Pith. sign in

REVIEW 4 major objections 5 minor 54 references

4DPV: 4D Pet from Videos by Coarse-to-Fine Non-Rigid Radiance Fields

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A coarse-to-fine neural deformation model recovers camera pose and 4D shape of unknown deformable objects from casual multi-view videos, without 3D templates or 3D training data.

desk verdict Solid incremental extension of BANMo with a local quadratic deformation model, but the 'unknown object' claim is unvalidated because all experiments depend on pretrained category-specific DensePose CSE. read the letter →

arxiv 2411.10275 v1 pith:UKQFAGBW submitted 2024-11-15 cs.CV

Sergio M. de Paco , Antonio Agudo This is my paper · ORCID
classification cs.CV
keywords 4Dreconstructionnon-rigidradiancefieldsneuraldeformationmodelcoarse-to-finecameraposeestimationself-supervisedlearningcontinuoussurfaceembeddingsanimal
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 claims that simultaneous camera-pose estimation and high-detail 4D reconstruction of an unknown, freely deforming object can be achieved from multiple RGB videos in the wild, with no 3D template and no 3D supervision. The key is a coarse-to-fine neural deformation model: a linear blend skinning component captures articulated motion, while a newly introduced local quadratic deformation model with spatio-temporal smoothness captures fine, non-rigid details. Combined with canonical-space embeddings that provide cross-sequence correspondences, the model is trained self-supervised by rendering and comparing against input images. If correct, this would make detailed 4D capture of animals and other deformable objects practical from ordinary phone video, removing the need for category-specific 3D models or controlled capture setups.

What carries the argument

The method combines three learned modules: (1) a coarse neural radiance field with Linear Blend Skinning (LBS) that maps points between canonical and image space, estimating root and bone poses from latent codes; (2) a fine deformation network, the Local Quadratic Model (LQM), which takes the coarse canonical point and a deformation code and outputs a 3x9 matrix of linear, quadratic, and cross-term coefficients, applied to an extended coordinate vector to produce a refined point, with losses enforcing spatial and temporal smoothness of the coefficient matrices; and (3) a fine canonical network with extra samples to refine shape and appearance. Cross-sequence correspondence is anchored by Continuous Surface Embeddings (CSE) learned in canonical space and matched to DensePose predictions in 2D, plus 2D/3D cycle consistency and optical-flow reprojection losses.

What would settle it

Run the pipeline on a deformable category with no pre-trained CSE model (e.g., a horse or a cloth toy captured in several videos) and check whether the recovered canonical shape, per-frame meshes, and camera trajectories stay consistent across sequences; also ablate by removing the CSE losses (Eq. 17) and measuring how much the Chamfer distance and novel-view PSNR degrade on the existing cat and dog datasets.

Watch

Extended reading notes

Core claim

The central claim is that decomposing the deformation into a coarse articulated part (linear blend skinning) and a fine part (a neural local quadratic model with spatial and temporal coherence) lets a single self-supervised pipeline jointly estimate camera poses, an implicit canonical shape, and a time-varying deformation field from multiple monocular RGB videos, even under uncontrolled lighting and unknown camera motion. The fine quadratic model, applied per point via an MLP, captures bending, stretching, and twisting that piecewise-linear models miss, while the fine canonical network avoids over-constraining the surface and preserves detail. The paper reports quantitative and qualitative improvements over BANMo and ViSER on synthetic and real animal and human datasets, including higher mesh fidelity and fewer artifacts such as extra legs or heads.

Load-bearing premise

The method relies on pre-trained category-specific surface embeddings (DensePose/CSE) to supply dense correspondences across sequences; for an object category without such a model, the cross-sequence anchor disappears and the self-supervised pipeline has no way to align the videos.

Editorial extensions

If this is right

  • If the method works as claimed, detailed 4D reconstruction of pets and other deformable animals becomes possible from casual multi-view phone footage, with no 3D scanning or category-specific template.
  • The coarse-to-fine decomposition suggests a general recipe: use a physically motivated coarse model (e.g., skinning) to avoid local minima, then add a local higher-order deformation to recover fine surface detail without losing global consistency.
  • The learned canonical shape and deformation can be reused for novel-view synthesis at arbitrary time steps, enabling free-viewpoint video of the reconstructed object.
  • The self-supervised nature of the pipeline means it could scale to large video collections of many objects, since it does not depend on 3D ground-truth data.
  • The ablation study indicates that both the fine deformation model and the fine canonical network contribute to quality, and the spatio-temporal smoothness terms are necessary to avoid artifacts.

