Pith. sign in

REVIEW 4 major objections 5 minor 65 references

DCHM: Depth-Consistent Human Modeling for Multiview Detection

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read DCHM claims that a label-free pipeline built on superpixel-wise Gaussian splatting and refined monocular depth can detect pedestrians in sparse-view crowded scenes with MODA 84.2 on Wildtrack, 80.1 on Terrace, and 78.4 on MultiviewX.

desk verdict Plausible and useful label-free multiview detection pipeline, but the Wildtrack train/test split ambiguity undercuts the headline numbers until resolved. read the letter →

arxiv 2507.14505 v1 pith:V35ITIBW submitted 2025-07-19 cs.CV

classification cs.CV
keywords multiviewpedestriandetectionlabel-freedepthconsistencyGaussiansplattingpseudo-depthlearningsuperpixelinitializationsegmentationlocalization
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

Multiview pedestrian detection needs to fuse sparse camera views into one consistent 3D picture, and this paper argues that the bottleneck is depth consistency, not annotations. DCHM is a fully label-free pipeline: it initializes Gaussian splatting with per-pedestrian superpixels, optimizes it against segmentation masks to obtain pseudo-depth, filters those depths across views, fine-tunes a monocular depth network on them, and repeats the loop. The authors claim this yields point clouds precise enough that simple clustering recovers pedestrians, reaching MODA 84.2 on Wildtrack, 80.1 on Terrace, and 78.4 on MultiviewX without any 3D labels. They also claim it is the first method to reconstruct pedestrians and perform multiview segmentation in sparse-view, crowded scenes. If correct, the contribution matters because it removes the costly 3D annotation requirement and makes detection geometry accurate enough to later feed supervised localization, reaching 95.5 MODA on Wildtrack.

What carries the argument

The load-bearing machinery is superpixel-wise Gaussian Splatting used as a pseudo-depth generator. Gaussian splatting normally needs good 3D initialization and densely overlapping views, so the paper replaces Structure-from-Motion with a uniform sampling plus filtering initialization: rays through superpixel centers are sampled uniformly, with Gaussian scale fixed by superpixel area and ray distance. A combined loss enforces superpixel-wise photometric agreement, mask agreement, opacity near zero or one, and small depth variance inside each pedestrian mask. Two geometric filters then decide which rendered depths become labels: cross-view foreground filtering requires reprojection into the foreground in every reference view, and cross-view depth consistency filtering requires agreement with at least one other view's rendered depth. A second machinery piece is multiview label matching, which propagates Gaussian IDs through masks view by view using blending weights, so the same pedestrian gets one ID across cameras and its Gaussians can be clustered.

What would settle it

Run the pipeline on a scene with ground-truth depth and masks, then measure the precision of the retained pseudo-depth labels: if many labels surviving cross-view filtering exceed a small depth-error tolerance, or if randomly dropping 20% of per-view masks leaves MODA unchanged, the claimed dependence of the geometry on complete masks would be falsified.

Watch

Extended reading notes

Core claim

The central discovery is that cross-view depth consistency can be self-learned from segmentation masks alone, without 3D annotations, and that this consistency is what makes multiview human modeling usable. Concretely, for each pedestrian the method builds K superpixels, initializes a cloud of Gaussians by ray-sampling through superpixel centers, and optimizes with a loss combining superpixel photometric error, mask overlap, opacity regularization, and within-mask depth flatness. Because Gaussian splatting needs multiview support, only pixels that reproject into the foreground in every view and whose depth agrees with another view's rendered depth are kept as pseudo-depth labels; those labels fine-tune a monocular depth network, and the network's improved depth reinitializes the next splatting round. At inference, per-view masks are matched across views by assigning shared Gaussian IDs, and DBSCAN clustering of each ID's Gaussians gives detections. The claimed result is state-of-the-art label-free detection with higher localization precision, plus the first 3D pedestrian reconstruction and multiview segmentation under sparse, crowded, wide-baseline conditions.

Load-bearing premise

The load-bearing premise is that the per-view pedestrian masks produced by YOLOv11 are complete and accurate enough to supervise every downstream stage; mask misses or over-segmentation in the many pedestrians seen by only one to three views would directly corrupt which Gaussians survive, which IDs get assigned, and whether neighboring people merge.

