so running inference once per mask lets the same object articulate different parts in separate passes.
- A single-stage multi-view motion generator avoids the structural distortions the paper observes when a two-stage image-to-video then video-to-multi-view pipeline is used.
- On PartNet-Mobility and ACD, the recovered axes beat the few-shot baselines on mean angular and position error, indicating that text-plus-mask inputs can substitute for direct 3D motion annotations at inference.
- The method is currently limited to revolute and prismatic joints and to one moving part per inference pass, with multi-part articulation handled by repeated inference.
Reading between the lines
- A testable consequence the authors do not develop: the fixed azimuth set of 45, 135, 225 and 315 degrees should bias the recovered axes toward directions visible in those silhouettes, so grouping benchmark results by the angle between the true axis and the viewing directions would reveal how much accuracy comes from the learned motion prior versus the camera arrangement.
- If the pipeline is combined with an upstream vision-language segmenter, it becomes a fully text-driven animator for asset libraries, where a prompt such as 'open the refrigerator door' both locates the part and supplies its motion.
- The OBB search converts continuous axis regression into discrete classification, which suggests that failures will cluster at symmetries of the bounding box; comparing the discrete search against direct chamfer optimization on clean synthetic inputs would isolate that effect.
- The few-shot budget is a natural scaling variable: measuring articulation accuracy as the number of reference videos grows from one to eight would show whether the method's generalization is chiefly driven by the pretrained diffusion prior or by the personalization data.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. ATOP proposes a two-stage method for articulating a static 3D mesh from a text prompt and a part mask. First, a pre-trained multi-view image diffusion model (ImageDream) is inflated into a video model and finetuned on a small number of reference multi-view videos to learn part-specific motion patterns; at test time it generates multi-view motion frames for an unseen mesh. Second, the motion is transferred to 3D by optimizing per-frame 3D Gaussians with an SDS loss and then fitting a revolute or prismatic axis with an OBB-constrained search. The paper reports large improvements over Shape2Motion and OPD on PartNet-Mobility and ACD, and over WAN2.1+SV4D and PuppetMaster+SV4D baselines on video quality metrics, plus qualitative results on Objaverse.
Significance. The proposed problem, adding part-level articulation to a static 3D mesh from a text prompt and a part mask, is timely and practically relevant. The single-stage few-shot personalization of a multi-view diffusion model is a sensible way to avoid error accumulation in two-stage I2V/V2MV pipelines, and the qualitative results on Objaverse suggest genuine generalization potential. The OBB-constrained axis search is a pragmatic and transparent way to regularize the otherwise noisy point-cloud optimization. If the pipeline truly transfers the generated 2D motion to 3D, this would be a useful contribution to animating large 3D asset repositories. However, the central claim is currently undercut by the gap between the generated video in Eq. (5) and the SDS loss in Eq. (7), by the overstatement of the method as “annotation-free”, and by the repeated deferral of key algorithmic details to an appendix that is not present in the manuscript.
major comments (4)
- [Sec. 4.4.2, Eq. (7); Sec. 4.3.1] The SDS loss in Eq. (7) conditions the denoiser only on (tau, I_r, B, C) and not on the generated multi-view video defined in Eq. (5). As written, the optimized 3D Gaussians are guided by the category-level motion prior learned during finetuning, not by the particular hallucinated motion in V_hat, so the claimed transfer of motion from the generated frames to the mesh is not implemented by the stated loss. Please add an explicit conditioning mechanism on V_hat (for example, a video embedding, a reference-frame attention term, or a reconstruction loss) or clearly state that V_hat is not used in this stage and discuss what the reported axis accuracy then measures.
- [Abstract; Introduction; Sec. 5.1; Sec. 7] The claim that ATOP is “annotation-free” is not supported by the described training protocol. The few-shot finetuning in Sec. 5.1 uses reference multi-view videos rendered from ground-truth PartNet-Mobility articulations, and rendering those videos requires exactly the motion annotations that the paper claims to avoid. Please replace “annotation-free” with a precise statement such as “no 3D motion supervision at test time” and specify what annotations are used to create the reference videos and how they are obtained.
- [Sec. 4.3.2; Sec. 5.1] The motion representation and the OBB-based axis estimation algorithm are repeatedly deferred to an appendix that is not included in the manuscript. Because the correctness of the axis estimation and the role of V_hat in Eq. (7) cannot be checked without this material, the appendix must be included (or the references to it removed) before the claims can be verified.
- [Tables 1 and 2] The baselines S2M and OPD are retrained with only 8 shapes per category, a regime that is known to disadvantage supervised methods, and no error bars or retraining-seed statistics are reported. The large MAE and MPE improvements should be accompanied by standard deviations over retraining seeds and, if possible, over multiple finetuning runs of ATOP, and the fully supervised performance of the baselines should be reported for context.
minor comments (7)
- [Throughout] The word “generalizibility” appears repeatedly (Introduction, Related Work, Sec. 4.2, Sec. 5.3); it should be corrected to “generalizability”.