Reading between the lines

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

  • The method is not fully category-agnostic as stated, because the CSE correspondences rely on a pre-trained DensePose model that exists only for a few categories (cat, dog, human); a direct test on a category without such an embedding would show how far the self-supervised pipeline can stretch.
  • A natural extension would be to replace the fixed CSE with online-learned embeddings from the video collection itself, which would break the dependency on category-specific models and potentially generalize to arbitrary deformable objects.
  • The local quadratic model is a second-order polynomial deformation field per point; one could view it as a truncated Taylor expansion, and higher-order terms or a learned basis might capture even finer effects such as fur or cloth wrinkles, at the cost of more regularization.
  • The requirement for silhouettes and optical flow from off-the-shelf networks suggests that end-to-end learning of these prerequisites within the same framework, or making the method robust to their noise, would be a direct testable improvement.
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 / 5 minor

Summary. The paper proposes 4DPV, a coarse-to-fine neural deformation model for jointly recovering camera pose and 4D shape of a dynamic object from multiple RGB videos without a 3D template or 3D training data. The method combines a Linear Blend Skinning coarse model, a local quadratic deformation network with spatio-temporal smoothness, and a fine canonical network, trained self-supervised with photometric, silhouette, optical-flow, CSE, and cycle-consistency losses. It is evaluated on four synthetic datasets (AMA-swing, AMA-samba, Eagle, Hands) and three real datasets (casual-cat, adult-5, casual-dog), with comparisons to BANMo and ViSER and an ablation on view synthesis.

Significance. If the claims hold, the contribution is a plausible extension of BANMo that improves fine detail in non-rigid NeRF reconstruction, and the authors release code, which aids reproducibility. The core architectural idea of combining LBS coarse deformation with a local quadratic deformation model is reasonable and is evaluated on both synthetic and in-the-wild datasets. However, the paper's central generality claim ('unknown object') is not fully supported because the method depends on category-specific DensePose CSE supervision, and the empirical support for 'outperforms state-of-the-art' is weakened by inconsistent results in Table 1 and the absence of error bars or significance tests. The manuscript is worth a major revision rather than rejection, since the identified issues are addressable with additional ablations, corrected metrics, and phrase revision.

major comments (4)
  1. [Abstract; Section 3.1; Eq. (17); Eq. (19)] The claim that the method recovers an 'unknown object' is load-bearing, but the optimization relies on category-specific Continuous Surface Embeddings from a pretrained DensePose network [26] through L_cse2D and L_cse3D in Eq. (17), which are part of the global objective in Eq. (19). All evaluated categories (cat, dog, human) already have pretrained CSE models; for a genuinely novel deformable category without a pretrained embedding model, the cross-sequence correspondence anchor disappears. The paper does not provide a controlled ablation that removes the CSE losses, and the qualitative note in Fig. 3 about 'without 3D CSE constraining' is not quantified in Table 2. Thus the abstract's 'unknown object' and 'no 3D training data' wording goes beyond what the formulation and experiments demonstrate.
  2. [Section 6.3; Table 1] The claim of outperforming state-of-the-art approaches is not consistently supported by Table 1. The full method is worse than its own baseline on Hands (CD 6.0 vs. 5.4), worse than the '+ fcn' variant on Eagle (5.1 vs. 4.7) and Hands (6.0 vs. 4.8), and worse than the original BANMo on AMA-swing (10.0 vs. 9.1). No error bars, multiple-seed runs, or significance tests are reported, so the 'best on average' conclusion is fragile. The authors should report variance and/or a paired comparison to substantiate the superiority claim.
  3. [Eq. (15); Section 5] The fine silhouette loss is defined as ||s_c(\bar{x}) - s_f(\bar{x})||^2, comparing the coarse and fine silhouette predictions against each other rather than comparing the fine prediction to the ground-truth mask s(\bar{x}). This contradicts the preceding sentence, which states the silhouette loss penalizes deviations between the predicted 2D shape and the segmentation mask. If implemented exactly as written, the fine network is not directly supervised by the ground-truth silhouette; the authors should correct the formula or clarify the intended teacher-student behavior.
  4. [Section 6.3; Table 2; Fig. 3] The ablation study in Table 2 evaluates view synthesis quality (PSNR/SSIM) rather than 3D reconstruction accuracy, while the main quantitative comparison in Table 1 reports Chamfer distance and F-score. These two rankings can differ, and indeed the '+ fcn' variant already outperforms the full method on two datasets in Table 1. The qualitative claim in Fig. 3 about omitting 3D CSE constraining is not backed by a quantitative ablation isolating the CSE terms. The component analysis therefore does not clearly establish which proposed module is responsible for the reported gains.
minor comments (5)
  1. [Table 1] Several F@2% entries are missing (BANMo on AMA-samba, ViSER on AMA-samba and possibly others); the authors should either report the values or explicitly state why they are omitted.
  2. [Section 6.1] The sentence 'We obtain the weight coefficients experimentally and fix them for all experiments, obtaining a non-overfitted solution' makes an unsupported claim; reporting a small hyperparameter sweep or a sensitivity check would be more appropriate.
  3. [Eq. (12) and Eq. (13)] The notation A^t_{s,k} appears in the loss formulas but is not defined; clarify that it denotes MLP_Q evaluated at the sampled neighboring point.
  4. [Section 6.3] The phrase 'we also employ this algorithm in our training setup' is ambiguous because it is unclear whether 'this algorithm' refers to BANMo or ViSER; rephrase for clarity.
  5. [General] The text contains many OCR-like artifacts (e.g., 'coarse-to-\ufb01ne', 'de\ufb01ne', 'con\ufb01gurations') and inconsistent hyphenation; a careful proofreading pass is needed.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the method is a self-supervised fitting pipeline supervised by external measurements (pixels, masks, optical flow, DensePose CSE) and evaluated on held-out views and ground-truth meshes.

