Pith. sign in

REVIEW 4 major objections 6 minor 69 references

4D-Animal: Freely Reconstructing Animatable 3D Animals from Videos

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read 4D-Animal reconstructs animatable 3D animals from monocular videos without sparse keypoint annotations, using dense 2D cues and hierarchical alignment to fit SMAL models, and reports state-of-the-art shape and texture quality on the COP3D…

desk verdict Useful keypoint-free 4D animal reconstruction, but the headline margin over Avatars is unverified and the paper overclaims on model-free baselines. read the letter →

arxiv 2507.10437 v1 pith:TUCH26AX submitted 2025-07-14 cs.CV

Shanshan Zhong , Jiawei Peng , Zehan Zheng , Zhongzhan Huang , Wufei Ma , Guofeng Zhang , Qihao Liu , Alan Yuille
show 1 more author
Jieneng Chen
This is my paper
classification cs.CV
keywords 3DanimalreconstructionanimatableavatarSMALfittingkeypoint-freehierarchicalalignmentdensefeaturesmonocularvideoquadruped
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that sparse semantic keypoints, long considered necessary for fitting parametric animal models like SMAL to videos, are not required for high-quality animatable 3D reconstruction. Instead, weak but dense cues from pretrained 2D vision models—silhouettes, part masks, dense pixel-to-vertex correspondences, and point tracks—can provide enough guidance. If this is right, labor-intensive keypoint annotation and unreliable keypoint detectors can be dropped from animal reconstruction pipelines, making large-scale in-the-wild animal modeling practical. The paper supports the claim by showing that its keypoint-free method outperforms both model-based and model-free baselines on IoU, perceptual texture quality, and depth accuracy.

What carries the argument

The central machinery is the hierarchical alignment loss combined with a dense feature network. The SMAL parametric model (a skinned quadruped template with shape and pose parameters) is fitted per frame by a learnable network that takes DINO-ViT features and a frame-index embedding and outputs SMAL parameters. A multi-stage training schedule first stabilizes global pose with weak object and temporal weights, then enables vertex offsets and fine alignment while decreasing part and pixel weights and increasing temporal weight. The four geometric losses—silhouette Chamfer distance, part-based point-to-vertex distance, CSE-derived pixel-to-vertex distance, and BootsTAP trajectory re-projection distance—each target a different granularity of alignment, and their calibrated combination is what substitutes for keypoint supervision.

What would settle it

Run 4D-Animal on a set of dog videos where the PartGLEE masks or CSE correspondences are known to be incorrect, for example fast-motion sequences with motion blur or occluded body parts, and measure IoU and LPIPS; if a keypoint-supervised baseline such as BITE or Avatars outperforms 4D-Animal on these subsets, the claim that keypoint-free fitting is universally superior would be falsified. A simpler test is to corrupt the CSE-to-SMAL mapping (e.g., by random vertex shuffles) and observe whether the reported COP3D gains vanish.

Watch

Extended reading notes

Core claim

The central claim is that a keypoint-free, optimization-based fitting framework can achieve state-of-the-art animatable 3D animal reconstruction from monocular videos. The method, 4D-Animal, maps DINO-ViT image features to SMAL parameters through a lightweight dense feature network, then aligns the resulting mesh to the video using a hierarchical set of geometric losses: object-level silhouette matching with SAM masks, part-level alignment with semantic masks (head, body, feet, tail) from PartGLEE, pixel-to-vertex supervision from CSE correspondences transferred to the SMAL mesh via Zoom-Out, and temporal consistency from BootsTAP point tracks. On 50 dog videos from COP3D, 4D-Animal reports IoU 0.84, IoUw5 0.71, PSNR 21.28, and LPIPS 0.061, outperforming keypoint-dependent baselines BARC, BITE, and Avatars, and the model-free RAC; it also achieves lower depth error on TracksTo4D and roughly 3x faster convergence than Avatars. The paper further shows that removing the object-level loss drastically degrades performance (IoU drops to 0.51), while removing part, pixel, or temporal losses each causes measurable degradation, supporting the claim that the combination of hierarchical cues is what makes keypoint-free fitting work.

Load-bearing premise

The method assumes that the pretrained 2D models—especially the CSE-to-SMAL zoom-out mapping and PartGLEE part masks—produce accurate enough cues on the target dog videos; if these cues are unreliable, the hierarchical alignment loses its guidance and reconstruction quality drops.

Editorial extensions

If this is right

  • If keypoint annotations are truly unnecessary, animal reconstruction pipelines can scale to arbitrary casual videos without manual labeling.
  • The reported speedup in convergence suggests that dense feature initialization can make optimization-based fitting practical for larger video collections.
  • The generated animatable 3D assets can be used to fine-tune large image-to-3D generative models, reducing the synthetic-to-real domain gap for unseen animal images.
  • The method's generality beyond dogs is suggested by preliminary cat reconstructions using a dog-specific template, indicating potential extension to other quadrupeds.
  • Better depth accuracy than Avatars on RGB-D dog videos implies that keypoint-free fitting can capture geometry at least as reliably as keypoint-supervised approaches.

