Pith. sign in

REVIEW 3 major objections 7 minor 36 references

The paper claims that a single head trajectory can be turned into a natural full-body mesh in one generation step, without iterative diffusion, by aligning the latent space of a head-conditioned variational autoencoder with that of a full-b

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

A one-step guided VAE reconstructs full-body SMPL-H meshes from head pose alone, outperforming diffusion-based ego-body baselines with a claimed >50x speedup.

T0 review reviewed 2026-08-01 challenge →

load-bearing objection A solid, incremental VAE-based alternative to diffusion for head-only ego-body reconstruction, with good ablations, a real speed advantage, and one mechanism question that is worth raising but not fatal. the 3 major comments →

arxiv 2607.27755 v1 pith:OA6EIZUJ submitted 2026-07-30 cs.CV

EgoGVAE: Ego-body Mesh Reconstruction via Guided Variational Autoencoder

Jaehun Jung , Wonjun Kim This is my paper
classification cs.CV
keywords ego-body mesh reconstructionhead-to-motion generationvariational autoencoderlatent-space guidancetransformerone-step sampling3D human poseegocentric perception
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

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 full-body meshes can be reconstructed from only the wearer's head pose using a guided variational autoencoder, avoiding the expensive iterative denoising of diffusion-based approaches. Its central idea is to train a head-to-motion network whose Gaussian latent distribution is pulled toward the latent distribution of a motion-to-motion network that sees full-body poses. At inference, one sample from the aligned head-conditioned latent decodes into a plausible complete body pose. This is reported to cut mean per-joint error by about 11 percent relative to the strongest diffusion-based prior method, while running more than 50 times faster (0.026 seconds for 128 frames). If true, the result would make ego-body avatar reconstruction practical for real-time head-mounted and smart-glass applications.

Core claim

The method's central claim is that a variational autoencoder trained on full-body pose sequences can serve as a guide for a head-to-motion network: by enforcing similarity between the Gaussian latent distribution of the guidance network and that of the head-to-motion network, a latent sampled once from the head-conditioned distribution can be decoded into a natural full-body pose. The guidance network and head-to-motion network share a motion decoder; during training, a symmetric Kullback-Leibler divergence aligns the two Gaussians, a regularizer pulls both toward the standard normal, and learnable tokens appended to head embeddings stand in for unobserved body parts. The guidance is removed

What carries the argument

The central object is the guided variational autoencoder (EgoGVAE): two transformer-based VAEs, one motion-to-motion guide and one head-to-motion network, sharing a single motion decoder. The workhorse is the alignment of sequence-level Gaussian latents via a symmetric KL divergence, combined with a normal-prior regularizer; learnable tokens concatenated to the head embeddings provide the network with placeholders for missing limbs and let the head-conditioned encoder place its latent inside the guided distribution.

Load-bearing premise

The whole pipeline rests on a single sequence-level Gaussian latent being rich enough to summarize the full-body pose distribution, so that aligning two such Gaussians transfers enough information; if the true pose distribution is multimodal or the latent capacity is too small, one-step sampling cannot deliver reliable poses.

What would settle it

Train the same architecture on a motion set with clearly separated modes (e.g., walking, climbing, dancing), then sample repeatedly from the aligned head-conditioned latent for a fixed head trajectory: if the decoded poses cluster only around an average mode and miss rare modes, or if per-mode reconstruction error exceeds the reported MPJPE, the Gaussian-alignment mechanism is not carrying the full distribution.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Head-only egocentric pose estimation becomes fast enough for real-time wearable use: 0.026 seconds per 128 frames, more than 50 times faster than diffusion-based approaches, with lower reported joint error.
  • Because the guidance network is used only during training, deployment can use a lightweight head-to-motion network without extra hand or IMU sensors.
  • The aligned Gaussian latents produce smoother motion, with lower jitter and foot sliding than the diffusion baseline, and retain temporal consistency on longer sequences via sliding-window inference.
  • The same online sliding-window procedure yields a prediction every 26 ms per frame, making interactive avatar applications feasible.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • An implication the paper does not explore: the same latent-alignment trick could transfer other privileged signals (e.g., hand positions, sparse IMU data) into a cheap inference network, so the guidance source need not be head pose only.
  • A risk the paper does not address: a single sequence-level Gaussian latent may collapse genuinely multimodal pose distributions; for highly varied motions, one-step sampling could produce average-looking poses, and a mixture or hierarchical latent would be a natural extension to test.
  • The speed advantage is demonstrated on a fixed model size (13.88M parameters); the online setting still costs 26 ms per frame, so the real-time claim depends on hardware and sequence length.
  • The gains on a real-world dataset are smaller than on synthetic motion capture, so testing on noisier in-the-wild head trajectories would clarify whether the guidance benefit persists outside clean benchmarks.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 7 minor

