Pith. sign in

REVIEW 4 major objections 6 minor 72 references

HAMSt3R: Human-Aware Multi-view Stereo 3D Reconstruction

T0 review · 4 major / 6 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read The paper claims that HAMSt3R, a feed-forward MASt3R extension, reconstructs dense 3D scenes and humans jointly from sparse uncalibrated images, attaching instance and body-surface semantics to each 3D point.

desk verdict Useful integration, but the headline human-pose claim rests on an unablated SMPL-fitting pipeline; the new heads are never directly validated. read the letter →

arxiv 2508.16433 v1 pith:42CXOTFN submitted 2025-08-22 cs.CV

Sara Rojas , Matthieu Armando , Bernard Ghamen , Philippe Weinzaepfel , Vincent Leroy , Gregory Rogez This is my paper · ORCID
classification cs.CV
keywords multi-viewstereo3DhumanreconstructionuncalibratedimagesdensepointmapsPoseinstancesegmentationencoderdistillationSMPLfitting
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

HAMSt3R is the paper's proposal for making multi-view stereo reconstruction human-aware without giving up the speed of a single network. It extends the MASt3R pointmap architecture, which regresses dense 3D geometry from two uncalibrated images, by swapping in a distilled encoder that has learned both scene geometry and human-body representations, and by adding heads that segment people and regress continuous body-surface maps. The network therefore outputs, in one forward pass, both the 3D scene and, attached to every human point, which instance and which body-surface location it belongs to. The paper reports that this beats optimization-based joint reconstruction on world-coordinate pose error on one egocentric benchmark, runs in roughly a quarter of the time, and retains competitive accuracy on classical multi-view stereo and pose-estimation benchmarks. The significance, if true, is that human and scene understanding stop being separate optimization problems and become one feed-forward prediction.

What carries the argument

The load-bearing mechanism is the DUNE image encoder: a ViT backbone distilled from three teachers, a generalist self-supervised encoder, a multi-person human-mesh-recovery model, and the MASt3R encoder, so one frozen network provides features that are simultaneously good at scene geometry and human bodies. On top of that, the MASt3R dual-decoder with cross-attention produces view-consistent features, and three linear heads predict the pointmap, instance segmentation, and DensePose. The DensePose head regresses a continuous three-channel body-surface map plus a binary mask, an L2 regression rather than discrete part classification, which lets the same head attach SMPL-template coordinates to

What would settle it

Run HAMSt3R, MASt3R, and HSfM on a common held-out set of egocentric scenes with people far from the camera; if HAMSt3R's depth relative error is substantially above MASt3R's while its Procrustes-aligned pose error is above HSfM's, the central claim of preserving both scene and human performance fails.

Watch

Extended reading notes

Core claim

At its core, the paper claims that human semantics and scene geometry can be predicted from the same cross-attended image features, and that this joint prediction is what makes human-aware reconstruction efficient. Starting from MASt3R's dual-ViT decoder and pointmap regression, HAMSt3R uses a distilled encoder trained with both a scene-geometry teacher and a multi-person mesh-recovery teacher, so each image token carries both geometric and human-body knowledge. Three additional lightweight heads, instance segmentation, continuous DensePose regression, and a binary SMPL-region mask, turn those features into per-pixel human labels. Because the heads share the decoder's cross-view features, th

Load-bearing premise

The central claim rests on the frozen distilled encoder retaining MASt3R-grade scene geometry while gaining human understanding, and on synthetic indoor training transferring to real egocentric video; the paper's own large-scene and small-subject failure cases show this transfer is the fragile point.

Editorial extensions

If this is right

  • Feed-forward joint reconstruction removes the multi-stage optimization pipelines of prior joint methods, cutting runtime from roughly two minutes to about half a minute on a 4-view, 3-person scene.
  • Because human semantics are attached to 3D points, one can fit SMPL meshes directly to the reconstruction, enabling tracking, behavior analysis, and human–scene interaction applications.
  • The method remains usable for standard multi-view stereo and pose regression, with accuracy close to DUSt3R and MASt3R on non-human scenes despite the human-centric training mix.
  • For arbitrary numbers of views, pairwise predictions can be globally aligned with MASt3R's alignment procedure and ID correspondences resolved, producing a unified 3D scene with consistent human semantics.
  • The monocular case is a special case: feeding the same image twice yields human-aware reconstruction from a single image.