Reading between the lines

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

  • The approach implicitly shifts the annotation burden from sparse keypoints to the choice and quality of pretrained 2D models; as those models improve, keypoint-free reconstruction should improve without changing the pipeline.
  • A testable extension would be to replace the SMAL template with a species-specific or learned template and measure whether the hierarchical cues alone can drive fitting for non-canine quadrupeds, a regime the paper only samples with cats.
  • The failure cases point to a concrete boundary: when part masks, correspondences, or tracks are unreliable—fast motion, occlusion, atypical ears—the method loses its competitive edge, suggesting that robustness gains will come from making the 2D cues themselves more reliable rather than from the fitting objective.
  • The claimed benefit to image-to-3D generation could be quantified more directly by measuring downstream model performance before and after fine-tuning on 4D-Animal assets across multiple generator architectures and unseen animal categories.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper proposes 4D-Animal, an optimization-based method for reconstructing animatable 3D dog models from monocular videos without sparse keypoint annotations. It regresses SMAL parameters from DINO-ViT features through a shallow feature network and supervises the fit with a hierarchical set of cues: SAM silhouettes, PartGLEE part masks, CSE pixel-to-vertex correspondences mapped to SMAL via Zoom-Out, BootsTAP trajectories, and a perceptual LPIPS texture loss. The method is evaluated on 50 COP3D dog videos with a contiguous train/test frame split, reporting IoU 0.84, IoUw5 0.71, PSNR 21.28, and LPIPS 0.061, with comparisons to model-based (BARC, BITE, Avatars) and model-free (RAC) baselines; depth metrics on TracksTo4D and qualitative results are also included. The central claim is that keypoint-free fitting can outperform prior keypoint-based and model-free approaches.

Significance. If the reported results are validated, this is a useful contribution: it removes the need for manually annotated sparse keypoints, integrates dense but weak 2D cues in a principled hierarchical manner, and releases code. The held-out-frame evaluation split and the anchoring of losses to external pretrained models make the protocol substantially less prone to circularity than many fitting papers. The method also demonstrates potential downstream value for image-to-3D generation. However, the evaluation as presented has load-bearing gaps: the Avatars baseline is a re-run whose published numbers are withheld, no uncertainty estimates are provided, and the paper's own table contradicts the abstract claim of outperforming model-free baselines on PSNR.

major comments (4)
  1. [§4.2, Table 1, and Abstract] The abstract and Section 4.2 claim that 4D-Animal 'outperforms both model-based and model-free baselines' and achieves 'the best overall reconstruction.' Table 1, however, shows RAC (model-free) with PSNR 21.86 vs. 4D-Animal's 21.28, so the unqualified 'best overall' claim is internally contradicted by the paper's own reported metric. Please either state the aggregation criterion explicitly (e.g., wins on 4 of 5 metrics, with IoU/LPIPS as priorities), or temper the claim. The PSNR trade-off should be discussed, since it may reflect differences in texture-model capacity rather than geometry quality.
  2. [Appendix B.4 and Table 1] The Avatars* results in Table 1 come from a re-run that, by the authors' own admission, is 'slightly lower than those reported in the original paper.' The original published Avatars numbers are never reported, while the BARC, BITE, and RAC rows are copied from the Avatars paper. This mixes a possibly degraded re-run with published numbers for the other baselines, so the headline margins (IoU 0.84 vs. 0.81, PSNR 21.28 vs. 19.91) cannot be verified as reflecting a genuine improvement over Avatars. Please provide a table with both the original published values and the re-run values, and justify why the re-run is the appropriate basis for comparison rather than the original numbers.
  3. [§4.2 and §5 (RQ1, RQ2)] All quantitative results are point estimates over 50 videos with no standard deviations, confidence intervals, or significance tests. The margins over Avatars are small (e.g., IoU 0.84 vs. 0.81, PSNR 21.28 vs. 19.91), and even the shortfall to RAC on PSNR (0.58) could be within run-to-run or video-level variability. Please report per-video statistics with bootstrap confidence intervals or paired significance tests for the main metrics, and state the number of random seeds used.
  4. [§3.2 and Table 3] The dense feature network is a stated core contribution (abstract and Section 1), but its contribution is never ablated. Table 3 ablates only the hierarchical losses; there is no comparison against, for example, directly optimizing SMAL parameters per frame with the same losses and no feature network. The efficiency claim in Figure 6/12 (up to 3x speedup) is measured against Avatars rather than against an internal ablation, so the reader cannot attribute the improvements to the feature network specifically. Please add an ablation that removes or replaces the feature network to isolate its effect on accuracy and convergence speed.
minor comments (6)
  1. [§5 (Analysis)] There are two research questions labeled 'RQ4' in Section 5: one on downstream 3D generation and one on failure cases. The second should be renumbered to RQ5.
  2. [§3.3, §3.4, and Appendix cross-references] Several appendix cross-references do not match the actual appendix labels: 'Appendix 1.3' should be 'Appendix A.3,' 'Appendix 1.1' should be 'Appendix A.1,' and 'Appendix 2' / 'Appendix 2.1' should be 'Appendix B' / 'Appendix B.1.' Please correct them.
  3. [Table 4] The table's column structure is difficult to parse: under the three condition labels 'Initialization Org EPnP EPnP-RANSAC' there are seven numeric columns with checkmark rows that are not clearly aligned with the described configurations. Please restructure the table so each configuration has a single column, and state clearly what 'Org' means.
  4. [Figure 6 and Figure 12] The axis labels 'RNSP' and 'SPIPL' appear to be typos for 'PSNR' and 'LPIPS.' Please correct them in both figures.
  5. [Appendix B.7] The sentence 'As we have mentioned in Sec. 6 (main text)' is inaccurate because the downstream-task RQ4 appears in Section 5, not Section 6. Please correct the cross-reference.
  6. [Table 5] It is not stated whether the Avatars rows in Table 5 correspond to the same re-run as in Table 1. Please add a footnote so the reader can track which numbers are re-run results and which are from the original publication.