full rationale

All loss terms in Eq. (19) are supervised by external measurements: ground-truth pixels in Eq. (14), object masks in Eq. (15), off-the-shelf optical flow in Eq. (16), and the pretrained DensePose/CSE network in Eq. (17); none of these targets is produced by the model being trained. The learned canonical embeddings are matched to DensePose outputs rather than defining those outputs, so the correspondence signal is an external prior, not a self-referential fit. The quantitative claims (Table 1 Chamfer distance/F-score, Table 2 PSNR/SSIM) are computed against ground-truth synthetic meshes and held-out rendered views, not against training quantities, so no fitted constant is renamed as a prediction. The fine silhouette term in Eq. (15) uses the coarse silhouette as a teacher target, but the coarse silhouette is itself pinned to external masks; this is a student-teacher design, not circularity. The only self-citations (e.g., [3], [33]) are contextual related-work references and are not load-bearing. The paper's reliance on DensePose CSE does limit the abstract's 'unknown object' claim to categories with a pretrained embedding model, but that is an unvalidated generality and scope concern, not a circular derivation.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The central claim rests on the volume rendering model and LBS articulation priors inherited from BANMo and NeRF, plus the new local quadratic deformation assumption. The paper adds no new physical entities. The learned latent codes and loss weights are optimization/hyperparameters, not independent evidence. The key external dependencies are CSE embeddings, silhouettes, and optical flow from off-the-shelf networks.

free parameters (3)
  • loss weights λ in Eq. (19) = λ_pf=0.1, λ_sf=1, λ_Q=1e3, λ_pc=0.1, λ_sc=1, λ_of=0.1, λ_reg=0.02
    Set experimentally and fixed across all experiments; no sensitivity analysis or validation split reported, so results may depend on these choices.
  • number of bones B = 25
    Follows the LBS setup of BANMo [48]; affects the capacity of the coarse deformation model.
  • smoothness neighborhood size K = 6
    Number of sampled neighbors in the spatial/temporal smoothness losses (Eq. 12-13), chosen by hand.
assumptions (6)
  • standard math Volume rendering equation (Eq. 1) is an accurate model of image formation.
    The rendering equation integrates density and color along rays; this is the standard NeRF assumption.
  • domain assumption Linear blend skinning with 25 bones can represent the coarse deformation of an arbitrary animal or human.
    LBS is designed for articulated bodies (Eq. 6-7); for non-articulated objects or very soft bodies, the coarse initialization may be insufficient.
  • domain assumption A local quadratic map (Eq. 9) with per-point MLP coefficients can capture fine non-rigid deformations.
    Assumes the residual deformation after LBS is locally quadratic; this is a modeling assumption inherited from NRSfM.
  • domain assumption Continuous Surface Embeddings (CSE) provide correct dense correspondences across sequences for the object category.
    Used in Eq. (17) to anchor canonical embeddings; DensePose/CSE is trained on specific categories and may fail on others.
  • domain assumption Off-the-shelf silhouette and optical flow estimates are sufficiently accurate.
    Segmentation (PointRend) and flow (VCN-robust) are precomputed and used as supervision (Section 6.2); errors in these propagate to the reconstruction.
  • ad hoc to paper Photometric, silhouette, flow, CSE, and cycle losses are sufficient to disambiguate the non-rigid shape and camera pose.
    The optimization landscape may still get stuck in local minima without 3D supervision; the paper observes over-constraining from flow and CSE and removes some constraints (Section 4.2), suggesting the balance is delicate.