Reading between the lines

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

  • An untested extension: unfreezing or fine-tuning the distilled encoder on a scene-heavy mixed dataset might close the depth-accuracy gap on general benchmarks while keeping the human-pose gains, since the paper's own analysis points to the frozen encoder and loss scaling as the source of the drop.
  • The continuous DensePose regression head is a template-based surface regressor, so the same design could be applied to other articulated categories, such as animal bodies, with a suitable mesh template; the paper does not claim this.
  • The confidence-weighted aggregation of DensePose across pairs and the overlap-based ID matching could be upgraded to temporal tracking for video input, a direction the paper names as future work but does not test.
  • Because the method attaches per-point semantic labels before any optimization, it could serve as a fast differentiable initialization for optimization-based pipelines, potentially combining feed-forward efficiency with optimization accuracy.
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. HAMSt3R extends MASt3R to jointly reconstruct 3D scenes and humans from sparse, uncalibrated image pairs. It replaces the MASt3R encoder with the distilled DUNE encoder, adds instance segmentation, DensePose, and binary-mask heads, and trains on a mix of original MASt3R data and human-centric datasets (notably synthetic HumGen3D+Infinigen data). The claimed output is a dense point map with per-point human semantic information; for evaluation, SMPL is fitted to the predicted pointmaps using an external multi-view fitting framework. The paper evaluates human pose/camera metrics on EgoHumans and EgoExo4D, plus standard multi-view stereo and pose-regression benchmarks. The central claim is that this feed-forward architecture reconstructs humans effectively while preserving strong general-scene reconstruction performance.

Significance. If the contributions are substantiated, HAMSt3R would be a useful step toward unified, efficient human-scene reconstruction: it attaches semantics directly to 3D pointmaps and avoids the heavy per-scene optimization of methods like HSfM. The synthetic data pipeline is a practical contribution. However, the current evidence is conditional: Table 2 shows only mixed improvements over HSfM, the human-pose numbers are obtained through a separate SMPL-fitting optimization rather than the feed-forward network alone, and the segmentation/DensePose heads are not evaluated directly or ablated. The depth/pose regression tables also show a consistent drop relative to MASt3R and DUNE. With targeted ablations and a more calibrated set of claims, this could become a solid paper; as written, the central claims outrun the evidence.

major comments (4)
  1. [§4.1 and Table 2] The human-pose evaluation does not measure the feed-forward model: 3D joints are obtained by fitting SMPL to the predicted pointmaps via MvSMPLFitting with VPoser priors. No ablation compares (a) MASt3R pointmaps + the same SMPL fitting, (b) the frozen DUNE encoder with the original MASt3R pointmap head and no semantic heads, or (c) HAMSt3R without the new heads. Since DUNE already distills Multi-HMR, the W-MPJPE gains could come from the encoder or the external optimizer rather than from the proposed segmentation/DensePose heads. The segmentation and DensePose outputs themselves are never scored (no mAP, correspondences, or fitting accuracy), so the core 'semantic 3D human representation' is unverified. This ablation is load-bearing for the paper's main claim.
  2. [§4.2, Table 2, and Conclusion] The conclusion states that the method 'outperforms prior methods in estimating human poses', but Table 2 contradicts this on most metrics. On EgoHumans, HSfM is better on W-MPJPE (1.04 vs 3.80), GA-MPJPE (0.21 vs 0.42), and PA-MPJPE (0.05 vs 0.14). On EgoExo4D, HSfM is better on PA-MPJPE (0.06 vs 0.09) and HAMSt3R is better only on W-MPJPE (0.51 vs 0.56). The abstract's 'reconstruct humans effectively' is defensible only if framed as 'comparable or better in some world-coordinate metrics', not as general outperformance. The claim needs to be narrowed or the experiments need to support it.
  3. [§3.3 and §4.1] The abstract and introduction describe the approach as 'fully feed-forward', but the quantitative human-pose numbers require an optimization-based SMPL fitting stage, and the multi-view extension runs all image pairs and performs global alignment. Only the pairwise pointmap/semantics inference is feed-forward; the reported pose metrics are not end-to-end. This is a material overstatement when comparing against optimization-based baselines like HSfM, since the optimization is doing part of the work. Please state clearly which results are from the network alone and which require post-processing.
  4. [Tables 5 and 6] The claim that the method 'preserves strong performance in general 3D reconstruction tasks' is weakened by the numbers in the paper. In Table 6, HAMSt3R's average rel error is 4.26 versus 3.13 for MASt3R and 3.38 for DUNE, and it is worse than MASt3R on every dataset; in Table 5, CO3Dv2 mAA drops to 76.3 from MASt3R's 81.8. The paper acknowledges the drop, but the conclusion should be calibrated: the method does not preserve MASt3R-level geometry, it trades geometry accuracy for human semantics. If the authors wish to claim 'strong performance', they should specify a comparison threshold or a practical use case where this trade-off is acceptable.
minor comments (6)
  1. [Table 4] The value '16.303' in the EgoHumans Large AE row appears to be a typo (likely 16.30). Please check formatting.
  2. [§3.3 / Eq. (1)] The notation LMASt3R is used without a formal definition of its components; define it explicitly (pointmap loss, confidence loss, InfoNCE loss) so that the weighted sum in Eq. (1) is self-contained.
  3. [§4.1] PA-MPJPE and GA-MPJPE are not precisely defined. Please state which Procrustes/groups are used (e.g., aligned over all joints, per-person alignment, group alignment) since these metrics are central to the comparison.
  4. [Table 6 and §4.2] The row label 'DeepV2D (ScanNet)' is confusing because the column already says ScanNet; clarify whether DeepV2D is trained on ScanNet and evaluated on all datasets.
  5. [Supplementary, Loss Weight Selection] The loss weights are given as λ1=0.01, λ2=1, λ3=1, but there is no sensitivity analysis. Given the central role of the segmentation and DensePose heads, even a small ablation (e.g., setting λ1=0 or λ2=0) would strengthen the paper considerably.
  6. [General] No error bars or multiple-seed results are reported for any table. Since some differences are small (e.g., EgoExo4D W-MPJPE 0.51 vs 0.56), reporting variance or significance would help the reader assess robustness.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: DUNE/MASt3R/Multi-HMR are prior published models and the SMPL fitting used for pose metrics is disclosed post-processing; the absence of ablations is an attribution gap, not a circular reduction.