Summary. The paper proposes EgoGVAE, a two-branch transformer-based variational autoencoder for reconstructing full-body SMPL-H meshes from a sequence of head poses only. A 'motion-to-motion' network encodes ground-truth full-body pose sequences into a Gaussian latent N(µ_M, Σ_M); a 'head-to-motion' network encodes head embeddings plus learnable tokens into a second Gaussian N(µ_H, Σ_H). Training minimizes reconstruction, velocity, and two KL terms: one aligning the two latent distributions (symmetric KL) and one pulling them toward N(0,I). At inference, the head-to-motion branch samples z_H once and decodes it with a shared transformer decoder. Experiments on AMASS, RICH, and EgoBody report lower MPJPE/PA-MPJPE than diffusion baselines (e.g., 106.7/89.9 vs 119.7/101.1 for EgoAllo on AMASS) and much faster batch inference (0.026 s for 128 frames). Ablations in Tables 5–7 assess guidance, learnable tokens, loss terms, and guidance design.

Significance. If the empirical results are reproducible, the paper offers a practical non-iterative alternative to diffusion-based head-to-body pose generation, with the added benefit of public code and model release. The joint training of a guidance VAE and a head-conditioned VAE is a reasonable idea, and the ablations show that both the guidance loss and the learnable tokens contribute to accuracy. The method is tested on three datasets, including real VR head-pose data (EgoBody), which is a strength. However, the central mechanism is presented as a transfer of latent information, but the current experiments do not exclude a simpler regularizer effect, and the claimed speed advantage is not maintained under the paper's own online sliding-window protocol. These issues need to be resolved before the contribution can be fully assessed.

major comments (3)
  1. [§4.3, Table 4; §4.5, Table 9] The paper's headline speed claim ('one-step sampling achieves remarkably fast inference (more than 50 times faster)', Abstract) is based on Table 4, which reports 0.026 s for a single 128-frame forward pass. However, §3.4 states that for online/arbitrary-length inputs the window slides one frame at a time, and Table 9 reports EgoGVAE as 26 ms per frame — a factor of ~128 larger than the batch-per-frame cost. Under this online protocol the method is not 50× faster than diffusion baselines; indeed, Table 9 shows it is 15–17× slower than AvatarPoser/EgoPoser. The efficiency comparison must be conducted under the same protocol (online sliding-window for all methods) or the speed claim should be explicitly restricted to offline batch processing. This is load-bearing because real-time inference is a core motivation of the paper.
  2. [§3.2, Fig. 2; §4.4, Table 5] The paper's causal claim in §3.2 and the Conclusion — that aligning N(µ_H,Σ_H) with N(µ_M,Σ_M) allows the head-to-motion branch to decode natural poses — is not isolated by the ablations. The two decoder branches use different queries: the head branch uses head/learnable-token embeddings, the motion branch uses full-body motion tokens (Fig. 2). A symmetric KL on the latent key/value space does not align the query distributions, so the observed gain from the guidance term (Table 5, 125.6→106.7 MPJPE) could stem from the KL loss acting as a regularizer on the head encoder rather than from 'latent transfer'. To substantiate the mechanism, the authors should add an ablation replacing the guidance KL with an equally strong regularizer, or exchanging the decoder queries, or measuring whether z_H can be decoded by the motion-branch decoder. Without this, the proposed guidance mechanism is not e
  3. [§4.2, Tables 1–3, 5–7] All tables report a single run with no variance information. Given the stochastic nature of transformer training, the ≈10% MPJPE improvement over EgoAllo (Table 1) and the ablations (Tables 5–7) may be within run-to-run noise. The authors should report mean±std over at least 3 seeds, and preferably a paired evaluation for ablations. This is needed to establish the significance of the reported gains.
minor comments (7)
  1. [Fig. 1, §3.2] Typo: 'EgoGAVE' in the caption of Fig. 1 should read 'EgoGVAE'.
  2. [Eq. (1)] The covariance matrices Σ in Eq. (1) are not stated to be diagonal. For the closed-form Gaussian KL to apply, Σ must be diagonal. Please state this explicitly.
  3. [§4.2] The definition of the Ground metric is ambiguous. It should be made clear whether the vertical distance is absolute (always non-negative) and how the ground plane is obtained in each dataset.
  4. [Table 2] Typesetting error: the EgoPoser row in Table 2 shows '5.129.5' for Jitter; this should be '5.1' and '29.5' or similar.
  5. [§3.2] The text uses 'motion encoder' for both the shared encoder and the head-to-motion encoder; the weight-sharing arrangement between the two networks should be clarified.
  6. [Fig. 3] The t-SNE visualization is qualitative; a quantitative measure of latent alignment (e.g., MMD or average symmetric KL) during training would strengthen the claim of Fig. 3.
  7. [§4.3] A deterministic (non-variational) head-to-motion baseline that uses the same transformer architecture but without latent sampling would help isolate the benefit of the variational/guided formulation.