Editorial extensions

If this is right

  • DCHM reports the best label-free results on the three benchmarks: MODA 84.2 on Wildtrack, 80.1 on Terrace, and 78.4 on MultiviewX, with Wildtrack MODP of 80.3 versus 61.2 for the previous label-free baseline UMPD.
  • When the same human modeling is wrapped with the supervised localization head used in reference [31], it reaches 95.5 MODA on Wildtrack at 6.1 FPS, so the learned geometry transfers to annotation-based pipelines as well.
  • The iterative loop converges in three rounds: after the third optimization cycle, valid pseudo-depth regions stop growing and accuracy gains become minimal, giving a practical stopping rule.
  • Because no 3D annotations are needed, the same training recipe can be applied to a new calibrated camera setup using only images and per-view segmentation masks, as done for the 4200 sampled Wildtrack frames per camera.
  • At inference, human modeling takes about 0.1 seconds per frame and full label-free detection runs at 1.2 FPS, a real-time operating point for surveillance-scale use.

Reading between the lines

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

  • An editorial inference: nothing in the depth-consistency loop is pedestrian-specific, so the same superpixel Gaussian pseudo-depth fine-tuning could be applied to other sparse-view multiview detection tasks, such as vehicles or animals, whenever per-view masks are available.
  • The paper's own limitation note implies that detection-compensation quality is capped by depth accuracy; a natural extension would be to gate SAM prompts with an estimated per-projection depth uncertainty and to validate the compensated masks before accepting them.
  • Because 78.67% of Wildtrack pedestrians appear in only one to three views, the method's gains likely concentrate on the multi-view majority of pixels; incorporating temporal consistency is the most direct test of whether the remaining single-view-only cases can be pushed further.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes DCHM, a label-free pipeline for multiview pedestrian detection. The method first generates pseudo-depth labels for pedestrians by optimizing a 3D Gaussian Splatting scene with superpixel-wise supervision, filters those labels by cross-view foreground and depth-consistency checks, and uses the filtered labels to fine-tune a monocular depth estimator. Missed detections are compensated by projecting source-view masks into a reference view and refining them with SAM. At inference, the fine-tuned depth maps are fused into a global Gaussian point cloud, matched across views via a label-matching algorithm, and clustered with DBSCAN to produce pedestrian locations. Experiments on Wildtrack, Terrace, and MultiviewX report label-free state-of-the-art results (MODA 84.2, 80.1, and 78.4 respectively), and the supplementary adds a supervised variant that reaches 95.5 MODA on Wildtrack.

Significance. If the reported numbers are correct under a clean protocol, the paper makes a useful contribution: it demonstrates a fully label-free way to obtain multiview-consistent depth for crowded, sparse-view scenes, and it shows that the resulting point clouds can support not only detection but also multiview pedestrian segmentation. The paper is also commendable for including a decomposition against UMPD, ablations for superpixel-wise optimization, detection compensation, and the iterative loop, plus a DTU-based evaluation of depth quality. The code is promised on a project page. These strengths are, however, conditional on resolving the evaluation-protocol overlap and the self-referential nature of the pseudo-depth loop, as detailed in the major comments.