- [References] Reference [39] is cited for CLIP but is attributed to Miller et al.; the canonical CLIP reference is Radford et al. 2021. Please correct the citation.
- [Eqs. (1) and (7)] The latent variable is written as z_t in Eq. (1) and as z_g^t in Eq. (7); unify the notation to avoid confusion about which render is being distilled.
- [Table 2] The NightStand MPE value of 0.54 for ATOP is a conspicuous outlier relative to the other categories; please verify the entry and discuss why this category is harder.
- [Table 3] The WAN2.1 baseline uses a distilled demo version optimized for fast inference, not the full 14B model; this may understate the baseline and should be acknowledged in the comparison.
- [Sec. 5.1] The four fixed viewpoints with elevation 10 or 30 degrees are a core design choice, but the paper provides no sensitivity analysis and no failure cases for parts that are occluded or whose motion is mostly along the viewing direction. Please add such an analysis or acknowledge this limitation.
- [Sec. 5.3] No statistical significance tests are provided for any of the quantitative tables; given the small number of training videos and categories, reporting confidence intervals would strengthen the claims.
Circularity Check
No significant circularity: the 3D axis is obtained by SDS optimization and an independent OBB-chamfer search, not by fitting a parameter that is then renamed as a prediction.
full rationale
ATOP's derivation chain is: (i) finetune the multi-view motion model on few-shot reference videos (Eq. 6); (ii) generate multi-view motion using the finetuned model (Eq. 5); (iii) optimize per-frame Gaussian states using the SDS loss (Eq. 7); (iv) select the motion axis and origin by an OBB-candidate search with chamfer distance against the optimized point cloud. I checked each stage for a reduction of the prediction to its inputs by construction. The SDS loss is not defined in terms of the target axis, and the axis selection is an independent geometric search over OBB-derived candidates, not a function of the text prompt or the generated video. The finetuning uses ground-truth articulation videos rendered from PartNet-Mobility, which weakens the 'annotation-free' wording, but it does not make the test-time axis estimate equal to the training labels by construction: evaluation is on held-out shapes and the axis is selected by optimization. A more serious concern is that Eq. 7 conditions on (tau, Ir, B, C) but not on the generated multi-view motion, so the generated frames may be causally inert for the axis stage; this is a correctness or architecture gap, not circularity, because no predicted quantity is identical to an input by definition. The self-citations in the paper are confined to related work and are not load-bearing. Therefore no circular step is present.
Assumptions & free parameters
free parameters (3)
- Elevation angle for rendered views =
10 or 30 degrees depending on category
- Classifier-free guidance scale =
5.0
- Number of reference videos per category =
8 shapes
assumptions (5)
- domain assumption Articulation is piecewise rigid and limited to revolute and prismatic joints.
- domain assumption Pre-trained ImageDream's correspondence-aware attention remains geometrically consistent when inflated for video and finetuned on a few examples.
- domain assumption Off-the-shelf part segmentation methods provide correct masks for unseen meshes.
- domain assumption Reference videos rendered from PartNet-Mobility ground truth are representative of the target category's motion distribution.
- domain assumption Score distillation sampling provides gradients that move the 3D gaussians toward physically plausible articulation states.
Cite this review
Pith. "Pith review of Articulate That Object Part (ATOP): 3D Part Articulation via Text and Motion Personalization." pith.science (2026). https://pith.science/paper/BISHWXHZ
@misc{pith2026250207278,
author = {Pith},
title = {Pith review of: Articulate That Object Part (ATOP): 3D Part Articulation via Text and Motion Personalization},
year = {2026},
howpublished = {\url{https://pith.science/paper/BISHWXHZ}},
note = {Machine review of arXiv:2502.07278}
}
read the original abstract
We present ATOP (Articulate That Object Part), a novel few-shot method based on motion personalization to articulate a static 3D object with respect to a part and its motion as prescribed in a text prompt. Given the scarcity of available datasets with motion attribute annotations, existing methods struggle to generalize well in this task. In our work, the text input allows us to tap into the power of modern-day diffusion models to generate plausible motion samples for the right object category and part. In turn, the input 3D object provides ``image prompting'' to personalize the generated motion to the very input object. Our method starts with a few-shot finetuning to inject articulation awareness to current diffusion models to learn a unique motion identifier associated with the target object part. Our finetuning is applied to a pre-trained diffusion model for controllable multi-view motion generation, trained with a small collection of reference motion frames demonstrating appropriate part motion. The resulting motion model can then be employed to realize plausible motion of the input 3D object from multiple views. At last, we transfer the personalized motion to the 3D space of the object via differentiable rendering to optimize part articulation parameters by a score distillation sampling loss. Experiments on PartNet-Mobility and ACD datasets demonstrate that our method can generate realistic motion samples with higher accuracy, leading to more generalizable 3D motion predictions compared to prior approaches in the few-shot setting.