full rationale

The paper's derivation chain is not circular. It explicitly builds on prior published models: 'we exploit DUNE, a strong image encoder obtained by distilling, among others, the encoders from MASt3R and from a state-of-the-art HMR model, multi-HMR.' DUNE, MASt3R, and Multi-HMR are externally evaluated published models whose weights are used directly; citing them is normal transfer learning, not an unverified self-citation theorem. The human-pose numbers are not claimed as direct network outputs. The paper states: 'To obtain 3D joint predictions with our method, we fit SMPL to our predictions using an optimization procedure that minimizes the distance between all predicted 3D points of the person, and the corresponding vertices on the SMPL model,' and the supplement adds that 'SMPL fitting is performed only as a post-processing step for evaluation.' This is a disclosed evaluation-time fitting procedure, not a fitted parameter renamed as a prediction. The remaining concern — that the segmentation/DensePose heads are not ablated, so pose improvements could stem from the DUNE encoder or the SMPL optimizer — is a legitimate experimental-design/attribution gap, but it is not circularity: there is no equation in which an output is equivalent to an input by construction, and the paper evaluates on external benchmarks (EgoHumans, EgoExo4D, KITTI, ScanNet, etc.) with results that show trade-offs rather than tautology. No self-citation is used as a uniqueness theorem or as an unverified foundational premise. Therefore the appropriate finding is no significant circularity.

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

The central claim rests on the DUNE encoder and synthetic data, both from the authors' prior work or generated by the authors, with no external verification. The loss weights and data mix are tuned hyperparameters.

free parameters (3)
  • Loss weights lambda1, lambda2, lambda3 = 0.01, 1, 1
    Selected by scaling to typical loss magnitudes on a held-out validation set (Supp. Material). They directly affect the balance of segmentation, DensePose, and mask losses.
  • Data mixing ratio = 50% MASt3R / 50% human data
    The paper states this ratio is used per epoch with no systematic study of its effect.
  • Input resolution = max dimension 518 px
    All images are downscaled to this size, which affects performance on small humans (acknowledged in failure cases).
assumptions (3)
  • domain assumption Synthetic human data (HumGen3D + Infinigen, BEDLAM, etc.) is representative of real ego-centric scenes
    The model is trained largely on synthetic and studio data but evaluated on EgoHumans/EgoExo4D; the paper provides no direct evidence of domain transfer beyond the reported metrics.
  • domain assumption The DUNE encoder preserves MASt3R's geometric competence while adding human understanding
    Section 3.2 asserts this, but Table 6 shows depth accuracy degrades substantially, so the assumption is only partially satisfied.
  • domain assumption SMPL fitting from predicted 3D pointmaps is a valid way to evaluate human pose
    The MPJPE metrics are obtained by an offline SMPL fit to the pointmaps, not by the network directly, so the metric is sensitive to the fitting procedure.

how reviews work

0 comments
Cite this review

Pith. "Pith review of HAMSt3R: Human-Aware Multi-view Stereo 3D Reconstruction." pith.science (2026). https://pith.science/paper/42CXOTFN

@misc{pith2026250816433,
  author       = {Pith},
  title        = {Pith review of: HAMSt3R: Human-Aware Multi-view Stereo 3D Reconstruction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/42CXOTFN}},
  note         = {Machine review of arXiv:2508.16433}
}
read the original abstract