how reviews work

0 comments
Cite this review

Pith. "Pith review of 4DPV: 4D Pet from Videos by Coarse-to-Fine Non-Rigid Radiance Fields." pith.science (2026). https://pith.science/paper/UKQFAGBW

@misc{pith2026241110275,
  author       = {Pith},
  title        = {Pith review of: 4DPV: 4D Pet from Videos by Coarse-to-Fine Non-Rigid Radiance Fields},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UKQFAGBW}},
  note         = {Machine review of arXiv:2411.10275}
}
read the original abstract

We present a coarse-to-fine neural deformation model to simultaneously recover the camera pose and the 4D reconstruction of an unknown object from multiple RGB sequences in the wild. To that end, our approach does not consider any pre-built 3D template nor 3D training data as well as controlled illumination conditions, and can sort out the problem in a self-supervised manner. Our model exploits canonical and image-variant spaces where both coarse and fine components are considered. We introduce a neural local quadratic model with spatio-temporal consistency to encode fine details that is combined with canonical embeddings in order to establish correspondences across sequences. We thoroughly validate the method on challenging scenarios with complex and real-world deformations, providing both quantitative and qualitative evaluations, an ablation study and a comparison with respect to competing approaches. Our project is available at https://github.com/smontode24/4DPV.

Figures

Figures reproduced from arXiv: 2411.10275 by the authors.

Figure 1
Figure 1. 4D Shapes from multiple videos. Given multiple monocular videos of an unknown and dynamic object shape with non-controlled illumination conditions, our method can learn a coarse￾to-ne neural deformation model without considering any 3D template or the camera locations. Our algorithm retrieves RGB appearance, an implicit canonical 3D shape as well as a time￾varying deformation model. Once learned, our algorithm can … view at source ↗
Figure 2
Figure 2. Self-supervised coarse-to-ne approach to capture 4D shapes from multiple videos. LBS model points at image t are mapped into the coarse canonical space. Then, they are rened with the ne regularized non-rigid deformation and an additional ne canonical network to render the RGB image. 6.1 Optimization and Implementation details Our implementation of implicit shape and appearance models uses NeRF [24] and BANMo [48… view at source ↗
Figure 3
Figure 3. Qualitative ablation in the casual-cat dataset. From left to right: BANMo base￾line [48]; Our method with a coarse canonical network together with the ne deformation model; Our full approach with ne canonical and deformation networks. two intermediate solutions (just adding the ne canonical network and just adding the ne deformation one) and the one with our full algorithm [PITH_FULL_IMAGE:figures/full_fig_p012… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: CSE embeddings of deformed views in the casual-dog dataset. From left to right: Input frame, CSE-colored mesh visualized in camera, side and top views, respectively. Mean PSNR ↑ Mean SSIM ↑ Baseline 33.439 0.957 Coarse + deformation code 35.185 0.954 Coarse + ray smoot…
Figure 5
Figure 5. Figure 5: Qualitative evaluation. The same information is provided in every column for the datasets casual-cat, adult-5, eagle and casual-dog. Top: Some pictures in the input dataset. Mid￾dle: 3D color mesh inferred by BANMo [48]. Bottom: Our estimation. In all cases, meshes are…
Figure 6
Figure 6. Figure 6: Qualitative evaluation of deformed views in the [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 51 canonical work pages

  1. [26]

    NeurIPS (2020)

    Neverova, N., Novotny, D., Szafraniec, M., Khalidov, V ., Labatut, P ., V edaldi, A.: Continuous surface embeddings. NeurIPS (2020)

    work page 2020

  2. [1]

    In: ICCV (2009)