Figures
Figures from the paper (16 more)
Forward citations
Cited by 2 Pith papers
-
Dual-Stream EEG Decoding for 3D Visual Perception
Dual-stream EEG decoder separates identity and orientation to support 3D reconstruction from neural signals via circular regression and conditioned diffusion.
-
Advances in 4D Representation: Geometry, Motion, and Interaction
A representation-centric survey of 4D generation and reconstruction, organized by geometry, motion, and interaction, with qualitative trade-off comparisons across seven representation families.
Reference graph
Works this paper leans on
-
[1]
Learning to generalize kinematic models to novel objects
Ben Abbatematteo, Stefanie Tellex, and George Konidaris. Learning to generalize kinematic models to novel objects. In Proceedings of the 3rd Conference on Robot Learning, 2019. 2, 3
2019
-
[2]
Smite: Segment me in time.arXiv preprint arXiv:2410.18538, 2024
Amirhossein Alimohammadi, Sauradip Nag, Saeid Asgari Taghanaki, Andrea Tagliasacchi, Ghassan Hamarneh, and Ali Mahdavi Amiri. Smite: Segment me in time.arXiv preprint arXiv:2410.18538, 2024. 4
arXiv 2024
-
[3]
Yogesh Balaji, Seungjun Nah, Xun Huang, Arash Vahdat, Jiaming Song, Karsten Kreis, Miika Aittala, Timo Aila, Samuli Laine, Bryan Catanzaro, et al. ediffi: Text-to-image diffusion models with an ensemble of expert denoisers.arXiv preprint arXiv:2211.01324, 2022. 4
arXiv 2022
-
[4]
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. 2
arXiv 2023
-
[5]
ShapeNet: An information-rich 3D model repos- itory.arXiv preprint arXiv:1512.03012, 2015
Angel X Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio Savarese, Mano- lis Savva, Shuran Song, Hao Su, Jianxiong Xiao, Li Yi, and Fisher Yu. ShapeNet: An information-rich 3D model repos- itory.arXiv preprint arXiv:1512.03012, 2015. 2
arXiv 2015
-
[6]
ABO: dataset and benchmarks for real-world 3D object understanding
Jasmine Collins, Shubham Goel, Kenan Deng, Achlesh- war Luthra, Leon Xu, Erhan Gundogdu, Xi Zhang, Tomas F Yago Vicente, Thomas Dideriksen, Himanshu Arora, Matthieu Guillaumin, and Jitendra Malik. ABO: dataset and benchmarks for real-world 3D object understanding. In CVPR, pages 21126–21136, 2022. 2, 9
2022
-
[7]
Flatten: optical flow- guided attention for consistent text-to-video editing
Yuren Cong, Mengmeng Xu, Christian Simon, Shoufa Chen, Jiawei Ren, Yanping Xie, Juan-Manuel Perez-Rua, Bodo Rosenhahn, Tao Xiang, and Sen He. Flatten: optical flow- guided attention for consistent text-to-video editing. 2024. 5
2024
-
[8]
Matt Deitke, Ruoshi Liu, Matthew Wallingford, Huong Ngo, Oscar Michel, Aditya Kusupati, Alan Fan, Christian Laforte, 14 Figure 14.Multi-part articulation:We demonstrate that by running inference multiple times with our personalized diffusion model, each time modifying the spatial location of the mask (indicated by different colors for each object), we can...
work page 2023
Show all 83 references
-
[9]
Objaverse: A universe of annotated 3d objects
Matt Deitke, Dustin Schwenk, Jordi Salvador, Luca Weihs, Oscar Michel, Eli VanderBilt, Ludwig Schmidt, Kiana Ehsani, Aniruddha Kembhavi, and Ali Farhadi. Objaverse: A universe of annotated 3d objects. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Reco...
2023
-
[10]
Objaverse-xl: A universe of 10m+ 3d objects.Advances in Neural Informa- tion Processing Systems, 36, 2024
Matt Deitke, Ruoshi Liu, Matthew Wallingford, Huong Ngo, Oscar Michel, Aditya Kusupati, Alan Fan, Christian Laforte, Vikram V oleti, Samir Yitzhak Gadre, et al. Objaverse-xl: A universe of 10m+ 3d objects.Advances in Neural Informa- tion Processing Systems, 36, 2024. 1, 9, 10, 18
2024
-
[11]
Qualitative results of generalization of Multi-view video generation and 3D Motion Axis prediction on ACD dataset [17]
Huan Fu, Rongfei Jia, Lin Gao, Mingming Gong, Binqiang 15 Figure 16. Qualitative results of generalization of Multi-view video generation and 3D Motion Axis prediction on ACD dataset [17]. Zhao, Steve Maybank, and Dacheng Tao. 3d-future: 3d fur- niture shape with texture.Inter...
2021
-
[12]
Qualitative results of generalization of Multi-view video generation and 3D Motion Axis prediction on ACD dataset [17]
Songwei Ge, Seungjun Nah, Guilin Liu, Tyler Poon, An- 16 Figure 17. Qualitative results of generalization of Multi-view video generation and 3D Motion Axis prediction on ACD dataset [17]. 17 Figure 18. Qualitative results of generalization of Multi-view video generation and 3D...
