acceptodds
Withdrawn from ICLR 2027

A Diagnostic Study of World-Space Human Recovery Under High Camera Dynamics

Withdrawn from ICLR 2027, so left out of the paper lists and the map.

Abstract

Monocular world-space human recovery methods excel on current benchmarks, but these rarely include fast camera motion. We introduce the K-Pop stress test, broadcast dance videos that combine fast, large camera motion with dense multi-person choreography. Leading methods produce plausible per-frame meshes on these videos, yet their world-space trajectories drift far from the performers' stage positions. Through a systematic diagnosis, we show that the camera and human estimates each remain self-consistent, and that the drift appears when they are combined at mismatched scales. Each is recovered at its own scale, so a mismatch between the two can turn camera motion into human drift that grows with camera displacement and becomes severe in broadcast dance videos. We derive a closed-form correction that estimates this mismatch and applies one scale to all performers, removing the drift while keeping their relative positions in one shared world frame. It requires no retraining and works with existing backends. Combined with physical constraints, the correction yields stable world-space reconstruction for choreography modeling, scene editing, data generation, and virtual reshooting.

Primary area:
applications to computer vision, audio, language, and other modalities
Keywords:
computer vision, human mesh recovery, closed form, training free

Reviews

Among:
Shape:
1234this paper, Bottom 2%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.33 over 3 reviews, above 2% of the others.
Reviewer_2NKHOpenReview ↗
Rating
1
Confidence
4
Informativeness
1
Presentation
2
Soundness
1
Read review
Paper summary

This paper investigates the failure of compositional world-space human mesh recovery (HMR) methods under high camera dynamics, using a curated set of K-Pop broadcast dance videos as a “K-Pop stress test.” The authors attribute severe world-space trajectory drift to a single, constant sequence-level scalar scale mismatch between independently estimated visual odometry (VO) translations and camera-space HMR translations. They propose a closed-form weighted least-squares estimator for this scale, fitted over detected stationary foot-contact intervals, and integrate it into a multi-stage pipeline that includes Physically Grounded Optimization (PGO).

The application scenario is relevant, and the integrated system produces visually coherent multi-person trajectories. However, the central diagnostic claim is not adequately established. The constant-scale assumption is questionable under broadcast cameras with continuously changing focal lengths; clips that fail the VO time-reversal shuttle test are discarded, undermining the claim that VO remains self-consistent under high camera dynamics; and the paired-camera evaluation uses monocular PromptHMR predictions as pseudo-ground truth. The evaluation also relies heavily on 2D reprojection and smoothness/contact metrics that do not establish world-space accuracy and are directly related to terms optimized by PGO. On EMDB-2, the proposed method degrades accuracy relative to the PromptHMR baseline. Finally, the extensive preprocessing and 3,000-iteration PGO stage make it difficult to isolate the contribution of the claimed closed-form scale estimator, while the reported ablations suggest that PGO can absorb much of the estimator’s contribution.

Critical strengths
  1. Practical Problem Scenario: Highlighting the vulnerability of compositional HMR pipelines under aggressive camera movements (e.g., dolly zooms and rapid panning) and dense multi-person choreography is relevant for in-the-wild media applications.

  2. Visually Coherent Qualitative Outputs: The integrated system produces visually pleasing multi-person trajectories and reduces inter-person collisions in complex dance sequences.

Critical weaknesses
  1. The core diagnostic claim and evaluation do not establish reliable world-space accuracy. The proposed single, constant sequence-level scale mismatch is theoretically questionable for broadcast videos with continuous optical zooming and time-varying perspective effects. The paper also discards clips that fail the VO time-reversal shuttle test, uses monocular PromptHMR predictions as pseudo-ground truth in the paired-camera reference set, and evaluates the large K-Pop set mainly with proxy metrics that cannot establish world-space accuracy and are directly affected by the optimization objectives. On EMDB-2, the proposed method degrades W-MPJPE from 388.66 mm to 415.03 mm and WA-MPJPE from 139.65 mm to 158.50 mm.

  2. The contribution of the closed-form scale estimator is not isolated from preprocessing and PGO. The pipeline combines ViPE, custom dynamic-foreground prompts, filtering, SAM2.1/YOLO11x tracking, RANSAC floor fitting, heuristic gap filling and contact blending, and 3,000 iterations of PGO that itself includes a learned residual scale. Appendix ablations suggest that PGO can absorb substantial differences in the closed-form initialization and can improve trajectory smoothness even without the proposed closed-form correction. The main baseline comparisons do not apply the same preprocessing and PGO, making the contribution of the proposed estimator difficult to determine.