major comments (4)
  1. [§4.1, §4.2, §3.1.4] The Wildtrack evaluation protocol does not state whether the 40 labeled evaluation frames are excluded from the 4200 frames per camera used for mono-depth fine-tuning. Section 4.1 says evaluation uses 40 labeled frames, and Section 4.2 says 4200 images per camera are sampled at 2 fps from the same 35-minute video; the default reading is that the 40 labeled frames lie inside the fine-tuning set. Section 3.1.4 describes an iterative loop in which pseudo-depth generation, mono-depth fine-tuning, and detection compensation all operate on those frames. If the evaluation frames are in the fine-tuning set, the reported MODA values in Tables 2 and 3 and the comparison with UMPD could reflect test-time adaptation rather than a generalizable depth-consistent model. The authors should state whether the 40 frames are excluded, and if they are not, rerun the full pipeline with a clean split.
  2. [§3.1.1, §3.1.3, §3.2.2, Algorithms 1–4] Several hyperparameters that the pipeline depends on are never reported: the loss weights λ_sp, λ_m, λ_d, λ_o in Eq. (4); the depth-consistency threshold τ in the pseudo-depth filter; the visibility threshold τ_vis in Algorithm 1; the cluster threshold τ_cluster, DBSCAN ϵ and minPts in Algorithm 2; and the blending-weight threshold used in localization. Without these values, the results in Tables 2 and 6 are not reproducible, and the paper provides no sensitivity analysis for parameters that directly control how much pseudo-depth is retained and how many Gaussians form a detection. At minimum, the full parameter setting should be reported, and a sensitivity experiment over τ and τ_vis should be added to show that the state-of-the-art claim is not an artifact of a single threshold choice.
  3. [§3.1.1, supplementary Fig. 10.B, §13] The pseudo-depth filtering strategy requires a source-view pixel to reproject into foreground in every reference view and to agree with GS-rendered depth in at least one other view. Supplementary Fig. 10.B reports that 78.67% of Wildtrack pedestrians appear in only one to three views. Pedestrians visible in only one view can never satisfy the depth-consistency condition, and mask errors in any of the few supporting views cause their pixels to be pruned by the foreground filter. Since YOLOv11 masks are the only supervision entering the GS loss and the filtering steps, the paper should quantify mask quality per view and report, for the final Wildtrack detections, how many pedestrians have zero valid pseudo-depth labels, how many detections come from refined multi-view geometry versus compensation, and how sensitive Table 2 is to mask completeness. The limitation statement in §13 acknowledges single-view instability qualitatively, but the main claim needs a quantitative account.
  4. [§3.1.4, §3.1.1, Fig. 5] The pseudo-depth loop has a self-referential component that is not discussed. Section 3.1.4 states that the fine-tuned model's updated depth predictions serve as initialization for the next Gaussian optimization cycle, while Section 3.1.1 accepts a pseudo-depth label only if it agrees with GS-rendered depth in another view. Once GS is initialized from the fine-tuned monocular depth, agreement between reprojected depth and GS-rendered depth may simply reflect the initializer rather than independent multiview evidence. The paper should break or analyze this circularity, for example by comparing against fine-tuning with unfiltered monocular depth, by measuring the agreement between pseudo-labels and an independent source (e.g., a stereo method or ground-truth depth on a subset), or by reporting how much of the improvement remains after only one loop without re-initialization.
minor comments (5)
  1. [§4.3] The word 'Supersingly' should be 'Surprisingly'.
  2. [§3.1.1] 'scaler weights' should be 'scalar weights' in the description of Eq. (4).
  3. [References] The Terrace dataset is cited as reference [12], but reference [12] is the probabilistic occupancy map paper; the correct Terrace dataset citation should be provided.
  4. [Table 2] Several baselines have '-' entries for Terrace and MultiviewX; the paper should state whether those results are unavailable from the original papers or were omitted for another reason, since the comparison is uneven otherwise.
  5. [§4.3, Table 2] Report whether the numbers in Table 2 are single runs or averaged over multiple runs, and if averaged, give standard deviations; if the pipeline is deterministic, say so explicitly.

Circularity Check

2 steps flagged · score 7.0 of 10

Wildtrack evaluation frames are arithmetically inside the fine-tuning set, and the iterative pseudo-depth loop trains on labels derived from the same network's own depth predictions.

  1. fitted input called prediction [Section 4.1 (Experimental settings) and Section 4.2 (Implementation details)]
    "We sample 4200 frames per camera for fine-tuning and use 40 labeled frames for evaluation. ... we extract 4200 images per camera from 35-minute videos sampled at 2 frames per second."

    A 35-minute video sampled at 2 fps yields exactly 4200 frames per camera, so the fine-tuning set covers the whole timeline of the video from which the 40 labeled evaluation frames are drawn; the paper never states that the evaluation frames were excluded. Consequently the mono-depth model is, on the face of the text, fine-tuned on the same frames whose MODA is reported in Table 2. The headline numbers 84.2, 80.1, and 78.4 are then values fitted on the evaluation frames rather than held-out predictions, fitting the pattern of a fitted input renamed as a prediction.

  2. self definitional [Section 3.1.4 (Iterative improvement) and Section 3.1.1 (Pseudo-depth filtering), with Eqs. 12-14 in Supplementary Section 11]
    "Gaussian optimization generates pseudo-labels to supervise monocular depth estimation, and the fine-tuned network’s updated depth predictions serve as initialization for the next Gaussian optimization cycle."

    After the first cycle, the network being fine-tuned supplies the initialization for the GS optimization that produces the pseudo-depth labels, and the cross-view filter retains a depth only when it agrees with GS-rendered depth in another view. The training target is therefore a function of the same network's previous outputs: 'multiview depth consistency' is defined as agreement with a reconstruction initialized from the model itself, not with externally measured 3D geometry. The growing 'valid pseudo-depth' regions in Figure 7 can reflect convergence to a self-consistent fixed point, so the claimed consistency improvement is partly self-confirming rather than independently verified.