2023 arXiv
-
[13]
Animatediff: Animate your personalized text- to-image diffusion models without specific tuning.arXiv preprint arXiv:2307.04725, 2023
Yuwei Guo, Ceyuan Yang, Anyi Rao, Zhengyang Liang, Yaohui Wang, Yu Qiao, Maneesh Agrawala, Dahua Lin, and Bo Dai. Animatediff: Animate your personalized text- to-image diffusion models without specific tuning.arXiv preprint arXiv:2307.04725, 2023. 2, 4
2023 arXiv
-
[14]
Denoising dif- fusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020
Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising dif- fusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020. 4
2020
-
[15]
Cogvideo: Large-scale pretraining for text-to-video generation via transformers.arXiv preprint arXiv:2205.15868, 2022
Wenyi Hong, Ming Ding, Wendi Zheng, Xinghan Liu, and Jie Tang. Cogvideo: Large-scale pretraining for text-to-video generation via transformers.arXiv preprint arXiv:2205.15868, 2022. 4
2022 arXiv
-
[16]
Learning to predict part mo- bility from a single static snapshot.ACM Transactions On Graphics (TOG), 36(6):1–13, 2017
Ruizhen Hu, Wenchao Li, Oliver Van Kaick, Ariel Shamir, Hao Zhang, and Hui Huang. Learning to predict part mo- bility from a single static snapshot.ACM Transactions On Graphics (TOG), 36(6):1–13, 2017. 2, 3
2017
-
[17]
S2o: Static to openable enhancement for articulated 3d objects.arXiv preprint arXiv:2409.18896,
Denys Iliash, Hanxiao Jiang, Yiming Zhang, Manolis Savva, and Angel X Chang. S2o: Static to openable enhancement for articulated 3d objects.arXiv preprint arXiv:2409.18896,
-
[18]
Screwnet: Category-independent articulation model estimation from depth images using screw theory
Ajinkya Jain, Rudolf Lioutikov, Caleb Chuck, and Scott Niekum. Screwnet: Category-independent articulation model estimation from depth images using screw theory. In 2021 IEEE International Conference on Robotics and Au- tomation (ICRA), pages 13670–13677. IEEE, 2021. 2, 3
2021
-
[19]
Vmc: Video motion customization using temporal attention adaption for text-to-video diffusion models.arXiv preprint arXiv:2312.00845, 2023
Hyeonho Jeong, Geon Yeong Park, and Jong Chul Ye. Vmc: Video motion customization using temporal attention adaption for text-to-video diffusion models.arXiv preprint arXiv:2312.00845, 2023. 4
2023 arXiv
-
[20]
Opd: Single-view 3d openable part detection
Hanxiao Jiang, Yongsen Mao, Manolis Savva, and Angel X Chang. Opd: Single-view 3d openable part detection. In European Conference on Computer Vision, pages 410–426. Springer, 2022. 2, 3, 9, 13
2022
-
[21]
Ditto: Building digital twins of articulated objects from interaction
Zhenyu Jiang, Cheng-Chun Hsu, and Yuke Zhu. Ditto: Building digital twins of articulated objects from interaction. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5616–5626, 2022. 3
2022
-
[22]
Approximate differ- entiable rendering with algebraic surfaces
Leonid Keselman and Martial Hebert. Approximate differ- entiable rendering with algebraic surfaces. InEuropean Con- ference on Computer Vision, pages 596–614. Springer, 2022. 8
2022
-
[23]
Flexible techniques for differentiable rendering with 3d gaussians.arXiv preprint arXiv:2308.14737, 2023
Leonid Keselman and Martial Hebert. Flexible techniques for differentiable rendering with 3d gaussians.arXiv preprint arXiv:2308.14737, 2023. 8
2023 arXiv
-
[24]
Habitat synthetic scenes dataset (hssd-200): An analysis of 3d scene scale and realism tradeoffs for objectgoal naviga- tion
Mukul Khanna, Yongsen Mao, Hanxiao Jiang, Sanjay Haresh, Brennan Shacklett, Dhruv Batra, Alexander Clegg, Eric Undersander, Angel X Chang, and Manolis Savva. Habitat synthetic scenes dataset (hssd-200): An analysis of 3d scene scale and realism tradeoffs for objectgoal naviga-...