    Agarwal, S., Snavely, N., Simon, I., Seitz, S.M., Szeliski, R.: Building Rome in a day. In: ICCV (2009)

    work page 2009

  3. [2]

    In: ICASSP (2022)

    Agudo, A.: Safari from visual signals: Recovering volumetric 3D shapes. In: ICASSP (2022)

    work page 2022

  4. [3]

    TPAMI44(1), 519–532 (2022)

    Agudo, A.: Unsupervised 3D reconstruction and grouping of rigid and non-rigid categories. TPAMI44(1), 519–532 (2022)

    work page 2022

  5. [4]

    In: CVPR (2017)

    Agudo, A., Moreno-Noguer, F.: DUST: Dual union of spatio-temporal subspaces for monoc- ular multiple object 3D reconstruction. In: CVPR (2017)

    work page 2017

  6. [5]

    TPAMI40(9), 2137–2150 (2018)

    Agudo, A., Moreno-Noguer, F.: Force-based representation for non-rigid shape and elastic model estimation. TPAMI40(9), 2137–2150 (2018)

    work page 2018

  7. [6]

    TPAMI38(5), 979–994 (2016)

    Agudo, A., Moreno-Noguer, F., Calvo, B., Montiel, J.M.M.: Sequential non-rigid structure from motion using physical priors. TPAMI38(5), 979–994 (2016)

    work page 2016

  8. [7]

    TPAMI33(7), 1442–1456 (2011)

    Akhter, I., Sheikh, Y ., Khan, S., Kanade, T.: Trajectory space: A dual representation for nonrigid structure from motion. TPAMI33(7), 1442–1456 (2011)

    work page 2011

Show all 54 references
  1. [8]

    In: ICCV (2021)

    Chen, X., Zheng, Y ., Black, M.J., Hilliges, O., Geiger, A.: Snarf: Differentiable forward skinning for animating non-rigid neural implicit shapes. In: ICCV (2021)

    2021

  2. [9]

    In: ECCV (2010)

    Fayad, J., Agapito, L., Del Bue, A.: Piecewise quadratic reconstruction of non-rigid surfaces from monocular sequences. In: ECCV (2010)

    2010

  3. [10]

    In: BMVC (2009)

    Fayad, J., Del Bue, A., Agapito, L., Aguiar, P .: Non-rigid structure from motion using quadratic deformation models. In: BMVC (2009)

    2009

  4. [11]

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

    Fridovich-Keil, S., Meanti, G., Warburg, F.R., Recht, B., Kanazawa, A.: K-planes: Explicit radianceelds in space, time, and appearance. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 12479–12488 (2023)

    2023

  5. [12]

    In: CVPR (2013)

    Garg, R., Roussos, A., Agapito, L.: Dense variational reconstruction of non-rigid surfaces from monocular video. In: CVPR (2013)

    2013

  6. [13]

    TPAMI33(10), 2051–2065 (2011)

    Gotardo, P .F.U., Martinez, A.M.: Computing smooth time-trajectories for camera and de- formable shape in structure from motion with occlusion. TPAMI33(10), 2051–2065 (2011)

    2011

  7. [14]

    In: CVPR (2020)

    Kirillov, A., Wu, Y ., He, K., Girshick, R.: Pointrend: Image segmentation as rendering. In: CVPR (2020)

    2020

  8. [15]

    In: ICCV (2019)

    Kong, C., Lucey, S.: Deep non-rigid structure from motion. In: ICCV (2019)

    2019

  9. [16]

    In: CVPR (2023)

    Kuai, T., Karthikeyan, A., Kant, Y ., Mirzaei, A., Gilitschenski, I.: CAMM: Building category-agnostic and animatable 3D models from monocular videos. In: CVPR (2023)

    2023

  10. [17]

    In: CVPR (2020)

    Kulkarni, N., Gupta, A., Fouhey, D., Tulsiani, S.: Articulation-aware canonical surface map- ping. In: CVPR (2020)

    2020

  11. [18]

    PR77(11), 428–443 (2017)

    Kumar, S., Dai, Y ., Li, H.: Spatio-temporal union of subspaces for multi-body non-rigid structure-from-motion. PR77(11), 428–443 (2017)

    2017

  12. [19]

    In: CVPR (2016)

    Lee, M., Cho, J., Oh, S.: Consensus of non-rigid reconstructions. In: CVPR (2016)

    2016

  13. [20]