Recovering the 3D geometry of a scene from a sparse set of uncalibrated images is a long-standing problem in computer vision. While recent learning-based approaches such as DUSt3R and MASt3R have demonstrated impressive results by directly predicting dense scene geometry, they are primarily trained on outdoor scenes with static environments and struggle to handle human-centric scenarios. In this work, we introduce HAMSt3R, an extension of MASt3R for joint human and scene 3D reconstruction from sparse, uncalibrated multi-view images. First, we exploit DUNE, a strong image encoder obtained by distilling, among others, the encoders from MASt3R and from a state-of-the-art Human Mesh Recovery (HMR) model, multi-HMR, for a better understanding of scene geometry and human bodies. Our method then incorporates additional network heads to segment people, estimate dense correspondences via DensePose, and predict depth in human-centric environments, enabling a more comprehensive 3D reconstruction. By leveraging the outputs of our different heads, HAMSt3R produces a dense point map enriched with human semantic information in 3D. Unlike existing methods that rely on complex optimization pipelines, our approach is fully feed-forward and efficient, making it suitable for real-world applications. We evaluate our model on EgoHumans and EgoExo4D, two challenging benchmarks con taining diverse human-centric scenarios. Additionally, we validate its generalization to traditional multi-view stereo and multi-view pose regression tasks. Our results demonstrate that our method can reconstruct humans effectively while preserving strong performance in general 3D reconstruction tasks, bridging the gap between human and scene understanding in 3D vision.

Figures

Figures reproduced from arXiv: 2508.16433 by the authors.

Figure 1
Figure 1. Given a set of unposed images, HAMSt3R reconstructs the 3D scene as a dense point map with human semantics, attaching [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of HAMSt3R. From left to right: (1) Input stereo images are processed via a Siamese ViT encoder, (2) extracted features are passed to dual decoders with cross-attention, (3) separate heads generate 3D pointmaps and dense human semantic information, in the form of instance segmentation, DensePose, and binary mask predictions. (4) These outputs can be lifted to 3D using the Pointmaps and can be used, for exam… view at source ↗
Figure 3
Figure 3. Results on HumGen3D data (on a scene not seen dur￾ing training), using global alignment: Given a set of images of a scene (three out of eight of them are shown in the first column), we run our model on all possible image pairs, and aggregate pre￾dictions from the human heads in 2D, for each view (second and third columns). We apply the alignment method of MASt3R [33] to align the individual pointmaps in 3D (fourth c… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Illustration of the human-centric datasets used in this paper and listed in [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Qualitative results of HAMSt3R. Results from EgoExo4D (left) and EgoHumans (right). Each example includes point clouds (pink), instance segmentation (blue), and dense pose (green), with corresponding input images below (Best viewed when zoomed in). on RealEstate10K. Th…
Figure 6
Figure 6. Figure 6: Qualitative results of monocular prediction on in-the￾wild images taken from Pexels [2]. Each 2-row group shows: (top row, left to right) input image, high-confidence reconstructed point cloud overlaid with color (PointMap), and segmentation results; (bottom row) Dense…
Figure 7
Figure 7. Figure 7: Examples of failure cases. Top row: reconstruction failures on a scene from EgoHumans. Due to the large scene scale and wide camera angle (left), human subjects appear very small after downscaling, leading to noisy point clouds (right) and incor￾rect orientation in the…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

72 extracted references · 66 canonical work pages

  1. [1]

    https://www.humgen3d.com/

    Humgen3d. https://www.humgen3d.com/. 2, 5

    work page

  2. [2]

    https://www.pexels.com/

    Pexels. https://www.pexels.com/. Accessed: 2025- 07-29. 12

    work page 2025

  3. [3]

    Large-scale data for multiple-view stereopsis

    Henrik Aanæs, Rasmus Ramsbøl Jensen, George V ogiatzis, Engin Tola, and Anders Bjorholm Dahl. Large-scale data for multiple-view stereopsis. IJCV, 2016. 6

    work page 2016

  4. [4]

    Bundle adjustment in the large

    Sameer Agarwal, Noah Snavely, Steven M Seitz, and Richard Szeliski. Bundle adjustment in the large. In ECCV,

    work page

  5. [5]

    Cross-view and cross- pose completion for 3d human understanding

    Matthieu Armando, Salma Galaaoui, Fabien Baradel, Thomas Lucas, Vincent Leroy, Romain Br ´egier, Philippe Weinzaepfel, and Gr ´egory Rogez. Cross-view and cross- pose completion for 3d human understanding. In CVPR,

    work page

  6. [6]

    Multi-hmr: Multi-person whole-body hu- man mesh recovery in a single shot

    Fabien Baradel*, Matthieu Armando, Salma Galaaoui, Ro- main Br ´egier, Philippe Weinzaepfel, Gr ´egory Rogez, and Thomas Lucas*. Multi-hmr: Multi-person whole-body hu- man mesh recovery in a single shot. In ECCV, 2024. 2, 3, 6, 7

    work page 2024

  7. [7]