-
[25]
Partstad: 2d-to-3d part segmentation task adaptation
Hyunjin Kim and Minhyuk Sung. Partstad: 2d-to-3d part segmentation task adaptation. InEuropean Conference on Computer Vision, pages 422–439. Springer, 2024. 4, 5, 9, 14, 15
2024
-
[26]
Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980,
Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980,
-
[27]
Auto-encoding varia- tional bayes.arXiv preprint arXiv:1312.6114, 2013
Diederik P Kingma and Max Welling. Auto-encoding varia- tional bayes.arXiv preprint arXiv:1312.6114, 2013. 6
2013 arXiv
-
[28]
Ccvs: context-aware controllable video synthesis.Advances in Neural Information Processing Systems, 34:14042–14055,
Guillaume Le Moing, Jean Ponce, and Cordelia Schmid. Ccvs: context-aware controllable video synthesis.Advances in Neural Information Processing Systems, 34:14042–14055,
-
[29]
Nap: Neural 3d articulated object prior.Advances in Neural Information Processing Systems, 36:31878–31894, 2023
Jiahui Lei, Congyue Deng, William B Shen, Leonidas J Guibas, and Kostas Daniilidis. Nap: Neural 3d articulated object prior.Advances in Neural Information Processing Systems, 36:31878–31894, 2023. 4, 14
2023
-
[30]
Vivid-zoo: Multi-view video generation with diffusion model.Advances in Neural Information Processing Systems, 37:62189–62222,
Bing Li, Cheng Zheng, Wenxuan Zhu, Jinjie Mai, Biao Zhang, Peter Wonka, and Bernard Ghanem. Vivid-zoo: Multi-view video generation with diffusion model.Advances in Neural Information Processing Systems, 37:62189–62222,
-
[31]
Puppet-master: Scaling interactive video gen- eration as a motion prior for part-level dynamics
Ruining Li, Chuanxia Zheng, Christian Rupprecht, and An- drea Vedaldi. Puppet-master: Scaling interactive video gen- eration as a motion prior for part-level dynamics. InICCV,
-
[32]
Paris: Part-level reconstruction and motion analysis for articulated objects
Jiayi Liu, Ali Mahdavi-Amiri, and Manolis Savva. Paris: Part-level reconstruction and motion analysis for articulated objects. InProceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 352–363, 2023. 3, 8
2023
-
[33]
Singapo: Single image controlled generation of articulated parts in object.arXiv preprint arXiv:2410.16499, 2024
Jiayi Liu, Denys Iliash, Angel X Chang, Manolis Savva, and Ali Mahdavi-Amiri. Singapo: Single image controlled generation of articulated parts in object.arXiv preprint arXiv:2410.16499, 2024. 4, 9
2024 arXiv
-
[34]
Cage: Controllable articulation generation
Jiayi Liu, Hou In Ivan Tam, Ali Mahdavi-Amiri, and Manolis Savva. Cage: Controllable articulation generation. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17880–17889, 2024. 4, 14
2024
-
[35]
Partslip: Low-shot part seg- mentation for 3d point clouds via pretrained image-language models
Minghua Liu, Yinhao Zhu, Hong Cai, Shizhong Han, Zhan Ling, Fatih Porikli, and Hao Su. Partslip: Low-shot part seg- mentation for 3d point clouds via pretrained image-language models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21...
2023
-
[36]
Videofusion: Decomposed diffusion mod- els for high-quality video generation
Zhengxiong Luo, Dayou Chen, Yingya Zhang, Yan Huang, Liang Wang, Yujun Shen, Deli Zhao, Jingren Zhou, and Tieniu Tan. Videofusion: Decomposed diffusion mod- els for high-quality video generation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogniti...
2023
-
[37]
Real2code: Reconstruct articulated objects via code genera- tion.arXiv preprint arXiv:2406.08474, 2024
Zhao Mandi, Yijia Weng, Dominik Bauer, and Shuran Song. Real2code: Reconstruct articulated objects via code genera- tion.arXiv preprint arXiv:2406.08474, 2024. 3
2024 arXiv
-
[38]
Customizing motion in text-to-video diffusion models.arXiv preprint arXiv:2312.04966, 2023
Joanna Materzynska, Josef Sivic, Eli Shechtman, Antonio Torralba, Richard Zhang, and Bryan Russell. Customizing motion in text-to-video diffusion models.arXiv preprint arXiv:2312.04966, 2023. 4
2023 arXiv
-
[39]
Miller, Will Feng, Dhruva Tirumala, Adam Fisch, Augustus Odena, Vivek Ramavajjala, Joel Z
Alexander H. Miller, Will Feng, Dhruva Tirumala, Adam Fisch, Augustus Odena, Vivek Ramavajjala, Joel Z. Leibo, Kelvin Guu andJesse Engel, Jack Clark, Maruan H. Ali, Nazneen Rajani, Iain J. Dunning, Jacob Andreas, Chris 20 Dyer, Dario Amodei, Jakob Uszkoreit, Douwe Pieksma, Tom...