Circularity Check

0 steps flagged · score 1.0 of 10

No load-bearing circularity: the fitting objective is anchored to external pretrained 2D cues and evaluated on held-out frames; the Avatars re-run and PSNR table discrepancy are verification concerns, not circular derivation.

full rationale

Walking the derivation chain: Sec. 3.3 defines the objective as L = Lgeo + λtex Ltex + R, where Lgeo is a sum of four losses (Lobj, Lpart, Lpix, Ltime), each supervised by pretrained external models (SAM, PartGLEE, CSE+Zoom-Out, BootsTAP). None of these loss terms is defined in terms of the SMAL parameters being optimized in a way that makes the fitted value equal to the supervision: Lobj is a Chamfer distance between projected vertices and SAM masks, Lpart and Lpix are reprojection errors to PartGLEE/CSE outputs, and Ltime is a tracking reprojection error. The dense feature network f: It -> (βt, θt) is trained per video under these losses, and evaluation uses interleaved 15-train/5-test frame blocks (Sec. 4.1), so test-frame metrics are not fit quantities. The only self-citation with author overlap is Animal3D [15], used to evaluate the BARC keypoint detector's unreliability in Sec. 1; this motivates the method but is not load-bearing for the reconstruction results. Appendix B.4's admission that the Avatars re-run is 'slightly lower than those reported in the original paper' is an evaluation/verification caveat: it affects whether the headline comparisons are confirmed, but it is not a case of a prediction reducing to its inputs by construction. Likewise, Table 1's RAC PSNR (21.86) exceeding 4D-Animal (21.28) contradicts the abstract's unqualified 'outperforms both model-based and model-free baselines,' but this is an internal-consistency error, not circularity. Overall, no equation, fitted parameter, or cited result is equivalent to the paper's own outputs, so the circularity score is minimal.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The paper introduces no new physical entities or mathematical axioms; the main assumptions are the reliability of pretrained 2D models and the expressiveness of the SMAL parameter space. The loss weights and sample counts are hand-set hyperparameters that affect the results.

free parameters (4)
  • lambda_obj, lambda_part, lambda_pix, lambda_time, lambda_tex = scheduled values, e.g., obj: 1, 100, 500, 800 at milestones 300, 1000, 6000
    Loss weights chosen by hand; different values affect final IoU and texture quality.
  • Ns (part sample count) = 200
    Number of 2D points sampled per part mask; manually set.
  • Nt (tracking point count) = 500 per key frame, 4 key frames
    Manual choice; ablation of this value is not reported.
  • confidence thresholds for part and CSE cues = 0.3 and 0.5
    Filters for unreliable pretrained outputs; chosen by hand.
assumptions (3)
  • domain assumption SMAL (dog-specific D-SMAL) parametrization plus vertex offsets is expressive enough to represent the target dogs' shape and pose.
    The method inherits the SMAL prior from [12],[14] and adds learned vertex offsets in the later training stage; if the template cannot represent the animal (e.g., unusual ear shape), reconstruction fails as acknowledged in Sec. 7.
  • domain assumption Pretrained 2D models (SAM, PartGLEE, CSE, BootsTAP) produce reliable and consistent cues on unseen videos.
    All four loss terms depend on these off-the-shelf models; failure cases in Sec. 7 are attributed to these cues being degraded (fast motion, occlusion, uncommon shapes).
  • domain assumption The Zoom-Out functional mapping between the CSE template and SMAL mesh preserves semantic correspondence.
    Used in Sec. 3.3 and A.3.3 to transfer CSE coordinates to SMAL vertices; if the mapping is misaligned, Lpix pulls vertices to wrong locations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of 4D-Animal: Freely Reconstructing Animatable 3D Animals from Videos." pith.science (2026). https://pith.science/paper/TUCH26AX

@misc{pith2026250710437,
  author       = {Pith},
  title        = {Pith review of: 4D-Animal: Freely Reconstructing Animatable 3D Animals from Videos},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TUCH26AX}},
  note         = {Machine review of arXiv:2507.10437}
}
read the original abstract

Existing methods for reconstructing animatable 3D animals from videos typically rely on sparse semantic keypoints to fit parametric models. However, obtaining such keypoints is labor-intensive, and keypoint detectors trained on limited animal data are often unreliable. To address this, we propose 4D-Animal, a novel framework that reconstructs animatable 3D animals from videos without requiring sparse keypoint annotations. Our approach introduces a dense feature network that maps 2D representations to SMAL parameters, enhancing both the efficiency and stability of the fitting process. Furthermore, we develop a hierarchical alignment strategy that integrates silhouette, part-level, pixel-level, and temporal cues from pre-trained 2D visual models to produce accurate and temporally coherent reconstructions across frames. Extensive experiments demonstrate that 4D-Animal outperforms both model-based and model-free baselines. Moreover, the high-quality 3D assets generated by our method can benefit other 3D tasks, underscoring its potential for large-scale applications. The code is released at https://github.com/zhongshsh/4D-Animal.