    Surf: Speeded up robust features

    Herbert Bay, Tinne Tuytelaars, and Luc Van Gool. Surf: Speeded up robust features. In ECCV, 2006. 2

    work page 2006

  8. [8]

    Bedlam: A synthetic dataset of bodies exhibiting de- tailed lifelike animated motion

    Michael J Black, Priyanka Patel, Joachim Tesch, and Jinlong Yang. Bedlam: A synthetic dataset of bodies exhibiting de- tailed lifelike animated motion. In CVPR, 2023. 5

    work page 2023

Show all 72 references
  1. [9]

    Gehler, Javier Romero, and Michael J

    Federica Bogo, Angjoo Kanazawa, Christoph Lassner, Pe- ter V . Gehler, Javier Romero, and Michael J. Black. Keep it SMPL: automatic estimation of 3d human pose and shape from a single image. In ECCV, 2016. 3

    2016

  2. [10]

    Scene coordinate reconstruction: Pos- ing of image collections via incremental learning of a relo- calizer

    Eric Brachmann, Jamie Wynn, Shuai Chen, Tommaso Cav- allari, ´Aron Monszpart, Daniyar Turmukhambetov, and Vic- tor Adrian Prisacariu. Scene coordinate reconstruction: Pos- ing of image collections via incremental learning of a relo- calizer. In ECCV, 2024. 2

    2024

  3. [11]

    Must3r: Multi-view network for stereo 3d reconstruc- tion

    Yohann Cabon, Lucas Stoffl, Leonid Antsfeld, Gabriela Csurka, Boris Chidlovskii, Jerome Revaud, and Vincent Leroy. Must3r: Multi-view network for stereo 3d reconstruc- tion. In CVPR, 2025. 3

    2025

  4. [12]

    HuMMan: Multi-modal 4d human dataset for versatile sensing and modeling

    Zhongang Cai, Daxuan Ren, Ailing Zeng, Zhengyu Lin, Tao Yu, Wenjia Wang, Xiangyu Fan, Yang Gao, Yifan Yu, Liang Pan, Fangzhou Hong, Mingyuan Zhang, Chen Change Loy, Lei Yang, and Ziwei Liu. HuMMan: Multi-modal 4d human dataset for versatile sensing and modeling. In ECCV, 2022. 5, 6

    2022

  5. [13]

    Per- pixel classification is not all you need for semantic segmen- tation

    Bowen Cheng, Alex Schwing, and Alexander Kirillov. Per- pixel classification is not all you need for semantic segmen- tation. In NeurIPS, 2021. 4

    2021

  6. [14]

    Masked-attention mask transformer for universal image segmentation

    Bowen Cheng, Ishan Misra, Alexander G Schwing, Alexan- der Kirillov, and Rohit Girdhar. Masked-attention mask transformer for universal image segmentation. In CVPR,

  7. [15]

    V . Choutas. Mvsmplfitting. https://github.com/ vchoutas/smplify-x. 6

  8. [16]

    Sfm with mrfs: Discrete-continuous optimization for large-scale structure from motion

    David J Crandall, Andrew Owens, Noah Snavely, and Daniel P Huttenlocher. Sfm with mrfs: Discrete-continuous optimization for large-scale structure from motion. IEEE trans. PAMI, 2012. 2

    2012

  9. [17]

    Hsfm: Hybrid structure-from-motion

    Hainan Cui, Xiang Gao, Shuhan Shen, and Zhanyi Hu. Hsfm: Hybrid structure-from-motion. In CVPR, 2017. 2

    2017

  10. [18]

    Scannet: Richly-annotated 3d reconstructions of indoor scenes

    Angela Dai, Angel X Chang, Manolis Savva, Maciej Hal- ber, Thomas Funkhouser, and Matthias Nießner. Scannet: Richly-annotated 3d reconstructions of indoor scenes. In CVPR, 2017. 6

    2017

  11. [19]

    Superpoint: Self-supervised Interest Point Detection and Description

    Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabi- novich. Superpoint: Self-supervised Interest Point Detection and Description. In CVPR, 2018. 2

    2018

  12. [20]

    An image is worth 16x16 words: Trans- formers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, et al. An image is worth 16x16 words: Trans- formers for image recognition at scale. In ICLR, 2021. 3

    2021

  13. [21]

    Mast3r- sfm: a fully-integrated solution for unconstrained structure- from-motion

    Bardienus Duisterhof, Lojze Zust, Philippe Weinzaepfel, Vincent Leroy, Yohann Cabon, and Jerome Revaud. Mast3r- sfm: a fully-integrated solution for unconstrained structure- from-motion. In 3DV, 2025. 2

    2025

  14. [22]