2020
-
[40]
Chang, Li Yi, Subarna Tripathi, Leonidas J
Kaichun Mo, Shilin Zhu, Angel X. Chang, Li Yi, Subarna Tripathi, Leonidas J. Guibas, and Hao Su. PartNet: A large- scale benchmark for fine-grained and hierarchical part-level 3D object understanding. InThe IEEE Conference on Com- puter Vision and Pattern Recognition (CVPR), 2019. 2
2019
-
[41]
Where2act: From pixels to actions for articulated 3d objects
Kaichun Mo, Leonidas J Guibas, Mustafa Mukadam, Abhi- nav Gupta, and Shubham Tulsiani. Where2act: From pixels to actions for articulated 3d objects. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 6813–6823, 2021. 2, 3
2021
-
[42]
Dreamix: Video diffusion models are general video editors.arXiv preprint arXiv:2302.01329, 2023
Eyal Molad, Eliahu Horwitz, Dani Valevski, Alex Rav Acha, Yossi Matias, Yael Pritch, Yaniv Leviathan, and Yedid Hoshen. Dreamix: Video diffusion models are general video editors.arXiv preprint arXiv:2302.01329, 2023. 4
2023 arXiv
-
[43]
Difftad: Temporal action detection with pro- posal denoising diffusion
Sauradip Nag, Xiatian Zhu, Jiankang Deng, Yi-Zhe Song, and Tao Xiang. Difftad: Temporal action detection with pro- posal denoising diffusion. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 10362– 10374, 2023. 4
2023
-
[44]
In-2-4d: Inbetweening from two single-view images to 4d generation.arXiv preprint arXiv:2504.08366, 2025
Sauradip Nag, Daniel Cohen-Or, Hao Zhang, and Ali Mahdavi-Amiri. In-2-4d: Inbetweening from two single-view images to 4d generation.arXiv preprint arXiv:2504.08366, 2025. 4
2025
-
[45]
Pytorch: An im- perative style, high-performance deep learning library.Ad- vances in neural information processing systems, 32, 2019
Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An im- perative style, high-performance deep learning library.Ad- vances in neural information processing systems, ...
2019
-
[46]
Sdxl: Improving latent diffusion mod- els for high-resolution image synthesis.arXiv preprint arXiv:2307.01952, 2023
Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas M ¨uller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion mod- els for high-resolution image synthesis.arXiv preprint arXiv:2307.01952, 2023. 4
2023 arXiv
-
[47]
Barron, and Ben Milden- hall
Ben Poole, Ajay Jain, Jonathan T. Barron, and Ben Milden- hall. Dreamfusion: Text-to-3d using 2d diffusion.arXiv,
-
[48]
Fatezero: Fus- ing attentions for zero-shot text-based video editing
Chenyang Qi, Xiaodong Cun, Yong Zhang, Chenyang Lei, Xintao Wang, Ying Shan, and Qifeng Chen. Fatezero: Fus- ing attentions for zero-shot text-based video editing. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 15932–15942, 2023. 5
2023
-
[49]
Articulate anymesh: Open-vocabulary 3d articulated objects modeling.arXiv preprint arXiv:2502.02590, 2025
Xiaowen Qiu, Jincheng Yang, Yian Wang, Zhehuan Chen, Yufei Wang, Tsun-Hsuan Wang, Zhou Xian, and Chuang Gan. Articulate anymesh: Open-vocabulary 3d articulated objects modeling.arXiv preprint arXiv:2502.02590, 2025. 4, 8
2025 arXiv
-
[50]
Zero-shot text-to-image generation
Aditya Ramesh, Mikhail Pavlov, Gabriel Goh, Scott Gray, Chelsea V oss, Alec Radford, Mark Chen, and Ilya Sutskever. Zero-shot text-to-image generation. InInternational Confer- ence on Machine Learning, pages 8821–8831. PMLR, 2021. 4
2021
-
[51]
Sam 2: Segment anything in images and videos,
Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman R¨adle, Chloe Rolland, Laura Gustafson, Eric Mintun, Junt- ing Pan, Kalyan Vasudev Alwala, Nicolas Carion, Chao- Yuan Wu, Ross Girshick, Piotr Doll´ar, and Christoph Feic...
-
[52]
L4gm: Large 4d gaussian reconstruction model.Advances in Neural Information Processing Systems, 37:56828–56858, 2024
Jiawei Ren, Cheng Xie, Ashkan Mirzaei, Karsten Kreis, Zi- wei Liu, Antonio Torralba, Sanja Fidler, Seung Wook Kim, Huan Ling, et al. L4gm: Large 4d gaussian reconstruction model.Advances in Neural Information Processing Systems, 37:56828–56858, 2024. 4
2024
-
[53]
High-resolution image synthesis with latent diffusion models
Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 5
2022
-
[54]
U- net: Convolutional networks for biomedical image segmen- tation
Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U- net: Convolutional networks for biomedical image segmen- tation. InMedical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, par...
2015
-
[55]
Photorealistic text-to-image diffusion models with deep language understanding.Advances in Neural Information Processing Systems, 35:36479–36494, 2022
Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. Photorealistic text-to-image diffusion models with deep language understanding.Advances in Neural Information ...