Figures

Figures reproduced from arXiv: 2507.10437 by the authors.

Figure 1
Figure 1. Visualization of 4D reconstruction results generated by the proposed 4D-Animal. Five [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Left: Evaluation of the keypoint detector from BARC [13] on the Animal3D dataset [15], using IoU and the 2D Percentage of Correct Keypoints (PCK) metric. The category is the super￾category defined in Animal3D. Right: Visualization of keypoint detector BARC and part segmentation model PartGLEE [16]. The leg keypoints are distinguished by connecting them with lines. 2 [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. (a) Left: Keypoint alignment (manually annotated) . Right: Part-level alignment (automatic) using semantic masks divided into head, body, feet, and tail, which are mapped to corresponding SMAL mesh regions. (b) Temporal tracking alignment. A tracked 2D point p1 in Frame 1 is matched to p2 in Frame 2. The nearest vertex v1 to p1 is found, and its corresponding vertex v2 is located via SMAL topology. The loss is compu… view at source ↗
Figures from the paper (20 more)
Figure 4
Figure 4. Figure 4: The overview of proposed 4D-Animal. (a) Mesh reconstruction. “C" stands for concatena [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Qualitative comparison. We compare our 4D-Animal to baselines by selecting images from [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Comparison of training efficiency between 4D-Animal and Avatars. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Comparison of the single-view image to 3D generation based on LGM [ [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Failure cases by 4D-Animal. 8 [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: The overview of proposed 4D-Animal. (a) Mesh reconstruction. “C" stands for concatena [PITH_FULL_IMAGE:figures/full_fig_p014_9.png]
Figure 10
Figure 10. Figure 10: (a) Left: Keypoint alignment (manually annotated) . Right: Part-level alignment (automatic) using semantic masks divided into head, body, feet, and tail, which are mapped to corresponding SMAL mesh regions. (b) Temporal tracking alignment. A tracked 2D point p1 in Fra…
Figure 11
Figure 11. Figure 11: (a) The initial pose of the mesh is often inaccurate. (b) In one-step fitting, the model [PITH_FULL_IMAGE:figures/full_fig_p017_11.png]
Figure 12
Figure 12. Figure 12: Comparison of training efficiency between 4D-Animal and Avatars. [PITH_FULL_IMAGE:figures/full_fig_p017_12.png]
Figure 13
Figure 13. Figure 13: Qualitative ablation of part-level alignment [PITH_FULL_IMAGE:figures/full_fig_p020_13.png]
Figure 14
Figure 14. Figure 14: Qualitative ablation of pixel-level alignment [PITH_FULL_IMAGE:figures/full_fig_p020_14.png]
Figure 15
Figure 15. Figure 15: Qualitative ablation of temporal-level alignment [PITH_FULL_IMAGE:figures/full_fig_p021_15.png]
Figure 16
Figure 16. Figure 16: Novel views of animatable 3D assets reconstructed using 4D-Animal. [PITH_FULL_IMAGE:figures/full_fig_p021_16.png]
Figure 17
Figure 17. Figure 17: Qualitative results of 4D-Animal on cat videos from COP3D. Despite using a dog-specific [PITH_FULL_IMAGE:figures/full_fig_p022_17.png]
Figure 18
Figure 18. Figure 18: Qualitative comparison. We compare our 4D-Animal with the state-of-the-art model [PITH_FULL_IMAGE:figures/full_fig_p023_18.png]
Figure 19
Figure 19. Figure 19: Qualitative comparison. We compare our 4D-Animal with the state-of-the-art model [PITH_FULL_IMAGE:figures/full_fig_p024_19.png]
Figure 20
Figure 20. Figure 20: Qualitative comparison. We compare our 4D-Animal with the state-of-the-art model [PITH_FULL_IMAGE:figures/full_fig_p025_20.png]
Figure 21
Figure 21. Figure 21: Qualitative comparison. We compare our 4D-Animal with the state-of-the-art model [PITH_FULL_IMAGE:figures/full_fig_p026_21.png]
Figure 22
Figure 22. Figure 22: Qualitative comparison. We compare our 4D-Animal with the state-of-the-art model [PITH_FULL_IMAGE:figures/full_fig_p027_22.png]
Figure 23
Figure 23. Figure 23: The performance of image-to-3D models SF3D [ [PITH_FULL_IMAGE:figures/full_fig_p028_23.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

69 extracted references · 46 canonical work pages

  1. [1]

    Animal avatars: Reconstructing animatable 3d animals from casual videos

    Remy Sabathier, Niloy J Mitra, and David Novotny. Animal avatars: Reconstructing animatable 3d animals from casual videos. In European Conference on Computer Vision, pages 270–287. Springer, 2024

    work page 2024

  2. [2]

    Fast encoder-based 3d from casual videos via point track processing

    Yoni Kasten, Wuyue Lu, and Haggai Maron. Fast encoder-based 3d from casual videos via point track processing. arXiv preprint arXiv:2404.07097, 2024

    arXiv 2024

  3. [3]

    Neural parametric gaussians for monocular non-rigid object reconstruction

    Devikalyan Das, Christopher Wewer, Raza Yunus, Eddy Ilg, and Jan Eric Lenssen. Neural parametric gaussians for monocular non-rigid object reconstruction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10715–10725, 2024

    work page 2024

  4. [4]