Circularity Check

0 steps flagged

No significant circularity: the only self-citation is non-load-bearing related work; all headline metrics come from held-out public dataset splits.

full rationale

The paper's central claim is that enforcing similarity between the latent distributions of the head-to-motion network and the motion-to-motion network (Eqs. 1–2) enables one-step sampling from the guided latent to decode natural full-body poses. This is a learned objective, not a definition of the evaluation targets. The reported MPJPE, PA-MPJPE, and related metrics in Tables 1–3, 8–9 are computed on held-out test splits of AMASS, RICH, and EgoBody following the EgoAllo split protocol, so the headline numbers are not fitted inputs or restatements of the training loss. The guidance network does observe full-body ground truth during training, but this is standard supervised learning, not circular fitting. The only author-overlapping reference is Eigenpose [9], discussed in Section 2.1 as related work; it is not used in the proposed method, in the loss equations, or in any comparison protocol. No uniqueness theorem, ansatz-by-citation, or renaming of a known result is present. Whether KL alignment alone is sufficient to make the guided latent decodable under the head-branch query distribution is a correctness/validation question, not a circularity concern. The self-citation is present but non-load-bearing, hence the low non-zero score.

Axiom & Free-Parameter Ledger

6 free parameters · 6 axioms · 1 invented entities

The method is an empirical ML system. It introduces one new model component (learnable tokens) and a loss-level guidance mechanism; the load-bearing assumptions are domain assumptions about data and SMPL-H, plus an unproven Gaussian-latent sufficiency assumption. Five hyperparameters are set by hand, and several architecture hyperparameters are unreported.