Questions for authors
  1. Unsubstantiated Core Claim and Oversimplified Problem Attribution (Scalar Scale Mismatch). The central claim is that world-space drift under high camera dynamics can be explained by a single, constant sequence-level scalar scale mismatch, ΔTij′=sΔTij\Delta T'_{ij}=s\Delta T_{ij} (Eq. 4). This modeling is theoretically questionable for broadcast K-Pop videos, where continuous optical zooming (time-varying focal length ftf_t) and nonlinear perspective distortions are common. As the authors acknowledge in Appendix B (line 962) and D.3, focal-length error directly shifts camera-space depth estimation proportionally. When a camera zooms continuously during a shot, the effective scale bias between HMR depth—which assumes a fixed per-clip focal length in Step 2—and VO translation may be time-varying (sts_t) and nonlinear rather than a constant scalar. How do the authors justify modeling a potentially dynamic focal-length and perspective-distortion problem as a static scalar mismatch?

  2. Contradiction in Time-Varying Scale Analysis (Table 7). In Table 7, the authors justify using a constant scale by showing that a time-varying scale severely degrades Jitter (26.87 to 131.57 mm) and Foot Skating (17.94 to 32.58 mm). However, this does not establish that the true physical mismatch is constant; it may instead indicate that the stationary-interval estimator is too ill-conditioned and noisy to estimate local scales without injecting discontinuous step changes into the camera trajectory. How can the authors conclude that the true error is a constant scalar when broadcast cameras actively zoom within shots?

  3. Selection Bias from Pre-Filtering Clips with the Shuttle Test. The paper claims in Section 1 and Section 5 that “camera estimation remains self-consistent under high camera dynamics, so the failure must lie in composition.” However, Section 4 (line 199) and Appendix F reveal that all K-Pop clips that failed the ViPE time-reversal shuttle test were explicitly discarded. The dataset is therefore filtered to retain clips where VO succeeds before the paper concludes that VO does not fail under high camera dynamics. How do the authors rule out this survivorship bias, and how many clips were excluded by the shuttle test?

  4. Flawed “Ground Truth” in the Paired-Camera Reference Set (Table 1). The paper evaluates metric accuracy (WA-MPJPE, W-MPJPE, and Scale Ratio) on 11 video pairs (989 frames) using PromptHMR predictions from a fixed venue camera as “Ground Truth” (Section 4, line 206). PromptHMR is a monocular estimation method, not a calibrated multi-view motion-capture or LiDAR system. On wide-angle venue shots where 6–12 performers are far from the camera, monocular HMR can suffer from severe depth ambiguity, scale bias, and inter-person depth-ordering errors. Using one monocular model’s predictions as ground truth to benchmark other models—and claiming superiority over PromptHMR itself—is scientifically questionable. Can the authors validate these results using independent calibrated multi-view or sensor-based ground truth?

  5. Circular Reasoning in Stationary-Interval Detection. In Section 6 (line 365), the authors detect stationary performers for estimating ss using a directional gate that rejects intervals if ∣cos⁡(ΔW^p,ij,ΔTij)∣≤0.8\left|\cos(\Delta\hat{W}_{p,ij},\Delta T_{ij})\right|\leq 0.8, arguing that “a stationary performer drifts only along the camera’s path (Eq. 4).” This assumes Eq. 4 to filter the data and then uses the filtered data to support Eq. 4 and fit ss. In dance choreography, performers may take small steps or sway parallel to the camera axis, passing both the partial foot-contact threshold (α=0.8\alpha=0.8) and the cosine gate and thereby corrupting s∗s^*. How do the authors quantify or rule out this source of bias?

  6. Severe Entanglement of the Closed-Form Scale Correction with PGO and Preprocessing. The paper presents the closed-form scale correction (Eq. 6) as its core technical contribution, but Appendices B and C describe a heavily engineered six-step pipeline: ViPE with custom dynamic-foreground text prompts, One Euro filtering, two-pass SAM2.1/YOLO11x tracking, RANSAC floor fitting, and 3,000 iterations of Adam-based PGO with seven loss terms (Lkp,Lsm,Lrep,Lfloor,Lcv,Lch,Lsf\mathcal{L}_{\text{kp}}, \mathcal{L}_{\text{sm}}, \mathcal{L}_{\text{rep}}, \mathcal{L}_{\text{floor}}, \mathcal{L}_{\text{cv}}, \mathcal{L}_{\text{ch}}, \mathcal{L}_{\text{sf}}). PGO also includes a learned residual scale Δs\Delta s. How much of the reported improvement is attributable to the closed-form estimator rather than the preprocessing and PGO stages?