    Reconstructing animatable categories from videos

    Gengshan Yang, Chaoyang Wang, N Dinesh Reddy, and Deva Ramanan. Reconstructing animatable categories from videos. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 16995–17005, 2023

    work page 2023

  5. [5]

    Banmo: Building animatable 3d neural models from many casual videos

    Gengshan Yang, Minh V o, Natalia Neverova, Deva Ramanan, Andrea Vedaldi, and Hanbyul Joo. Banmo: Building animatable 3d neural models from many casual videos. In CVPR, 2022

    work page 2022

  6. [6]

    Learning the 3d fauna of the web

    Zizhang Li, Dor Litvak, Ruining Li, Yunzhi Zhang, Tomas Jakab, Christian Rupprecht, Shangzhe Wu, Andrea Vedaldi, and Jiajun Wu. Learning the 3d fauna of the web. In CVPR, 2024

    work page 2024

  7. [7]

    Ppr: Physically plausible reconstruction from monocular videos

    Gengshan Yang, Shuo Yang, John Z Zhang, Zachary Manchester, and Deva Ramanan. Ppr: Physically plausible reconstruction from monocular videos. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 3914–3924, 2023

    work page 2023

  8. [8]

    Smpl: A skinned multi-person linear model

    Matthew Loper, Naureen Mahmood, Javier Romero, Gerard Pons-Moll, and Michael J Black. Smpl: A skinned multi-person linear model. In Seminal Graphics Papers: Pushing the Boundaries, V olume 2, pages 851–866. 2023

    2023

Show all 69 references
  1. [9]

    Ghum & ghuml: Generative 3d human shape and articulated pose models

    Hongyi Xu, Eduard Gabriel Bazavan, Andrei Zanfir, William T Freeman, Rahul Sukthankar, and Cristian Sminchisescu. Ghum & ghuml: Generative 3d human shape and articulated pose models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 6...

    2020

  2. [10]

    Accelerating 3d deep learning with pytorch3d

    Nikhila Ravi, Jeremy Reizenstein, David Novotny, Taylor Gordon, Wan-Yen Lo, Justin Johnson, and Georgia Gkioxari. Accelerating 3d deep learning with pytorch3d. arXiv preprint arXiv:2007.08501, 2020

    2007 arXiv

  3. [11]

    Soft rasterizer: A differentiable renderer for image- based 3d reasoning

    Shichen Liu, Tianye Li, Weikai Chen, and Hao Li. Soft rasterizer: A differentiable renderer for image- based 3d reasoning. In Proceedings of the IEEE/CVF international conference on computer vision , pages 7708–7717, 2019

    2019

  4. [12]

    Bite: Beyond priors for improved three-d dog pose estimation

    Nadine Rüegg, Shashank Tripathi, Konrad Schindler, Michael J Black, and Silvia Zuffi. Bite: Beyond priors for improved three-d dog pose estimation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8867–8876, 2023

    2023

  5. [13]

    Barc: Learning to regress 3d dog shape from images by exploiting breed information

    Nadine Rueegg, Silvia Zuffi, Konrad Schindler, and Michael J Black. Barc: Learning to regress 3d dog shape from images by exploiting breed information. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3876–3884, 2022

    2022

  6. [14]

    3d menagerie: Modeling the 3d shape and pose of animals

    Silvia Zuffi, Angjoo Kanazawa, David W Jacobs, and Michael J Black. 3d menagerie: Modeling the 3d shape and pose of animals. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 6365–6373, 2017

    2017

  7. [15]

    Animal3d: A comprehensive dataset of 3d animal pose and shape

    Jiacong Xu, Yi Zhang, Jiawei Peng, Wufei Ma, Artur Jesslen, Pengliang Ji, Qixin Hu, Jiehua Zhang, Qihao Liu, Jiahao Wang, et al. Animal3d: A comprehensive dataset of 3d animal pose and shape. arXiv preprint arXiv:2308.11737, 2023. 10

    2023 arXiv

  8. [16]

    Partglee: A foundation model for recognizing and parsing any objects

    Junyi Li, Junfeng Wu, Weizhi Zhao, Song Bai, and Xiang Bai. Partglee: A foundation model for recognizing and parsing any objects. In European Conference on Computer Vision, pages 475–494. Springer, 2024

    2024

  9. [17]

    Sam 2: Segment anything in images and videos

    Nikhila Ravi, Valentin Gabeur, Yuan-Ting Hu, Ronghang Hu, Chaitanya Ryali, Tengyu Ma, Haitham Khedr, Roman Rädle, Chloe Rolland, Laura Gustafson, et al. Sam 2: Segment anything in images and videos. arXiv preprint arXiv:2408.00714, 2024

    2024 arXiv

  10. [18]

    Continuous surface embeddings

    Natalia Neverova, David Novotny, Vasil Khalidov, Marc Szafraniec, Patrick Labatut, and Andrea Vedaldi. Continuous surface embeddings. 2020

    2020

  11. [19]

    Bootstap: Bootstrapped training for tracking-any-point

    Carl Doersch, Pauline Luc, Yi Yang, Dilara Gokay, Skanda Koppula, Ankush Gupta, Joseph Heyward, Ignacio Rocco, Ross Goroshin, João Carreira, et al. Bootstap: Bootstrapped training for tracking-any-point. In Proceedings of the Asian Conference on Computer Vision , pages 3257–3274, 2024

    2024

  12. [20]