free parameters (6)
  • KL balancing weight lambda_KL = 0.0004
    Hand-set in total loss Eq. (6); directly controls strength of latent alignment.
  • Velocity loss weight lambda_vel = 0.003
    Hand-set in Eq. (6); controls temporal smoothness.
  • Shape/contact loss weights lambda_shape, lambda_contact = 0.003 each
    Hand-set in Eq. (4); balance reconstruction terms.
  • Latent dimension of z (size of mu/Sigma) = not reported
    Core architecture hyperparameter not given; affects capacity of the Gaussian pose prior.
  • Learnable token count/dimension = not reported
    Central component hyperparameters not specified.
  • Transformer layers/heads/embedding dim = not reported
    Model capacity choices not disclosed; needed for reproduction.
axioms (6)
  • domain assumption A single sequence-level Gaussian latent is sufficient to represent full-body pose sequences (following TEMOS [24]).
    Invoked in Sec. 3.2; if pose distribution is multimodal or high-dimensional, the guidance objective has limited capacity.
  • domain assumption SMPL-H parametric model and forward kinematics are adequate ground-truth representations of full-body meshes.
    Sec. 3.1; all losses and metrics are computed through this model.
  • domain assumption Head pose representation (relative motion, canonicalized orientation, absolute height) is sufficient to determine body pose in the evaluation distribution.
    Sec. 3.1; the whole task definition assumes this.
  • domain assumption AMASS/RICH/EgoBody datasets and the annotation protocol of EgoAllo [31] are unbiased for cross-dataset evaluation.
    Sec. 4.2; all claims depend on these benchmarks.
  • ad hoc to paper Symmetric KL divergence between the two Gaussian latents is a sufficient objective to transfer generative capability.
    Eq. (1); the paper's core mechanism; only empirically validated by ablations.
  • standard math Reparameterization trick and Gaussian KL closed forms are valid.
    Sec. 3.2; standard background for VAE training.
invented entities (1)
  • Learnable body tokens L_token no independent evidence
    purpose: Appended to head embeddings to represent unobserved body parts in the head-to-motion encoder.
    New model parameters; no external falsifiable handle; only internal ablation evidence.

reviewed 2026-08-01 · how reviews work

0 comments
Cite this review

Pith. "Pith review of EgoGVAE: Ego-body Mesh Reconstruction via Guided Variational Autoencoder." pith.science (2026). https://pith.science/paper/OA6EIZUJ

@misc{pith2026260727755,
  author       = {Pith},
  title        = {Pith review of: EgoGVAE: Ego-body Mesh Reconstruction via Guided Variational Autoencoder},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OA6EIZUJ}},
  note         = {Machine review of arXiv:2607.27755}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

We address the problem of recovering the full-body mesh from only the head pose. This task has become essential for various applications based on head-mounted devices or smart glasses. The challenge of this task lies in estimating the pose information of unobserved body parts based solely on a single joint (i.e., head) trajectory. Several studies have begun to adopt head-conditioned generative models, however, such previous methods are costly and time-consuming due to the diffusion-based iterative process. As an alternative, we propose a simple yet novel method that leverages the latent space of the guidance network, which is designed as a variational autoencoder taking full-body poses as inputs. By enforcing latent distributions of this guidance network and our head-to-motion network to be similar, latent features sampled from the 'guided' distribution, i.e., distribution learned in our head-to-motion network, can be reliably decoded for natural representations of full-body poses even only with the head pose. One important advantage of the proposed method is that one-step sampling scheme achieves remarkably fast inference (more than 50 times faster) compared to diffusion-based approaches. Experimental results on benchmark datasets show that the proposed method efficiently improves the performance of ego-body mesh reconstruction.

Figures

Figures reproduced from arXiv: 2607.27755 by Jaehun Jung, Wonjun Kim.

Figure 1
Figure 1. Figure 1: Overview of the proposed method that estimates full-body poses only from head poses via the guided variational autoencoder, i.e., EgoGVAE. Note that the motion￾to-motion network guides our EgoGAVE in the latent space. In the beginning, to address this challenge, there have been meaningful at￾tempts to utilize the hand position, which can be obtained from motion tracking sensors such as hand controllers and… view at source ↗
Figure 2
Figure 2. Figure 2: The overall architecture of the proposed method for recovering full-body poses from head poses. A motion-to-motion network is used to guide the generation process of the head-to-motion network in the latent space. Note that both tokens for mean and variance are appended to input tokens of each network to formulate latent distributions as Gaussian distributions in a variational framework. 3 Proposed Method … view at source ↗
Figure 3
Figure 3. Figure 3: Visualization for the alignment process of latent distributions. (a) The t￾SNE [21] visualization of latent distributions over training epochs. (b) Results of ego￾body mesh reconstruction at corresponding epochs. (c) Ground Truth. Based on these two terms, the guidance is efficiently conducted in a variational scheme. The effect of the proposed guidance scheme is shown in [PITH_FULL_IMAGE:figures/full_fig… view at source ↗
Figure 4
Figure 4. Figure 4: Results of ego-body mesh reconstruction on the AMASS [22] dataset. (a) Re￾sults by AvatarPoser [13]. (b) Results by EgoPoser [12]. (c) Results by EgoEgo [18]. (d) Results by EgoAllo [31]. (e) Results by the proposed method (i.e., EgoGVAE). (f) Ground Truth. Procrustes-aligned mean per joint position error (PA-MPJPE) [35], grounding metric (Ground) [34], and mean head joint position error (Thead) [31]. Spec… view at source ↗
Figure 5
Figure 5. Figure 5: Results of ego-body mesh reconstruction on the RICH [10] dataset. (a) Results by AvatarPoser [13]. (b) Results by EgoPoser [12]. (c) Results by EgoEgo [18]. (d) Results by EgoAllo [31]. (e) Results by the proposed method (i.e., EgoGVAE). (f) Ground Truth. 4.3 Performance Evaluation Quantitative evaluation. To demonstrate the effectiveness of the proposed method in ego-body mesh reconstruction, we compare o… view at source ↗
Figure 6
Figure 6. Figure 6: (a) Result by removing both guidance and learnable tokens, i.e., baseline. (b) Result by baseline + guidance. (c) Result by baseline + guidance + learnable tokens (our method). (d) Ground Truth. can be seen, diffusion-based approaches perform with high computational costs, however, the proposed method shows a huge advantage in the efficiency of the inference time. Specifically, EgoEgo [18] has a low number… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