  7. Appendix Ablations Suggest That the Closed-Form Estimator May Be Largely Redundant. Appendix Tables 6 and 9 raise concerns about the necessity of the closed-form estimator.

    • In Table 6, reducing the pooled stationary intervals from 125 to 3 changes the closed-form estimate a∗a^* from 0.310 to 0.443, yet PGO absorbs almost the entire difference through its learned Δa\Delta a, producing nearly identical final scales (0.318 versus 0.320). If PGO’s learned Δs\Delta s can recover essentially the same scale regardless of the closed-form initialization, what is the actual contribution of the closed-form estimator?
    • In Table 9, on stationary clips, raw displacement is 2.334 m. PGO alone, without closed-form drift correction, reduces displacement by 75% to 0.581 m and reduces Jitter from 105.1 mm to 12.46 mm. Adding the closed-form drift correction leaves Jitter essentially unchanged at 12.38 mm.
    • In the main text (Tables 2 and 3), “Ours” is compared against PromptHMR, TRAM, and GVHMR without applying the authors’ PGO, SAM2.1 tracking, or ViPE preprocessing to the baselines. Please disentangle (a) backbone replacement (ViPE + SAM2.1), (b) closed-form s∗s^* alone without PGO, (c) PGO alone with learned Δs\Delta s initialized at zero and without closed-form s∗s^*, and (d) closed-form s∗s^* + PGO. These controlled comparisons should be reported consistently across the relevant tables.
  8. Questionable Validity and Value of Proxy Metrics (Tables 2 and 3). Evaluation on the 616 K-Pop clips relies entirely on 2D reprojection (PCK against YOLO11n-Pose and Mask IoU) and smoothness heuristics (Jitter and Foot Skating).

    • As the authors acknowledge, 2D reprojection is blind to world-space scale errors: TRAM and HAC achieve higher PCK and Mask IoU than “Ours” in Table 3 despite severe world-space drift.
    • In Appendix C, PGO explicitly optimizes 3D joints against ViTPose 2D keypoints through Lkp\mathcal{L}_{\text{kp}}, and directly penalizes acceleration/velocity through Lsm\mathcal{L}_{\text{sm}} (weight 1) and foot-contact velocity/stickiness through Lcv\mathcal{L}_{\text{cv}} (weight 500) and Lsf\mathcal{L}_{\text{sf}} (weight 100). Evaluating the method with Jitter and Foot Skating when these exact finite-difference terms are heavily penalized during test-time optimization risks being trivially gamed: setting world velocity to zero would achieve zero Jitter and Foot Skating. Thus, Tables 2 and 3 do not establish world-space trajectory accuracy. What independent metric or ground truth can validate the recovered world-space trajectories?
  9. Performance Degradation on the Standard Ground-Truth Benchmark (EMDB-2). On the only benchmark with real sensor ground truth evaluated in the paper (EMDB-2, Section 7.1, lines 454–459), the proposed method degrades world-space trajectory accuracy relative to PromptHMR: W-MPJPE worsens from 388.66 mm to 415.03 mm (+26.4 mm error), and WA-MPJPE worsens from 139.65 mm to 158.50 mm (+18.9 mm error). Furthermore, the PromptHMR reproduction on EMDB-2 reportedly obtains 388.66 mm W-MPJPE and 139.65 mm WA-MPJPE at line 456, whereas Appendix Table 4 reports PromptHMR-Vid achieving 216.5 mm W-MPJPE and 71.0 mm WA-MPJPE on EMDB-2. Please explain this substantial discrepancy and why a general drift-correction method harms accuracy on data with real ground truth.

  10. Missing Evaluation on Existing Multi-Person, High-Dynamic Ground-Truth Datasets (WorldPose/DCM). Table 5 lists WorldPose, which contains in-the-wild moving and zooming camera motion with LiDAR multi-camera ground truth for multiple people, and DCM, which contains moving and zooming dance videos with ground truth. Why is the method not evaluated on WorldPose or DCM instead of relying on 11 K-Pop clips with pseudo-ground truth? Alternatively, can the authors evaluate on synthetic high-camera-dynamic sequences, for example by rendering BEDLAM/AMASS dances with aggressive virtual broadcast-camera trajectories and focal zooms, where exact 3D ground truth is available?