    In: CVPR (2014)

    Lee, M., Choi, C.H., Oh, S.: A procrustean markov process for non-rigid structure recovery. In: CVPR (2014)

    2014

  14. [21]

    In: ACM SIGGRAPH (2000)

    Lewis, J., Cordner, M., Fong, N.: Pose space deformation: a unied approach to shape inter- polation and skeleton-driven deformation. In: ACM SIGGRAPH (2000)

    2000

  15. [22]

    In: CVPR (2021)

    Li, Z., Nuklaus, S., Snavely, N., Wang, O.: Neural sceneowelds for space-time view synthesis of dynamic scenes. In: CVPR (2021)

    2021

  16. [23]

    TOG34(6), 1–16 (2015)

    Loper, M., Mahmood, N., Romero, J., Pons-Moll, G., Black, M.J.: SMPL: A skinned multi- person linear model. TOG34(6), 1–16 (2015)

    2015

  17. [24]

    In: ECCV (2020)

    Mildenhall, B., Srinivasan, P ., Tancik, M., Barron, J., Ramamoorthi, R., Ng, R.: Nerf: Rep- resenting scenes as neural radianceelds for view synthesis. In: ECCV (2020)

    2020

  18. [25]

    arXiv preprint arXiv:2201.05989 (2022) 16 S

    Müller, T., Evans, A., Schied, C., Keller, A.: Instant neural graphics primitives with a mul- tiresolution hash encoding. arXiv preprint arXiv:2201.05989 (2022) 16 S. M. de Paco and A. Agudo

    2022 arXiv

  19. [27]

    In: CVPR (2021)

    Neverova, N., Sanakoyeu, A., Labatut, P ., Novotny, D., V edaldi, A.: Discovering relation- ships between object categories via universal canonical maps. In: CVPR (2021)

    2021

  20. [28]

    In: CVPR (2010)

    Newcome, R., Davison, A.J.: Live dense reconstruction with a single moving camera. In: CVPR (2010)

    2010

  21. [29]

    In: CVPR (2022)

    Niemeyer, M., Barron, J., Mildenhall, B., Sajjadi, M., Geiger, A., Radwan, N.: Regnerf: Regularizing neural radianceelds for view synthesis from sparse inputs. In: CVPR (2022)

    2022

  22. [30]

    In: ICCV (2021)

    Park, K., Sinha, U., Barron, J., Bouaziz, S., Goldman, D., Seitz, S., Martin-Brualla, R.: Ner- es: Deformable neural radianceelds. In: ICCV (2021)

    2021

  23. [31]

    In: CVPR (2021)

    Pumarola, A., Corona, E., Pons-Moll, G., Moreno-Noguer, F.: D-nerf: Neural radianceelds for dynamic scenes. In: CVPR (2021)

    2021

  24. [32]

    In: CVPR (2016)

    Schonberger, J., Frahm, J.: Structure-from-motion revisited. In: CVPR (2016)

    2016

  25. [33]

    In: ECCV (2020)

    Sidhu, V ., Tretschk, E., Golyanik, V ., Agudo, A., Theobalt, C.: Neural dense non-rigid struc- ture from motion with latent space constraints. In: ECCV (2020)

    2020

  26. [34]

    arXiv preprint arXiv:2306.05872 (2023)

    Sklyarova, V ., Chelishev, J., Dogaru, A., Medvedev, I., Lempitsky, V ., Zakharov, E.: Neu- ral haircut: Prior-guided strand-based hair reconstruction. arXiv preprint arXiv:2306.05872 (2023)

    2023 arXiv

  27. [35]

    IEEE Transactions on Visualization and Computer Graphics29(5), 2732–2742 (2023)

    Song, L., Chen, A., Li, Z., Chen, Z., Chen, L., Y uan, J., Xu, Y ., Geiger, A.: Nerfplayer: A streamable dynamic scene representation with decomposed neural radianceelds. IEEE Transactions on Visualization and Computer Graphics29(5), 2732–2742 (2023)

    2023

  28. [36]

    Tatarchenko, M., Richter, S., Stephan, R., Ranftl, R., Li, Z., Koltun, V ., Brox, T.: What do single-view 3D reconstruction networks learn? In: CVPR (2019)

    2019

  29. [37]

    TPAMI30(5), 878–892 (2008)

    Torresani, L., Hertzmann, A., Bregler, C.: Nonrigid structure-from-motion: estimating shape and motion with hierarchical priors. TPAMI30(5), 878–892 (2008)