36 extracted references · 1 linked inside Pith

  1. [1]

    In: Proc

    Aliakbarian, S., Saleh, F., Collier, D., Cameron, P., Cosker, D.: HMD-NeMo: On- line 3D avatar motion generation from sparse observations. In: Proc. IEEE/CVF Int. Conf. Comput. Vis. pp. 9622–9631 (2023) 4

    2023

  2. [2]

    In: Proc

    Barquero, G., Bertsch, N., Marramreddy, M., Chacón, C., Arcadu, F., Rigual, F., He, N.S., Palmero, C., Escalera, S., Ye, Y., Kips, R.: From sparse signal to smooth motion: Real-time motion generation with rolling prediction models. In: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recog. pp. 1850–1860 (2025) 4

    2025

  3. [3]

    In: Proc

    Chi, S., Huang, P.H., Sachdeva, E., Ma, H., Ramani, K., Lee, K.: Estimating ego- body pose from doubly sparse egocentric video data. In: Proc. Adv. Neural Inform. Process. Syst. vol. 37, pp. 55178–55203 (2024) 4

    2024

  4. [4]

    In: Proc

    Du, Y., Kips, R., Pumarola, A., Starke, S., Thabet, A., Sanakoyeu, A.: Avatars Grow Legs: Generating smooth human motion from sparse tracking inputs with diffusion model. In: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recog. pp. 481– 490 (2023) 4

    2023

  5. [5]

    In: Proc

    Dwivedi, S.K., Sun, Y., Patel, P., Feng, Y., Black, M.J.: TokenHMR: Advancing human mesh recovery with a tokenized pose representation. In: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recog. pp. 1323–1333 (2024) 3

    2024

  6. [6]

    arXiv:2308.13561 (2023) 1, 4

    Engel, J., Somasundaram, K., Goesele, M., Sun, A., Gamino, A., Turner, A., Ta- lattof, A., Yuan, A., Souti, B., Meredith, B., et al.: Project Aria: A new tool for egocentric multi-modal ai research. arXiv:2308.13561 (2023) 1, 4

    Pith/arXiv arXiv 2023

  7. [7]

    In: Proc

    Feng, H., Ma, W., Gao, Q., Zheng, X., Xue, N., Xu, H.: Stratified avatar generation from sparse observations. In: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recog. pp. 153–163 (2024) 4

    2024

  8. [8]

    Flash, T., Hogan, N.: The coordination of arm movements: an experimentally con- firmed mathematical model. J. Neurosci.5(7), 1688–1703 (1985) 10

    1985

  9. [9]

    IEEE Trans

    Gwon, M.G., Um, G.M., Cheong, W.S., Kim, W.: Eigenpose: Occlusion-robust 3D human mesh reconstruction. IEEE Trans. Image Process.34, 2379–2391 (2025) 4

    2025

  10. [10]