  11. Ad Hoc Heuristics in Gap Filling and Contact Blending. Appendix B (lines 1014–1018) states that missing detections are handled by linearly interpolating 3D translations, damping boundary velocities by 0.8 per frame, and artificially forcing contact logits toward 1.0 after 15 frames so that PGO plants the feet. How much of the stability reported in Table 3 comes from these hand-crafted track-filling heuristics rather than the proposed scale correction?

Reviewer_mVuZOpenReview ↗
Rating
2
Confidence
3
Informativeness
2
Presentation
2
Soundness
2
Read review
Paper summary

The paper studies world-space human recovery in crowded scenes with highly dynamic cameras, a setting underrepresented in standard HMR benchmarks. The authors observe drift in the world-space trajectories produced by existing pipelines and attribute it to a scale mismatch between independently estimated camera motion and camera-space human positions. They estimate a single sequence-level scale correction by solving a weighted least-squares problem over stationary foot-contact intervals pooled across performers. On the proposed K-Pop stress test, this simple correction, combined with physically grounded optimization, improves trajectory stability and reprojection consistency.

Critical strengths

The paper highlights a relevant failure mode of world-space HMR under large camera motion, a setting underrepresented in standard benchmarks. The scale-mismatch explanation is geometrically well motivated: independently estimated camera and human translations can appear plausible on their own, yet introduce camera-correlated drift when combined at inconsistent scales. Larger camera displacement amplifies this error, explaining why it becomes particularly visible in broadcast footage. The K-Pop stress test usefully exposes this weakness in existing pipelines.

Critical weaknesses

The experimental evaluation does not clearly isolate the contribution of the closed-form scale estimator. Although the LS scale estimator is presented as a major contribution, the pipeline also includes substantial physically grounded optimization (PGO), which directly penalizes motion instability and foot sliding and further optimizes the scale. Figure 8 supports the benefit of adding drift correction within the PGO pipeline, but does not establish the effectiveness of the LS scale correction alone. Table 2 does not clearly specify whether “+ drift correction” includes PGO, and no LS-only ablation is reported.

Crucially, it is unclear whether the residual scale remains optimizable in the “PGO without drift correction” ablation. This determines whether the comparison isolates the benefit of LS initialization or also changes the optimization variables. Matched comparisons between LS-only, PGO with freely optimized scale initialized at unity, and LS-initialized PGO are needed. Without these controls, the evidence does not adequately establish how much of the reported improvement is attributable to the paper’s central methodological contribution.

Questions for authors
  1. Does “+ drift correction” in Table 2 mean LS scale only or LS scale + PGO?

  2. Is scale still optimized in “PGO without drift correction”? Please report LS-only results and compare unity, HAC, and LS initializations with the same PGO.

  3. Does LS correction still help with a modern HMR backbone, such as SAM 3D Body (CVPR '26) + DA3, under the same PGO settings?

Reviewer_DLdQOpenReview ↗
Rating
1
Confidence
4
Informativeness
1
Presentation
3
Soundness
1
Read review
Paper summary

The paper claims that

  • it introduces the K-Pop stress test, broadcast dance videos with high camera dynamics and dense multi-person choreography;
  • compositional methods fail in this regime because the camera and human translations are estimated at mismatched scales;
  • a closed-form correction removes this mismatch across backbones without retraining.

What the paper studies

The paper presents itself as a diagnostic study, not as a paper that proposes a new problem.

  • Object: compositional world-space human recovery (e.g., TRAM, PromptHMR-Vid), which composes visual odometry with camera-space HMR.
  • Benchmark: the K-Pop stress test, 616 monocular broadcast dance clips with fast camera motion, zoom, and up to 12 performers.
    • It is proposed as an evaluation benchmark, but it has no 3D annotation (Table 5).
    • Its metrics are not convincing either. The 2D metrics compare against detector outputs. The smoothness metrics (jitter, foot skating) have no reference. The 3D metrics cover only 11 paired clips and use a PromptHMR reconstruction as the reference.