    Roma: Robust dense feature matching

    Johan Edstedt, Qiyu Sun, Georg B ¨okman, M ˚arten Wadenb¨ack, and Michael Felsberg. Roma: Robust dense feature matching. In CVPR, 2024. 2

    2024

  15. [23]

    Accurate, dense, and robust multiview stereopsis

    Yasutaka Furukawa and Jean Ponce. Accurate, dense, and robust multiview stereopsis. IEEE Trans. PAMI, 2010. 3

    2010

  16. [24]

    Vision meets robotics: The kitti dataset

    Andreas Geiger, Philip Lenz, Christoph Stiller, and Raquel Urtasun. Vision meets robotics: The kitti dataset. IJRR,

  17. [25]

    Humans in 4d: Re- constructing and tracking humans with transformers

    Shubham Goel, Georgios Pavlakos, Jathushan Rajasegaran, Angjoo Kanazawa, and Jitendra Malik. Humans in 4d: Re- constructing and tracking humans with transformers. In ICCV, 2023. 6

    2023

  18. [26]

    Densepose: Dense human pose estimation in the wild

    Rıza Alp G ¨uler, Natalia Neverova, and Iasonas Kokkinos. Densepose: Dense human pose estimation in the wild. In CVPR, 2018. 2, 3

    2018

  19. [27]

    Multiple view ge- ometry in computer vision

    Richard Hartley and Andrew Zisserman. Multiple view ge- ometry in computer vision . Cambridge university press,

  20. [28]

    Sherf: Generalizable human nerf from a single image

    Shoukang Hu, Fangzhou Hong, Liang Pan, Haiyi Mei, Lei Yang, and Ziwei Liu. Sherf: Generalizable human nerf from a single image. In ICCV, 2023. 3

    2023

  21. [29]

    Neuman: Neural human radiance field from a single video

    Wei Jiang, Kwang Moo Yi, Golnoosh Samei, Oncel Tuzel, and Anurag Ranjan. Neuman: Neural human radiance field from a single video. In ECCV, 2022. 3

    2022

  22. [30]

    3d gaussian splatting for real-time radiance field rendering

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk ¨uhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph., 2023. 3

    2023

  23. [31]

    Ego-humans: An ego- centric 3d multi-human benchmark

    Rawal Khirodkar, Aayush Bansal, Lingni Ma, Richard New- combe, Minh V o, and Kris Kitani. Ego-humans: An ego- centric 3d multi-human benchmark. In ICCV, 2023. 2, 6

    2023

  24. [32]

    Tanks and temples: Benchmarking large-scale scene reconstruction

    Arno Knapitsch, Jaesik Park, Qian-Yi Zhou, and Vladlen Koltun. Tanks and temples: Benchmarking large-scale scene reconstruction. ACM Trans. Graphics, 2017. 6

    2017

  25. [33]

    Ground- ing image matching in 3d with mast3r

    Vincent Leroy, Yohann Cabon, and Jerome Revaud. Ground- ing image matching in 3d with mast3r. In ECCV, 2024. 2, 3, 5, 6, 7, 8

    2024

  26. [34]

    Ego-exo: Transferring visual representations from third-person to first-person videos

    Yanghao Li, Tushar Nagarajan, Bo Xiong, and Kristen Grau- man. Ego-exo: Transferring visual representations from third-person to first-person videos. In CVPR, 2021. 2, 6

    2021

  27. [35]

    3d human pose and shape estimation through collaborative learning and multi-view model-fitting

    Zhongguo Li, Magnus Oskarsson, and Anders Heyden. 3d human pose and shape estimation through collaborative learning and multi-view model-fitting. In WACV, 2021. 3

    2021

  28. [36]

    Barf: Bundle-adjusting neural radiance fields

    Chen-Hsuan Lin, Wei-Chiu Ma, Antonio Torralba, and Si- mon Lucey. Barf: Bundle-adjusting neural radiance fields. In ICCV, 2021. 2

    2021

  29. [37]

    Joint optimization for 4d human-scene reconstruction in the wild

    Zhizheng Liu, Joe Lin, Wayne Wu, and Bolei Zhou. Joint optimization for 4d human-scene reconstruction in the wild. arXiv preprint arXiv:2501.02158, 2025. 2, 3

    2025 arXiv

  30. [38]

    Matthew Loper, Naureen Mahmood, Javier Romero, Gerard Pons-Moll, and Michael J. Black. SMPL: a skinned multi- person linear model. ACM Trans. Graphics, 2015. 2, 4

    2015

  31. [39]

    Sift-the scale invariant feature transform

    G Lowe. Sift-the scale invariant feature transform. IJCV,

  32. [40]

    Troje, Ger- ard Pons-Moll, and Michael J

    Naureen Mahmood, Nima Ghorbani, Nikolaus F. Troje, Ger- ard Pons-Moll, and Michael J. Black. AMASS: Archive of motion capture as surface shapes. In ICCV, 2019. 5