    2008

  30. [38]

    In: ICCV (2021)

    Tretschk, E., Tewari, A., Golyanik, V ., Zollhöfer, M., Lassner, C., Theobalt, C.: Non-rigid neural radianceelds: Reconstruction and novel view synthesis of a dynamic scene from monocular video. In: ICCV (2021)

    2021

  31. [39]

    In: CVPR (2016)

    Ulusoy, A.O., Black, M.J., Geiger, A.: Patches, planes and probabilities: A non-local prior for volumetric 3D reconstruction. In: CVPR (2016)

    2016

  32. [40]

    arXiv preprint arXiv:2112.03907 (2021)

    V erbin, D., Hedman, P ., Mildenhall, B., Zickler, T., Barron, J., Srinivasan, P .: Ref- nerf: Structured view-dependent appearance for neural radianceelds. arXiv preprint arXiv:2112.03907 (2021)

    2021 arXiv

  33. [41]

    In: ACM SIGGRAPH (2008)

    Vlasic, D., Baran, I., Matusik, W., Popovi ´c, J.: Articulated mesh animation from multi-view silhouettes. In: ACM SIGGRAPH (2008)

    2008

  34. [42]

    In: CVPR (2021)

    Wang, C., Lucey, S.: Paul: Procrustean autoencoder for unsupervised lifting. In: CVPR (2021)

    2021

  35. [43]

    In: CVPR (2022)

    Xu, Q., Xu, Z., Philip, J., Bi, S., Shu, Z., Sunkavalli, K., Neumann, U.: Point-nerf: Point- based neural radianceelds. In: CVPR (2022)

    2022

  36. [44]

    In: ICCV (2019)

    Xu, X., Dunn, E.: Discrete Laplace operator estimation for dynamic 3D reconstruction. In: ICCV (2019)

    2019

  37. [45]

    In: NeurIPS (2019)

    Y ang, G., Ramanan, D.: V olumetric correspondence networks for opticalow. In: NeurIPS (2019)

    2019

  38. [46]

    In: CVPR (2021)

    Y ang, G., Sun, D., Jampani, V ., Vlasic, D., Cole, F., Chang, H., Ramanan, D., Freeman, W., Liu, C.: Lasr: Learning articulated shape reconstruction from a monocular video. In: CVPR (2021)

    2021

  39. [47]

    In: NeurIPS (2021)

    Y ang, G., Sun, D., Jampani, V ., Vlasic, D., Cole, F., Liu, C., Ramanan, D.: Viser: Video- specic surface embeddings for articulated 3D shape recon- struction. In: NeurIPS (2021)

    2021

  40. [48]

    In: CVPR (2022) 4DPV 17

    Y ang, G., V o, M., Neverova, N., Ramanan, D., V edaldi, A., Joo, H.: BANMo: Building ani- matable 3D neural models from many casual videos. In: CVPR (2022) 4DPV 17

    2022

  41. [49]

    NeurIPS (2021)

    Y ariv, L., Gu, J., Kasten, Y ., Lipman, Y .: V olume rendering of neural implicit surfaces. NeurIPS (2021)

    2021

  42. [50]

    In: ICCV (2015)

    Y u, R., Russell, C., Campbell, N., Agapito, L.: Direct, dense, and deformable: Template- based non-rigid 3D reconstruction from rgb video. In: ICCV (2015)

    2015

  43. [51]

    In: CVPR (2022)

    Zhang, X., Bi, S., Sunkavalli, K., Su, H., Xu, Z.: Nerfusion: Fusing radianceelds for large- scale scene reconstruction. In: CVPR (2022)

    2022

  44. [52]

    In: CVPR (2014)

    Zhu, Y ., Huang, D., De La Torre, F., Lucey, S.: Complex non-rigid motion 3D reconstruction by union of subspaces. In: CVPR (2014)

    2014

  45. [53]

    in the wild

    Zuf, S., Kanazawa, A., Berger-Wolf, T., Black, M.J.: Three-D safari: Learning to estimate zebra pose, shape, and texture from images "in the wild". In: ICCV (2019)

    2019

  46. [54]

    In: CVPR (2018)

    Zuf, S., Kanazawa, A., Black, M.J.: Lions and tigers and bears: Capturing non-rigid, 3D, articulated shape from images. In: CVPR (2018)

    2018

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.