2022
-
[56]
Make-a-video: Text-to-video generation without text-video data.arXiv preprint arXiv:2209.14792,
Uriel Singer, Adam Polyak, Thomas Hayes, Xi Yin, Jie An, Songyang Zhang, Qiyuan Hu, Harry Yang, Oron Ashual, Oran Gafni, et al. Make-a-video: Text-to-video generation without text-video data.arXiv preprint arXiv:2209.14792,
-
[57]
Reacto: Reconstructing articulated ob- jects from a single video
Chaoyue Song, Jiacheng Wei, Chuan Sheng Foo, Guosheng Lin, and Fayao Liu. Reacto: Reconstructing articulated ob- jects from a single video. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5384–5395, 2024. 3
2024
-
[58]
Denoising diffusion implicit models.arXiv preprint arXiv:2010.02502, 2020
Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models.arXiv preprint arXiv:2010.02502, 2020. 4
2010 arXiv
-
[59]
Score-based generative modeling through stochastic differential equa- tions.arXiv preprint arXiv:2011.13456, 2020
Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equa- tions.arXiv preprint arXiv:2011.13456, 2020. 4
2011 arXiv
-
[60]
Fvd: A new metric for video generation
Thomas Unterthiner, Sjoerd van Steenkiste, Karol Kurach, Rapha¨el Marinier, Marcin Michalski, and Sylvain Gelly. Fvd: A new metric for video generation. 2019. 12
2019
-
[61]
Mo- tiondreamer: Zero-shot 3d mesh animation from video dif- fusion models.arXiv preprint arXiv:2405.20155, 2024
Lukas Uzolas, Elmar Eisemann, and Petr Kellnhofer. Mo- tiondreamer: Zero-shot 3d mesh animation from video dif- fusion models.arXiv preprint arXiv:2405.20155, 2024. 4
2024 arXiv
-
[62]
Attention is all you need.Advances in neural information processing systems, 30, 2017
Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in neural information processing systems, 30, 2017. 7
2017
-
[63]
Sv3d: Novel multi-view syn- thesis and 3d generation from a single image using latent 21 video diffusion
Vikram V oleti, Chun-Han Yao, Mark Boss, Adam Letts, David Pankratz, Dmitry Tochilkin, Christian Laforte, Robin Rombach, and Varun Jampani. Sv3d: Novel multi-view syn- thesis and 3d generation from a single image using latent 21 video diffusion. InEuropean Conference on Comput...
2025
-
[64]
Diffusers: State-of-the-art diffu- sion models.https://github.com/huggingface/ diffusers, 2022
Patrick von Platen, Suraj Patil, Anton Lozhkov, Pedro Cuenca, Nathan Lambert, Kashif Rasul, Mishig Davaadorj, Dhruv Nair, Sayak Paul, William Berman, Yiyi Xu, Steven Liu, and Thomas Wolf. Diffusers: State-of-the-art diffu- sion models.https://github.com/huggingface/ diffusers, 2022. 8
2022
-
[65]
Wan: Open and advanced large-scale video gen- erative models.arXiv preprint arXiv:2503.20314, 2025
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. 2, 3, 4, 5, 10, 11
2025 arXiv
-
[66]
Imagedream: Image-prompt multi-view diffusion for 3d generation.arXiv preprint arXiv:2312.02201, 2023
Peng Wang and Yichun Shi. Imagedream: Image-prompt multi-view diffusion for 3d generation.arXiv preprint arXiv:2312.02201, 2023. 3, 4, 5, 9
2023 arXiv
-
[67]
Shape2motion: Joint analysis of motion parts and attributes from 3d shapes
Xiaogang Wang, Bin Zhou, Yahao Shi, Xiaowu Chen, Qin- ping Zhao, and Kai Xu. Shape2motion: Joint analysis of motion parts and attributes from 3d shapes. InProceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 8876–8884, 2019. 2, 3, 9, 13
2019
-
[68]
Videocomposer: Compositional video synthesis with motion controllability
Xiang Wang, Hangjie Yuan, Shiwei Zhang, Dayou Chen, Ji- uniu Wang, Yingya Zhang, Yujun Shen, Deli Zhao, and Jin- gren Zhou. Videocomposer: Compositional video synthesis with motion controllability. InNeurIPS, 2024. 2
2024
-
[69]
Self-supervised neural articulated shape and appearance models
Fangyin Wei, Rohan Chabra, Lingni Ma, Christoph Lassner, Michael Zollhoefer, Szymon Rusinkiewicz, Chris Sweeney, Richard Newcombe, and Mira Slavcheva. Self-supervised neural articulated shape and appearance models. InProceed- ings IEEE/CVF Conference on Computer Vision and Pat...
2022
-
[70]
Tune-a-video: One-shot tuning of image diffusion models for text-to-video generation.arXiv preprint arXiv:2212.11565, 2022
Jay Zhangjie Wu, Yixiao Ge, Xintao Wang, Stan Weixian Lei, Yuchao Gu, Wynne Hsu, Ying Shan, Xiaohu Qie, and Mike Zheng Shou. Tune-a-video: One-shot tuning of image diffusion models for text-to-video generation.arXiv preprint arXiv:2212.11565, 2022. 4, 5
2022 arXiv
-
[71]
Lamp: Learn a motion pat- tern for few-shot-based video generation.arXiv preprint arXiv:2310.10769, 2023
Ruiqi Wu, Liangyu Chen, Tong Yang, Chunle Guo, Chongyi Li, and Xiangyu Zhang. Lamp: Learn a motion pat- tern for few-shot-based video generation.arXiv preprint arXiv:2310.10769, 2023. 4
2023 arXiv
-
[72]
Sapien: A simulated part-based interactive environment
Fanbo Xiang, Yuzhe Qin, Kaichun Mo, Yikuan Xia, Hao Zhu, Fangchen Liu, Minghua Liu, Hanxiao Jiang, Yifu Yuan, He Wang, et al. Sapien: A simulated part-based interactive environment. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 1109...