    2019

  33. [41]

    Reconstructing people, places, and cameras

    Lea M ¨uller, Hongsuk Choi, Anthony Zhang, Brent Yi, Jiten- dra Malik, and Angjoo Kanazawa. Reconstructing people, places, and cameras. In CVPR, 2025. 2, 3, 6, 7, 12

    2025

  34. [42]

    Dinov2: Learning robust visual features without supervision

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. TMLR, 2024. 3

    2024

  35. [43]

    Georgios Pavlakos, Vasileios Choutas, Nima Ghorbani, Timo Bolkart, Ahmed A. A. Osman, Dimitrios Tzionas, and Michael J. Black. Expressive body capture: 3d hands, face, and body from a single image. In CVPR, 2019. 6

    2019

  36. [44]

    Infinigen indoors: Photorealistic indoor scenes using procedural generation

    Alexander Raistrick, Lingjie Mei, Karhan Kayan, David Yan, Yiming Zuo, Beining Han, Hongyu Wen, Meenal Parakh, Stamatis Alexandropoulos, Lahav Lipson, Zeyu Ma, and Jia Deng. Infinigen indoors: Photorealistic indoor scenes using procedural generation. In CVPR, 2024. 2, 5

    2024

  37. [45]

    Vi- sion transformers for dense prediction

    Ren ´e Ranftl, Alexey Bochkovskiy, and Vladlen Koltun. Vi- sion transformers for dense prediction. In ICCV, 2021. 3

    2021

  38. [46]

    Com- mon objects in 3d: Large-scale learning and evaluation of real-life 3d category reconstruction

    Jeremy Reizenstein, Roman Shapovalov, Philipp Henzler, Luca Sbordone, Patrick Labatut, and David Novotn ´y. Com- mon objects in 3d: Large-scale learning and evaluation of real-life 3d category reconstruction. In ICCV, 2021. 2, 6, 7

    2021

  39. [47]

    Grounded sam: Assembling open-world models for diverse visual tasks

    Tianhe Ren, Shilong Liu, Ailing Zeng, Jing Lin, Kunchang Li, He Cao, Jiayu Chen, Xinyu Huang, Yukang Chen, Feng Yan, et al. Grounded sam: Assembling open-world models for diverse visual tasks. arXiv preprint arXiv:2401.14159 ,

    arXiv

  40. [48]

    R2d2: Reliable and repeatable detec- tor and descriptor

    Jerome Revaud, Cesar De Souza, Martin Humenberger, and Philippe Weinzaepfel. R2d2: Reliable and repeatable detec- tor and descriptor. In NeurIPS, 2019. 2

    2019

  41. [49]

    Orb: An efficient alternative to sift or surf

    Ethan Rublee, Vincent Rabaud, Kurt Konolige, and Gary Bradski. Orb: An efficient alternative to sift or surf. InICCV,

  42. [50]

    Unic: Universal classification models via multi-teacher distillation

    Mert B ¨ulent Sarıyıldız, Philippe Weinzaepfel, Thomas Lu- cas, Diane Larlus, and Yannis Kalantidis. Unic: Universal classification models via multi-teacher distillation. InECCV,

  43. [51]

    Dune: Distilling a universal encoder from heterogeneous 2d and 3d teachers

    Mert B ¨ulent Sarıyıldız, Philippe Weinzaepfel, Thomas Lu- cas, Pau de Jorge, Diane Larlus, and Yannis Kalantidis. Dune: Distilling a universal encoder from heterogeneous 2d and 3d teachers. In CVPR, 2025. 2, 3, 8

    2025

  44. [52]

    SuperGlue: Learning feature matching with graph neural networks

    Paul-Edouard Sarlin, Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabinovich. SuperGlue: Learning feature matching with graph neural networks. In CVPR, 2020. 2

    2020

  45. [53]

    Structure-from-motion revisited

    Johannes Lutz Sch ¨onberger and Jan-Michael Frahm. Structure-from-motion revisited. In CVPR, 2016. 1, 2

    2016

  46. [54]

    Pixelwise view selection for un- structured multi-view stereo

    Johannes Lutz Sch ¨onberger, Enliang Zheng, Marc Pollefeys, and Jan-Michael Frahm. Pixelwise view selection for un- structured multi-view stereo. In ECCV, 2016. 1, 2

    2016

  47. [55]

    A multi-view stereo benchmark with high- resolution images and multi-camera videos

    Thomas Schops, Johannes L Schonberger, Silvano Galliani, Torsten Sattler, Konrad Schindler, Marc Pollefeys, and An- dreas Geiger. A multi-view stereo benchmark with high- resolution images and multi-camera videos. In CVPR, 2017. 6

    2017

  48. [56]