    Humans in 4D: Reconstructing and tracking humans with transformers

    Shubham Goel, Georgios Pavlakos, Jathushan Rajasegaran, Angjoo Kanazawa*, and Jitendra Malik*. Humans in 4D: Reconstructing and tracking humans with transformers. In International Conference on Computer Vision (ICCV), 2023

    2023

  13. [21]

    Efros, and Jitendra Malik

    Angjoo Kanazawa, Shubham Tulsiani, Alexei A. Efros, and Jitendra Malik. Learning category-specific mesh reconstruction from image collections. In ECCV, 2018

    2018

  14. [22]

    Lasr: Learning articulated shape reconstruction from a monocular video

    Gengshan Yang, Deqing Sun, Varun Jampani, Daniel Vlasic, Forrester Cole, Huiwen Chang, Deva Ramanan, William T Freeman, and Ce Liu. Lasr: Learning articulated shape reconstruction from a monocular video. In CVPR, 2021

    2021

  15. [23]

    Viser: Video-specific surface embeddings for articulated 3d shape reconstruction

    Gengshan Yang, Deqing Sun, Varun Jampani, Daniel Vlasic, Forrester Cole, Ce Liu, and Deva Ramanan. Viser: Video-specific surface embeddings for articulated 3d shape reconstruction. Advances in Neural Information Processing Systems, 34:19326–19338, 2021

    2021

  16. [24]

    Lassie: Learning articulated shape from sparse image ensemble via 3d part discovery

    Chun-Han Yao, Wei-Chih Hung, Yuanzhen Li, Michael Rubinstein, Ming-Hsuan Yang, and Varun Jampani. Lassie: Learning articulated shape from sparse image ensemble via 3d part discovery. In NeurIPS, 2022

    2022

  17. [25]

    Hi-lassie: High-fidelity articulated shape and skeleton discovery from sparse image ensemble

    Chun-Han Yao, Wei-Chih Hung, Yuanzhen Li, Michael Rubinstein, Ming-Hsuan Yang, and Varun Jampani. Hi-lassie: High-fidelity articulated shape and skeleton discovery from sparse image ensemble. In CVPR, 2023

    2023

  18. [26]

    Emerging properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Hervé Jégou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerging properties in self-supervised vision transformers. In Proceedings of the International Conference on Computer Vision (ICCV), 2021

    2021

  19. [27]

    Dove: Learning deformable 3d objects by watching videos

    Shangzhe Wu, Tomas Jakab, Christian Rupprecht, and Andrea Vedaldi. Dove: Learning deformable 3d objects by watching videos. International Journal of Computer Vision, 131(10):2623–2634, 2023

    2023

  20. [28]

    MagicPony: Learning articulated 3d animals in the wild

    Shangzhe Wu, Ruining Li, Tomas Jakab, Christian Rupprecht, and Andrea Vedaldi. MagicPony: Learning articulated 3d animals in the wild. In CVPR, 2023

    2023

  21. [29]

    Dynamic gaussians mesh: Consistent mesh reconstruction from monocular videos, 2024

    Isabella Liu, Hao Su, and Xiaolong Wang. Dynamic gaussians mesh: Consistent mesh reconstruction from monocular videos, 2024

    2024

  22. [30]

    3d gaussian splatting for real-time radiance field rendering

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Transactions on Graphics, 42(4), July 2023

    2023

  23. [31]

    Birds of a feather: Capturing avian shape models from images

    Yufu Wang, Nikos Kolotouros, Kostas Daniilidis, and Marc Badger. Birds of a feather: Capturing avian shape models from images. In Computer Vision and Pattern Recognition (CVPR), 2021

    2021

  24. [32]

    in the wild

    Silvia Zuffi, Angjoo Kanazawa, Tanya Berger-Wolf, and Michael J. Black. Three-d safari: Learning to estimate zebra pose, shape, and texture from images "in the wild". In International Conference on Computer Vision, October 2019

    2019

  25. [33]

    Who left the dogs out?: 3D animal reconstruction with expectation maximization in the loop

    Benjamin Biggs, Oliver Boyne, James Charles, Andrew Fitzgibbon, and Roberto Cipolla. Who left the dogs out?: 3D animal reconstruction with expectation maximization in the loop. In ECCV, 2020

    2020

  26. [34]

    Coarse-to-fine animal pose and shape estimation

    Chen Li and Gim Hee Lee. Coarse-to-fine animal pose and shape estimation. Advances in Neural Information Processing Systems, 34:11757–11768, 2021

    2021

  27. [35]

    Rgbd-dog: Predicting canine pose from rgbd sensors

    Sinead Kearney, Wenbin Li, Martin Parsons, Kwang In Kim, and Darren Cosker. Rgbd-dog: Predicting canine pose from rgbd sensors. In IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2020. 11

    2020

  28. [36]

    Creatures great and SMAL: Recovering the shape and motion of animals from video

    Benjamin Biggs, Thomas Roddick, Andrew Fitzgibbon, and Roberto Cipolla. Creatures great and SMAL: Recovering the shape and motion of animals from video. In ACCV, 2018

    2018

  29. [37]

    Silvia Zuffi, Ylva Mellbin, Ci Li, Markus Hoeschle, Hedvig Kjellström, Senya Polikovsky, Elin Hernlund, and Michael J. Black. V AREN: Very accurate and realistic equine network. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2024

    2024

  30. [38]

    Emerging properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Hervé Jégou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerging properties in self-supervised vision transformers. In Proceedings of the IEEE/CVF international conference on computer vision , pages 9650–9660, 2021

    2021

  31. [39]

    Efficient geometry-aware 3d generative adversarial networks

    Eric R Chan, Connor Z Lin, Matthew A Chan, Koki Nagano, Boxiao Pan, Shalini De Mello, Orazio Gallo, Leonidas J Guibas, Jonathan Tremblay, Sameh Khamis, et al. Efficient geometry-aware 3d generative adversarial networks. In Proceedings of the IEEE/CVF conference on computer vis...