    In: Proc

    Huang, C.H.P., Yi, H., Höschle, M., Safroshkin, M., Alexiadis, T., Polikovsky, S., Scharstein, D., Black, M.J.: Capturing and inferring dense full-body human-scene contact. In: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recog. pp. 13274–13285 (2022) 9, 11, 12

    2022

  11. [11]

    IEEE Trans

    Ionescu, C., Papava, D., Olaru, V., Sminchisescu, C.: Human3.6M: Large scale datasets and predictive methods for 3D human sensing in natural environments. IEEE Trans. Pattern Anal. Mach. Intell.36(7), 1325–1339 (2013) 9

    2013

  12. [12]

    In: Proc

    Jiang, J., Streli, P., Meier, M., Holz, C.: EgoPoser: Robust real-time egocentric pose estimation from sparse and intermittent observations everywhere. In: Proc. Eur. Conf. Comput. Vis. pp. 277–294 (2024) 4, 8, 9, 10, 11, 12, 15

    2024

  13. [13]

    In: Proc

    Jiang, J., Streli, P., Qiu, H., Fender, A., Laich, L., Snape, P., Holz, C.: AvatarPoser: Articulated full-body pose tracking from sparse motion sensing. In: Proc. Eur. Conf. Comput. Vis. pp. 443–460 (2022) 2, 4, 8, 9, 10, 11, 12, 15

    2022

  14. [14]

    In: Proc

    Kanazawa, A., Black, M.J., Jacobs, D.W., Malik, J.: End-to-end recovery of human shape and pose. In: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recog. pp. 7122– 7131 (2018) 3

    2018

  15. [15]

    In: Proc

    Kingma, D.P., Welling, M.: Auto-encoding variational bayes. In: Proc. Int. Conf. Learn. Represent. (2014) 7

    2014

  16. [16]

    In: Proc

    Kocabas, M., Huang, C.H.P., Hilliges, O., Black, M.J.: PARE: Part attention re- gressor for 3D human body estimation. In: Proc. IEEE/CVF Int. Conf. Comput. Vis. pp. 11127–11137 (2021) 3 EgoGVAE 17

    2021

  17. [17]

    In: Proc

    Kolotouros, N., Pavlakos, G., Black, M.J., Daniilidis, K.: Learning to reconstruct 3D human pose and shape via model-fitting in the loop. In: Proc. IEEE/CVF Int. Conf. Comput. Vis. pp. 2252–2261 (2019) 3

    2019

  18. [18]

    In: Proc

    Li, J., Liu, K., Wu, J.: Ego-body pose estimation via ego-head pose estimation. In: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recog. pp. 17142–17151 (2023) 2, 4, 9, 10, 11, 12, 13, 14, 15

    2023

  19. [19]

    In: ACM Trans

    Loper, M., Mahmood, N., Romero, J., Pons-Moll, G., Black, M.J.: SMPL: A skinned multi-person linear model. In: ACM Trans. Graph. pp. 851–866 (2023) 3, 5, 8

    2023

  20. [20]

    In: Proc

    Loshchilov, I., Hutter, F.: Decoupled weight decay regularization. In: Proc. Int. Conf. Learn. Represent. (2019) 9

    2019

  21. [21]

    Maaten, L.v.d., Hinton, G.: Visualizing data using t-SNE. J. Mach. Learn. Res.9, 2579–2605 (2008) 7

    2008

  22. [22]

    In: Proc

    Mahmood, N., Ghorbani, N., Troje, N.F., Pons-Moll, G., Black, M.J.: AMASS: Archive of motion capture as surface shapes. In: Proc. IEEE/CVF Int. Conf. Com- put. Vis. pp. 5442–5451 (2019) 9, 10, 11, 12, 13, 14, 15

    2019

  23. [23]

    In: Proc

    Paszke, A., Gross, S., Chintala, S., Chanan, G., Yang, E., DeVito, Z., Lin, Z., Desmaison, A., Antiga, L., Lerer, A.: Automatic differentiation in pytorch. In: Proc. Adv. Neural Inform. Process. Syst. pp. 1–4 (2017) 9

    2017

  24. [24]