    Splatt3r: Zero-shot gaussian splatting from uncalibrated image pairs

    Brandon Smart, Chuanxia Zheng, Iro Laina, and Vic- tor Adrian Prisacariu. Splatt3r: Zero-shot gaussian splatting from uncalibrated image pairs. arXiv preprint arXiv:2408.13912, 2024. 2

    2024 arXiv

  49. [57]

    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. In 3DV, 2025. 2

    2025

  50. [58]

    Photo tourism: exploring photo collections in 3d

    Noah Snavely, Steven M Seitz, and Richard Szeliski. Photo tourism: exploring photo collections in 3d. In SIGGRAPH,

  51. [59]

    Surface capture for performance-based animation

    Jonathan Starck and Adrian Hilton. Surface capture for performance-based animation. IEEE Computer Graphics and Applications, 2007. 3

    2007

  52. [60]

    Loftr: Detector-free local feature matching with transformers

    Jiaming Sun, Zehong Shen, Yuang Wang, Hujun Bao, and Xiaowei Zhou. Loftr: Detector-free local feature matching with transformers. In CVPR, 2021. 2

    2021

  53. [61]

    Deepv2d: Video to depth with differentiable structure from motion

    Zachary Teed and Jia Deng. Deepv2d: Video to depth with differentiable structure from motion. arXiv preprint arXiv:1812.04605, 2018. 7, 8

    2018 arXiv

  54. [62]

    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 CVPR, 2024. 2

    2024

  55. [63]

    Continuous 3d per- ception model with persistent state

    Qianqian Wang, Yifei Zhang, Aleksander Holynski, Alexei A Efros, and Angjoo Kanazawa. Continuous 3d per- ception model with persistent state. In CVPR, 2025. 2, 3

    2025

  56. [64]

    Dust3r: Geometric 3d vi- sion made easy

    Shuzhe Wang, Vincent Leroy, Yohann Cabon, Boris Chidlovskii, and Jerome Revaud. Dust3r: Geometric 3d vi- sion made easy. In CVPR, 2024. 2, 3, 6, 7, 8, 12

    2024

  57. [65]

    Level- s2 fm: Structure from motion on neural level set of implicit surfaces

    Yuxi Xiao, Nan Xue, Tianfu Wu, and Gui-Song Xia. Level- s2 fm: Structure from motion on neural level set of implicit surfaces. In CVPR, 2023. 2

    2023

  58. [66]

    Multi-view multi-person 3d pose estimation with uncalibrated camera networks

    Yan Xu and Kris Kitani. Multi-view multi-person 3d pose estimation with uncalibrated camera networks. In BMVC,

  59. [67]

    Decoupling human and camera motion from videos in the wild

    Vickie Ye, Georgios Pavlakos, Jitendra Malik, and Angjoo Kanazawa. Decoupling human and camera motion from videos in the wild. In CVPR, 2023. 3

    2023

  60. [68]

    Multi- view human body reconstruction from uncalibrated cameras

    Zhixuan Yu, Linguang Zhang, Yuanlu Xu, Chengcheng Tang, Luan Tran, Cem Keskin, and Hyun Soo Park. Multi- view human body reconstruction from uncalibrated cameras. In NeurIPS, 2022. 3

    2022

  61. [69]

    Monst3r: A simple approach for estimating geometry in the presence of motion

    Junyi Zhang, Charles Herrmann, Junhwa Hur, Varun Jam- pani, Trevor Darrell, Forrester Cole, Deqing Sun, and Ming- Hsuan Yang. Monst3r: A simple approach for estimating geometry in the presence of motion. In ICLR, 2025. 2, 3

    2025

  62. [70]

    Ego- body: Human body shape and motion of interacting people from head-mounted devices

    Siwei Zhang, Qianli Ma, Yan Zhang, Zhiyin Qian, Taein Kwon, Marc Pollefeys, Federica Bogo, and Siyu Tang. Ego- body: Human body shape and motion of interacting people from head-mounted devices. In ECCV, 2022. 5, 6

    2022

  63. [71]

    Wang, Bhiksha Raj, Min Xu, Jimei Yang, and Chun-Hao Paul Huang

    Yizhou Zhao, Tuanfeng Y . Wang, Bhiksha Raj, Min Xu, Jimei Yang, and Chun-Hao Paul Huang. Synergistic global- space camera and human reconstruction from videos.CVPR,

  64. [72]

    Stereo magnification: Learning view syn- thesis using multiplane images

    Tinghui Zhou, Richard Tucker, John Flynn, Graham Fyffe, and Noah Snavely. Stereo magnification: Learning view syn- thesis using multiplane images. In SIGGRAPH, 2018. 2, 6, 7 Supplementary Material Loss Weight Selection The loss weights λ1, λ2, and λ3 were selected to balance t...

    2018

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.