    2022

  32. [40]

    Lightplane: Highly-scalable components for neural 3d fields

    Ang Cao, Justin Johnson, Andrea Vedaldi, and David Novotny. Lightplane: Highly-scalable components for neural 3d fields. arXiv preprint arXiv:2404.19760, 2024

    2024 arXiv

  33. [41]

    Creatures great and smal: Recovering the shape and motion of animals from video

    Benjamin Biggs, Thomas Roddick, Andrew Fitzgibbon, and Roberto Cipolla. Creatures great and smal: Recovering the shape and motion of animals from video. In Computer Vision–ACCV 2018: 14th Asian Conference on Computer Vision, Perth, Australia, December 2–6, 2018, Revised Select...

    2018

  34. [42]

    Zoomout: Spectral upsampling for efficient shape correspondence

    Simone Melzi, Jing Ren, Emanuele Rodola, Abhishek Sharma, Peter Wonka, and Maks Ovsjanikov. Zoomout: Spectral upsampling for efficient shape correspondence. arXiv preprint arXiv:1904.07865 , 2019

    1904 arXiv

  35. [43]

    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. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 586–595, 2018

    2018

  36. [44]

    As-rigid-as-possible shape manipulation

    Takeo Igarashi, Tomer Moscovich, and John F Hughes. As-rigid-as-possible shape manipulation. ACM transactions on Graphics (TOG), 24(3):1134–1141, 2005

    2005

  37. [45]

    Laplacian surface editing

    Olga Sorkine, Daniel Cohen-Or, Yaron Lipman, Marc Alexa, Christian Rössl, and H-P Seidel. Laplacian surface editing. In Proceedings of the 2004 Eurographics/ACM SIGGRAPH symposium on Geometry processing, pages 175–184, 2004

    2004

  38. [46]

    Laplacian mesh optimization

    Andrew Nealen, Takeo Igarashi, Olga Sorkine, and Marc Alexa. Laplacian mesh optimization. In Proceedings of the 4th international conference on Computer graphics and interactive techniques in Australasia and Southeast Asia, pages 381–389, 2006

    2006

  39. [47]

    Implicit fairing of irregular meshes using diffusion and curvature flow

    Mathieu Desbrun, Mark Meyer, Peter Schröder, and Alan H Barr. Implicit fairing of irregular meshes using diffusion and curvature flow. In Proceedings of the 26th annual conference on Computer graphics and interactive techniques, pages 317–324, 1999

    1999

  40. [48]

    Vggsfm: Visual geometry grounded deep structure from motion

    Jianyuan Wang, Nikita Karaev, Christian Rupprecht, and David Novotny. Vggsfm: Visual geometry grounded deep structure from motion. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 21686–21697, 2024

    2024

  41. [49]

    Flowmap: High-quality camera poses, intrinsics, and depth via gradient descent

    Cameron Smith, David Charatan, Ayush Tewari, and Vincent Sitzmann. Flowmap: High-quality camera poses, intrinsics, and depth via gradient descent. arXiv preprint arXiv:2404.15259, 2024

    2024 arXiv

  42. [50]

    Ep n p: An accurate o (n) solution to the p n p problem

    Vincent Lepetit, Francesc Moreno-Noguer, and Pascal Fua. Ep n p: An accurate o (n) solution to the p n p problem. International journal of computer vision , 81:155–166, 2009

    2009

  43. [51]

    Common pets in 3d: Dynamic new-view synthesis of real-life deformable categories

    Samarth Sinha, Roman Shapovalov, Jeremy Reizenstein, Ignacio Rocco, Natalia Neverova, Andrea Vedaldi, and David Novotny. Common pets in 3d: Dynamic new-view synthesis of real-life deformable categories. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern R...

    2023

  44. [52]

    Lgm: Large multi-view gaussian model for high-resolution 3d content creation

    Jiaxiang Tang, Zhaoxi Chen, Xiaokang Chen, Tengfei Wang, Gang Zeng, and Ziwei Liu. Lgm: Large multi-view gaussian model for high-resolution 3d content creation. In European Conference on Computer Vision, pages 1–18. Springer, 2024

    2024

  45. [53]

    Mvdream: Multi-view diffusion for 3d generation

    Yichun Shi, Peng Wang, Jianglong Ye, Mai Long, Kejie Li, and Xiao Yang. Mvdream: Multi-view diffusion for 3d generation. arXiv preprint arXiv:2308.16512, 2023. 12

    2023 arXiv

  46. [54]

    Consistent video depth estimation

    Xuan Luo, Jia-Bin Huang, Richard Szeliski, Kevin Matzen, and Johannes Kopf. Consistent video depth estimation. ACM Transactions on Graphics (ToG), 39(4):71–1, 2020

    2020

  47. [55]

    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. In Proceedings of the IEEE/CVF conference on computer vision and pattern rec...