2020
-
[73]
Sv4d: Dynamic 3d content generation with multi-frame and multi-view consistency.arXiv preprint arXiv:2407.17470, 2024
Yiming Xie, Chun-Han Yao, Vikram V oleti, Huaizu Jiang, and Varun Jampani. Sv4d: Dynamic 3d content generation with multi-frame and multi-view consistency.arXiv preprint arXiv:2407.17470, 2024. 2, 3, 5, 12
2024 arXiv
-
[74]
Dynamicrafter: Animating open-domain images with video diffusion priors
Jinbo Xing, Menghan Xia, Yong Zhang, Haoxin Chen, Wangbo Yu, Hanyuan Liu, Gongye Liu, Xintao Wang, Ying Shan, and Tien-Tsin Wong. Dynamicrafter: Animating open-domain images with video diffusion priors. InEu- ropean Conference on Computer Vision, pages 399–417. Springer, 2024. 2, 4
2024
-
[75]
Rpm-net: recurrent prediction of motion and parts from point cloud
Zihao Yan, Ruizhen Hu, Xingguang Yan, Luanmin Chen, Oliver Van Kaick, Hao Zhang, and Hui Huang. Rpm-net: recurrent prediction of motion and parts from point cloud. arXiv preprint arXiv:2006.14865, 2020. 2, 3
2006 arXiv
-
[76]
Cogvideox: Text-to-video diffusion models with an expert transformer.arXiv preprint arXiv:2408.06072, 2024
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. 2, 4
2024 arXiv
-
[77]
Sv4d 2.0: Enhancing spatio-temporal consistency in multi-view video diffusion for high-quality 4d generation.arXiv preprint arXiv:2503.16396, 2025
Chun-Han Yao, Yiming Xie, Vikram V oleti, Huaizu Jiang, and Varun Jampani. Sv4d 2.0: Enhancing spatio-temporal consistency in multi-view video diffusion for high-quality 4d generation.arXiv preprint arXiv:2503.16396, 2025. 2, 3, 4, 5, 12
2025 arXiv
-
[78]
Magvit: Masked generative video transformer
Lijun Yu, Yong Cheng, Kihyuk Sohn, Jos ´e Lezama, Han Zhang, Huiwen Chang, Alexander G Hauptmann, Ming- Hsuan Yang, Yuan Hao, Irfan Essa, et al. Magvit: Masked generative video transformer. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pa...
2023
-
[79]
Stag4d: Spatial-temporal anchored generative 4d gaussians
Yifei Zeng, Yanqin Jiang, Siyu Zhu, Yuanxun Lu, Youtian Lin, Hao Zhu, Weiming Hu, Xun Cao, and Yao Yao. Stag4d: Spatial-temporal anchored generative 4d gaussians. InEu- ropean Conference on Computer Vision, pages 163–179. Springer, 2024. 4
2024
-
[80]
4diffusion: Multi-view video dif- fusion model for 4d generation.Advances in Neural Infor- mation Processing Systems, 37:15272–15295, 2024
Haiyu Zhang, Xinyuan Chen, Yaohui Wang, Xihui Liu, Yun- hong Wang, and Yu Qiao. 4diffusion: Multi-view video dif- fusion model for 4d generation.Advances in Neural Infor- mation Processing Systems, 37:15272–15295, 2024. 4, 5
2024
-
[81]
The unreasonable effectiveness of deep features as a perceptual metric
Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. The unreasonable effectiveness of deep features as a perceptual metric. InCVPR, 2018. 12
2018
-
[82]
Motiondirector: Motion customization of text-to-video diffusion models
Rui Zhao, Yuchao Gu, Jay Zhangjie Wu, David Jun- hao Zhang, Jia-Wei Liu, Weijia Wu, Jussi Keppo, and Mike Zheng Shou. Motiondirector: Motion customization of text-to-video diffusion models. InEuropean Conference on Computer Vision, pages 273–290. Springer, 2024. 4, 5
2024
-
[83]
Magicvideo: Efficient video generation with latent diffusion models.arXiv preprint arXiv:2211.11018, 2022
Daquan Zhou, Weimin Wang, Hanshu Yan, Weiwei Lv, Yizhe Zhu, and Jiashi Feng. Magicvideo: Efficient video generation with latent diffusion models.arXiv preprint arXiv:2211.11018, 2022. 4 22
2022 arXiv
Reviewed August 8, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.