    In: Proc

    Petrovich, M., Black, M.J., Varol, G.: TEMOS: Generating diverse human motions from textual descriptions. In: Proc. Eur. Conf. Comput. Vis. pp. 480–497 (2022) 6

    2022

  25. [25]

    In: Proc

    Shin, S., Kim, J., Halilaj, E., Black, M.J.: WHAM: Reconstructing world-grounded humans with accurate 3D motion. In: Proc. IEEE/CVF Conf. Comput. Vis. Pat- tern Recog. pp. 2070–2080 (2024) 4

    2070

  26. [26]

    In: Proc

    Sun, Y., Bao, Q., Liu, W., Fu, Y., Black, M.J., Mei, T.: Monocular, one-stage, regression of multiple 3D people. In: Proc. IEEE/CVF Int. Conf. Comput. Vis. pp. 11179–11188 (2021) 3

    2021

  27. [27]

    In: Proc

    Sun, Y., Liu, W., Bao, Q., Fu, Y., Mei, T., Black, M.J.: Putting people in their place: Monocular regression of 3D people in depth. In: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recog. pp. 13243–13252 (2022) 3

    2022

  28. [28]

    In: Proc

    Vaswani,A.,Shazeer,N.,Parmar,N.,Uszkoreit,J.,Jones,L.,Gomez,A.N.,Kaiser, Ł., Polosukhin, I.: Attention is all you need. In: Proc. Adv. Neural Inform. Process. Syst. vol. 30 (2017) 7

    2017

  29. [29]

    In: Proc

    Wang, Y., Sun, Y., Patel, P., Daniilidis, K., Black, M.J., Kocabas, M.: PromptHMR: Promptable human mesh recovery. In: Proc. IEEE/CVF Conf. Com- put. Vis. Pattern Recog. pp. 1148–1159 (2025) 4

    2025

  30. [30]

    In: Proc

    Ye, V., Pavlakos, G., Malik, J., Kanazawa, A.: Decoupling human and camera motion from videos in the wild. In: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recog. pp. 21222–21232 (2023) 4

    2023

  31. [31]

    In: Proc

    Yi, B., Ye, V., Zheng, M., Li, Y., Müller, L., Pavlakos, G., Ma, Y., Malik, J., Kanazawa, A.: Estimating body and hand motion in an ego-sensed world. In: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recog. pp. 7072–7084 (2025) 2, 4, 5, 6, 9, 10, 11, 12, 13, 14, 15

    2025

  32. [32]

    In: Proc

    Zhang, S., Ma, Q., Zhang, Y., Qian, Z., Kwon, T., Pollefeys, M., Bogo, F., Tang, S.: EgoBody: Human body shapeand motion ofinteracting people from head-mounted devices. In: Proc. Eur. Conf. Comput. Vis. pp. 180–200 (2022) 12

    2022

  33. [33]

    In: Proc

    Zhang, Y., Zhang, H., Hu, L., Zhang, J., Yi, H., Zhang, S., Liu, Y.: ProxyCap: Real-time monocular full-body capture in world space via human-centric proxy- to-motion learning. In: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recog. pp. 1954–1964 (2024) 8 18 J. Jung and W. Kim

    1954

  34. [34]

    In: Proc

    Zheng, X., Su, Z., Wen, C., Xue, Z., Jin, X.: Realistic full-body tracking from sparse observations via joint-level modeling. In: Proc. IEEE/CVF Int. Conf. Comput. Vis. pp. 14678–14688 (2023) 2, 4, 8, 10

    2023

  35. [35]

    IEEE Trans

    Zhou, X., Zhu, M., Pavlakos, G., Leonardos, S., Derpanis, K.G., Daniilidis, K.: MonoCap: Monocular human motion capture using a cnn coupled with a geometric prior. IEEE Trans. Pattern Anal. Mach. Intell.41(4), 901–914 (2018) 10

    2018

  36. [36]

    In: Proc

    Zhou, Y., Barnes, C., Lu, J., Yang, J., Li, H.: On the continuity of rotation repre- sentations in neural networks. In: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recog. pp. 5745–5753 (2019) 5

    2019

This paper was first reviewed by deepseek-v4-flash on August 1, 2026.