full rationale

The paper's central derivation is partly compromised by two self-referential constructions. First, the Wildtrack evaluation frames are arithmetically contained in the fine-tuning set: 35 minutes at 2 fps yields 4200 frames per camera, matching the stated fine-tuning count, and the 40 labeled evaluation frames are not stated to be excluded. The final mono-depth model has therefore seen the exact test images on the face of the text, so the SOTA MODA numbers in Table 2 are fit values, not held-out predictions. Second, the iterative loop in Section 3.1.4 makes the pseudo-depth supervision depend on the same mono-depth network's own outputs: after the first iteration the fine-tuned network initializes GS, GS renders depth, cross-view filtering keeps depth agreeing with that rendering, and the filtered depth fine-tunes the same network. This makes the claimed multiview depth consistency a self-agreement criterion. The method is not wholly circular: the first GS optimization starts from ray-sampled superpixel Gaussians with photometric and mask supervision, and there is no load-bearing self-citation or imported uniqueness theorem. However, the two reductions above affect the paper's headline empirical claims, so the circularity score is 7 rather than lower.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The method relies on calibrated cameras, reliable 2D person masks, and the ability to bootstrap metric depth from relative monocular estimates via ground-plane alignment. The core self-training loop is not externally supervised, and thresholds and loss weights are hand-set rather than derived.

free parameters (7)
  • Loss weights lambda_sp, lambda_m, lambda_d, lambda_o = not reported
    Eq. 4 combines superpixel photometric, mask, opacity, and depth-constraint terms; the weights are hand-set scalars, but their values are never stated.
  • Number of superpixels per pedestrian K = 30
    Section 4.2 sets K=30; this controls the resolution of the superpixel-wise photometric supervision.
  • Cross-view depth consistency threshold tau = not reported
    Section 3.1.1 retains depth when |Ds->r - Dr| < tau; tau is pre-defined but its value does not appear in the paper.
  • Visibility threshold tau_vis = not reported
    Algorithm 1 assigns an ID only when the Gaussian blending weight exceeds tau_vis; the value is omitted.
  • Cluster threshold tau_cluster = not reported
    Algorithm 2 keeps an identity only if it has more than tau_cluster Gaussians; the value is omitted.
  • DBSCAN epsilon and minPts = not reported
    Algorithms 2 and 3 require epsilon and minPts; the paper never reports their values.
  • Gaussian blending weight threshold = 0.05
    Section 4.2 discards Gaussians with blending weight below 0.05 during localization.
assumptions (5)
  • domain assumption Cameras are calibrated with known intrinsics/extrinsics and frames are synchronized.
    Used throughout unprojection, fusion, cross-view filtering, and ground depth computation (Sec. 3.1.1, Eq. 3, Eq. 12-14).
  • domain assumption YOLOv11 pedestrian masks are accurate enough to serve as supervision for GS and matching.
    Eq. 5-7 define losses against M^v and Algorithm 1 matches IDs using these masks; no mask-quality analysis is provided.
  • domain assumption The ground plane is known and metric depth can be recovered by median scaling of relative depth with ground-truth ground depth.
    Section 4.3 computes scale = median(M_g * d_metric / M_g * d_rel) using camera calibration and a known ground plane.
  • ad hoc to paper Depth is approximately constant within each pedestrian mask (depth constraint loss).
    Eq. 7-8 penalize within-mask depth variance, flattening pedestrians; this biases geometry but aids clustering.
  • ad hoc to paper Pseudo-depth labels filtered by internal cross-view consistency are reliable enough to fine-tune a monocular depth foundation model.
    Sec. 3.1.1 and Fig. 5 filter by agreement with GS-rendered depth, which is itself part of the self-training loop.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DCHM: Depth-Consistent Human Modeling for Multiview Detection." pith.science (2026). https://pith.science/paper/V35ITIBW

@misc{pith2026250714505,
  author       = {Pith},
  title        = {Pith review of: DCHM: Depth-Consistent Human Modeling for Multiview Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/V35ITIBW}},
  note         = {Machine review of arXiv:2507.14505}
}
read the original abstract

Multiview pedestrian detection typically involves two stages: human modeling and pedestrian localization. Human modeling represents pedestrians in 3D space by fusing multiview information, making its quality crucial for detection accuracy. However, existing methods often introduce noise and have low precision. While some approaches reduce noise by fitting on costly multiview 3D annotations, they often struggle to generalize across diverse scenes. To eliminate reliance on human-labeled annotations and accurately model humans, we propose Depth-Consistent Human Modeling (DCHM), a framework designed for consistent depth estimation and multiview fusion in global coordinates. Specifically, our proposed pipeline with superpixel-wise Gaussian Splatting achieves multiview depth consistency in sparse-view, large-scaled, and crowded scenarios, producing precise point clouds for pedestrian localization. Extensive validations demonstrate that our method significantly reduces noise during human modeling, outperforming previous state-of-the-art baselines. Additionally, to our knowledge, DCHM is the first to reconstruct pedestrians and perform multiview segmentation in such a challenging setting. Code is available on the \href{https://jiahao-ma.github.io/DCHM/}{project page}.