The pipeline actually consists of

  • Off-the-shelf, frozen modules:
    • visual odometry (ViPE) for the camera poses;
    • camera-space HMR (PromptHMR-Vid) for the human meshes, root translations, and foot-contact confidence;
    • tracking, 2D-keypoint, and gravity estimators.
  • TRAM-style composition of the two outputs through a rigid transform (Eq. 1).
  • Two non-learned post-processing steps:
    1. a closed-form least-squares estimate of one global scale per sequence, fitted on intervals where performers are judged stationary and used to rescale the camera translation (Eqs. 6–7);
    2. a loss-based optimization (PGO) over reprojection, smoothness, contact, floor, and inter-person separation.

The paper is neither convincing nor sound because:

  1. The starting point is not justified. The paper takes the compositional paradigm as the object of the diagnosis. It dismisses the paradigm that predicts world-space motion directly (e.g., GVHMR) in one sentence. Figs. 1 and 2 motivate the problem only with PromptHMR, a compositional pipeline whose failure in this setting follows from the paper's own analysis.
  2. Reprojection is not justified as a target, yet it carries the argument.
    • The direct-prediction paradigm gives up reprojection for locally plausible world-space motion. The paper does not discuss whether this trade-off is wrong.
    • Nor does it argue that any of its listed applications needs both reprojection and coherent 3D. Scene editing, for instance, needs only meshes that reproject, with consistent identities. Virtual reshooting and choreography modeling need coherent 3D motion that can drive avatars.
    • Still, reprojection serves as the evidence of failure (Figs. 1 and 2) and as half of the evaluation. The paper itself, however, states that the composition preserves reprojection exactly and that reprojection cannot reveal a wrong scale.
  3. The problem is reduced rather than understood. The failure is attributed to a camera–human scale mismatch by elimination. By the paper's own account, the two checks behind this elimination cannot see an error shared across a clip. The problem then ends up as a single scalar.
  4. Consequently, the experimental setting does not support the claims. It inherits the starting point, since the baselines are compositional methods or methods adapted to the composition (GVHMR). Without 3D annotation, the evaluation relies on 2D and smoothness metrics. Neither kind can reveal a wrong scale. The gain over HAC is not separated from PGO, which HAC does not use. On EMDB-2, the only data with 3D ground truth, the method is worse than its reproduction of PromptHMR on the world-space errors.

The essential question, assuming that multi-person tracking is solved, is whether the camera and the humans can both be reconstructed accurately in world space under high camera dynamics. Existing methods answer it with one of two compromises:

  • Compositional methods keep reprojection under the estimated camera and give up global and temporal consistency. The resulting drift is what the paper studies.
  • The direct-prediction paradigm (e.g., GVHMR) keeps global and temporal consistency and gives up reprojection.

The title presents the work as a diagnostic study of world-space human recovery, but it studies only the first compromise. The trade-off itself appears only in passing, in captions (e.g., Fig. 7 and Table 2). The discussion does not make it clear, and no experiment examines it directly. Two questions therefore remain open: whether the compromise is necessary at all, and what the chosen compromise means in practice.

See "weaknesses" for details.

Critical strengths

The pipeline is clearly described in terms of data flow.

Critical weaknesses

1. The starting point and the role of reprojection are not justified

The paper analyzes only one of the two paradigms, and it does not explain why reprojection should be a target.

In this review, I use the two paradigms named in the paper's introduction:

  • Compositional paradigm: visual odometry estimates the camera, camera-space HMR estimates the humans, and a rigid transform combines the two (e.g., TRAM, PromptHMR-Vid).

  • Direct-prediction paradigm: world-space motion is predicted directly from human pose and motion (e.g., GLAMR, TRACE, WHAM, GVHMR).

  • The direct-prediction paradigm is set aside without analysis.

    • The paper discusses it in only one sentence (Sec. 2): "this local-to-global mapping is inherently ambiguous when motion priors do not match the scene, for example a subject dancing in place." A subject dancing in place is exactly the K-Pop setting. However, no experiment tests whether this paradigm fails there.
    • GVHMR appears in the experiments only in modified forms. In the main comparison, its trajectory is aligned once per track to a separate visual odometry (DPVO). Its low 2D scores may therefore come from this modification rather than from the method. The drift attributed to GVHMR (App. D.3) appears only after the authors combine its camera-space output with visual odometry themselves (App. B).
    • In my view, for performers on a stage, such a method followed by ground-plane estimation and physics-based simulation, as is common practice, can already give a plausible world-space structure. The paper does not test this.
    • The joint-optimization methods (e.g., SLAHMR, PACE) are cited in the same paragraph, but they are not compared either.
  • Reprojection is not justified as a target.