    2023

  48. [56]

    Shic: Shape-image correspondences with no keypoint supervision

    Aleksandar Shtedritski, Christian Rupprecht, and Andrea Vedaldi. Shic: Shape-image correspondences with no keypoint supervision. In European Conference on Computer Vision , pages 129–145. Springer, 2024

    2024

  49. [57]

    Unsupervised learning of category- level 3d pose from object-centric videos

    Leonhard Sommer, Artur Jesslen, Eddy Ilg, and Adam Kortylewski. Unsupervised learning of category- level 3d pose from object-centric videos. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 22787–22796, 2024

    2024

  50. [58]

    4dpv: 4d pet from videos by coarse-to-fine non-rigid radiance fields

    Sergio M de Paco and Antonio Agudo. 4dpv: 4d pet from videos by coarse-to-fine non-rigid radiance fields. In Proceedings of the Asian Conference on Computer Vision , pages 2596–2612, 2024

    2024

  51. [59]

    Lepard: Learning explicit part discovery for 3d articulated shape reconstruction

    Di Liu, Anastasis Stathopoulos, Qilong Zhangli, Yunhe Gao, and Dimitris Metaxas. Lepard: Learning explicit part discovery for 3d articulated shape reconstruction. Advances in Neural Information Processing Systems, 36:54187–54198, 2023

    2023

  52. [60]

    Self-supervised single-view 3d reconstruction via semantic consistency

    Xueting Li, Sifei Liu, Kihwan Kim, Shalini De Mello, Varun Jampani, Ming-Hsuan Yang, and Jan Kautz. Self-supervised single-view 3d reconstruction via semantic consistency. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Par...

    2020

  53. [61]

    Saor: Single-view articulated object reconstruction

    Mehmet Aygun and Oisin Mac Aodha. Saor: Single-view articulated object reconstruction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 10382–10391, 2024

    2024

  54. [62]

    Mosca: Dynamic gaussian fusion from casual videos via 4d motion scaffolds

    Jiahui Lei, Yijia Weng, Adam Harley, Leonidas Guibas, and Kostas Daniilidis. Mosca: Dynamic gaussian fusion from casual videos via 4d motion scaffolds. arXiv preprint arXiv:2405.17421, 2024

    2024 arXiv

  55. [63]

    Shape of motion: 4d reconstruction from a single video

    Qianqian Wang, Vickie Ye, Hang Gao, Jake Austin, Zhengqi Li, and Angjoo Kanazawa. Shape of motion: 4d reconstruction from a single video. arXiv preprint arXiv:2407.13764, 2024

    2024

  56. [64]

    Sf3d: Stable fast 3d mesh re- construction with uv-unwrapping and illumination disentanglement

    Mark Boss, Zixuan Huang, Aaryaman Vasishta, and Varun Jampani. Sf3d: Stable fast 3d mesh re- construction with uv-unwrapping and illumination disentanglement. arXiv preprint arXiv:2408.00653, 2024

    2024 arXiv

  57. [65]

    Triplane meets gaussian splatting: Fast and generalizable single-view 3d reconstruction with transformers

    Zi-Xin Zou, Zhipeng Yu, Yuan-Chen Guo, Yangguang Li, Ding Liang, Yan-Pei Cao, and Song-Hai Zhang. Triplane meets gaussian splatting: Fast and generalizable single-view 3d reconstruction with transformers. In Proceedings of the IEEE/CVF conference on computer vision and pattern...

    2024

  58. [66]

    Crm: Single image to 3d textured mesh with convolutional reconstruction model

    Zhengyi Wang, Yikai Wang, Yifei Chen, Chendong Xiang, Shuo Chen, Dajiang Yu, Chongxuan Li, Hang Su, and Jun Zhu. Crm: Single image to 3d textured mesh with convolutional reconstruction model. In European Conference on Computer Vision, pages 57–74. Springer, 2024

    2024

  59. [67]

    3dtopia-xl: High-quality 3d pbr asset generation via primitive diffusion

    Zhaoxi Chen, Jiaxiang Tang, Yuhao Dong, Ziang Cao, Fangzhou Hong, Yushi Lan, Tengfei Wang, Haozhe Xie, Tong Wu, Shunsuke Saito, et al. 3dtopia-xl: High-quality 3d pbr asset generation via primitive diffusion. arXiv preprint arXiv:2409.12957, 8, 2024

    2024 arXiv

  60. [68]

    Instantmesh: Efficient 3d mesh generation from a single image with sparse-view large reconstruction models

    Jiale Xu, Weihao Cheng, Yiming Gao, Xintao Wang, Shenghua Gao, and Ying Shan. Instantmesh: Efficient 3d mesh generation from a single image with sparse-view large reconstruction models. arXiv preprint arXiv:2404.07191, 2024

    2024 arXiv

  61. [69]

    C" stands for concatena- tion. “Frame Index Emb

    Jianfeng Xiang, Zelong Lv, Sicheng Xu, Yu Deng, Ruicheng Wang, Bowen Zhang, Dong Chen, Xin Tong, and Jiaolong Yang. Structured 3d latents for scalable and versatile 3d generation. arXiv preprint arXiv:2412.01506, 2024. 13 A Additional details of 4D-Animal A.1 The details of te...

    2024 arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.