    • The direct-prediction paradigm gives up reprojection to keep the motion locally plausible and consistent over space and time. The paper does not say what is lost by this.
    • By the paper's own statements, good reprojection is neither necessary nor sufficient for its core problem. The composition introduces no reprojection error (Sec. 5), and "Reprojection cannot reveal a wrong scale" (Sec. 7).
    • The paper names choreography modeling, scene editing, data generation, and virtual reshooting as applications. It does not argue that any of them needs both reprojection and coherent 3D. For example, scene editing needs only meshes that reproject, with consistent identities. Choreography modeling and virtual reshooting need coherent 3D motion that can drive avatars.
  • Besides, Figs. 1 and 2 do not motivate the problem. These two figures carry the motivation of the introduction, and both use only PromptHMR as the failing method. PromptHMR is itself compositional. By the paper's own analysis (Eq. 4), the drift of a compositional pipeline grows with camera displacement, so its failure under high camera dynamics is expected by construction. Fig. 2 nevertheless presents it as "World-space recovery that succeeds on existing benchmarks fails under high camera dynamics." The figures show a known weakness of one paradigm, not a failure of world-space human recovery in general. I consider this a critical problem in the logic of the introduction.

The compositional paradigm may be a reasonable choice on existing benchmarks, where these methods are strong (Table 4). Under high camera dynamics, however, the paper does not justify why it starts from this paradigm.

2. The diagnosis reduces the problem, and the evaluation cannot support the claims

The failure is reduced to a single scalar by elimination, and the benchmark cannot verify the 3D result.

  • The diagnosis relies on elimination, but its two checks cannot detect a clip-wide error.
    • HMR is checked on K-Pop only with qualitative overlays. Visual odometry is checked with a time-reversal test, and the clips themselves are selected by passing that same test (Sec. 4).
    • The paper itself states that "neither test can see an error shared across a whole clip" (Sec. 5). The conclusion that the error lies in the scale of the composition therefore follows from elimination, not from a measurement.
    • The problem then becomes one scale per sequence. Errors that change between frames are treated as residuals (App. D.3), and the paper acknowledges that a wrong depth for an individual performer cannot be repaired (Fig. 9).
  • Zoom is not modeled. The pipeline uses a single focal length per clip, even on zoom-heavy sequences (App. B). The paper itself shows that a focal-length error acts as a scale error (Fig. 13). During a zoom, this error changes over time, so one scale per sequence cannot represent it. The benchmark also cannot separate it from the scale mismatch that the paper diagnoses.
  • The benchmark has no 3D annotation. I note that this is related to Weakness 1: without 3D annotation, the evaluation falls back on reprojection, whose role is questioned there.
    • On the 616 clips, the 2D metrics compare against detector outputs. Since the composition preserves reprojection, TRAM, which only applies a rigid transform, obtains the highest PCK (Table 3).
    • Jitter and foot skating have no reference. A static mesh on the floor would score well on both.
    • 3D metrics exist only for 11 clips (Table 1). Their reference is a PromptHMR reconstruction from a fixed venue camera, without any described calibration, and from the same model family as the default backbone. No validation of this reference against known scene dimensions or other independent measurements is reported.
    • On EMDB-2, the only data with 3D ground truth, the method is worse than its reproduction of PromptHMR on both world-space errors (Sec. 7).
  • The contribution of the closed-form scale is not isolated. The HAC baseline is applied without PGO (App. B), while the proposed method always includes it. The paper's own ablation shows that PGO absorbs most of the change in the closed-form estimate (Table 6). It therefore remains unclear whether the gains over HAC come from the closed-form scale or from PGO.

Taken together, the paper claims a diagnosis of world-space human recovery, but it diagnoses one paradigm, mostly with its own composition, and without data that can verify a 3D result. It therefore does not answer whether the camera and the humans can both be reconstructed accurately under high camera dynamics.

Questions for authors

(Clarification) How was the absolute scale of the PromptHMR reference in Table 1 validated, for example against known scene dimensions or other independent measurements?

Beyond this one I have not listed detailed questions about the method, because my main concerns are the starting point of the diagnosis, the role of reprojection, and the evaluation without 3D annotation. If these concerns can be addressed, I will raise more specific questions about the method during the discussion.

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Discussion (0)

Sign in to comment.