Figures

Figures reproduced from arXiv: 2507.14505 by the authors.

Figure 1
Figure 1. Depth-Consistent Human Modeling. We propose depth-consistent human modeling in multi-view pedestrian detection by leveraging monocular depth estimation to fuse multi-view information into a global coordinate system represented by point clouds. These fused point clouds serve as inputs for pedestrian localization. Stereo-based methods [23] struggle with sparse views and severe occlusions, causing inaccurate geometry p… view at source ↗
Figure 2
Figure 2. Modeling humans as segmented Gaussians. Recon￾struction and multiview segmentation are performed in sparse￾view, large-scale, and occluded scenes. suitable solution. While existing monocular depth estima￾tion methods [3, 15, 17, 49] offer detailed depth from single images, backprojecting each inferred depth map to the 3D space leads to misaligned point cloud for 3D human mod￾eling as those depth maps lack cross-view… view at source ↗
Figure 3
Figure 3. Overview of the framework. The proposed multiview detection pipeline consists of separate training (left) and inference (right) stages. During training (Sec. 3.1), human modeling optimization refines mono-depth estimation for multiview consistent depth prediction, via pseudo-depth generation, mono-depth fine-tuning, and detection compensation. During inference (Sec. 3.2), the optimized mono-depth produces 3D point c… view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Superpixel-wise Gaussian initialization and optimization. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Pseudo-depth filtering strategies. rendered mask Mˆ v as: Lm = X v X u ∥Mv (u) − Mˆ v (u)∥ 2 2 . (6) Depth constraint loss Ld.We encourage a small variance of depth predictions for pixels inside each pedestrian mask: Ld = X v X i X u:Mv i (u)=1 ∥Dv (u) − ¯d v i ∥ 2 2 ,…
Figure 6
Figure 6. Figure 6: Multi-view compensation for missed detection. (A) Source-view masks to be projected into the reference view by predicted depth; (B) the projected elements form point-box prompts for SAM; (C) these prompts facilitate segmentation; (D) yielding compensatory detections in…
Figure 7
Figure 7. Figure 7: Increasing valid pseudo-depth during iterative learn￾ing. Binary masks illustrate the valid pseudo-depth regions, with white color indicating valid areas and black color denoting invalid ones. Green highlights newly added valid depth regions compared to the previous op…
Figure 9
Figure 9. Figure 9: Ablation study. Left: Pedestrian detection AP (%) for YOLOv9, YOLOv11, and their enhanced versions (“Ours”), showing improved performance. Right: MODA (%) and training time (hours) across loops, illustrating the accuracy-cost trade-off. label-free methods [25, 27, 28, …
Figure 10
Figure 10. Figure 10: (A) Error comparison with DA-v2 on DTU. (B) His￾togram of the number of pedestrians captured simultaneously by multiple cameras [PITH_FULL_IMAGE:figures/full_fig_p014_10.png]
Figure 11
Figure 11. Figure 11: Visual comparison of DepthAnything-v2 (DA-v2) with [PITH_FULL_IMAGE:figures/full_fig_p014_11.png]
Figure 12
Figure 12. Figure 12: Enlarged 3D reconstruction from various methods. We present enlarged views of 3D reconstructions generated by base￾lines [3, 15, 17, 23, 50], showcasing both front and back perspectives for each. Our method (“Ours”) produces more accurate and complete reconstructions …
Figure 13
Figure 13. Figure 13: Visualization of increasing valid pseudo-depth across all cameras. We illustrate the progression of valid depth regions for all viewpoints during each round of optimization. This serves as a supplement visualization for [PITH_FULL_IMAGE:figures/full_fig_p016_13.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

65 extracted references · 52 canonical work pages

  1. [1]

    Slic superpixels

    Radhakrishna Achanta, Appu Shaji, Kevin Smith, Aurelien Lucchi, Pascal Fua, and Sabine S ¨usstrunk. Slic superpixels

  2. [2]

    Deep oc- clusion reasoning for multi-camera multi-target detection

    Pierre Baqu ´e, Franc ¸ois Fleuret, and Pascal Fua. Deep oc- clusion reasoning for multi-camera multi-target detection. In Proceedings of the IEEE International Conference on Computer Vision, pages 271–279, 2017. 2, 3

  3. [3]

    Depth pro: Sharp monocular metric depth in less than a second

    Aleksei Bochkovskii, Ama ¨el Delaunoy, Hugo Germain, Marcel Santos, Yichao Zhou, Stephan R Richter, and Vladlen Koltun. Depth pro: Sharp monocular metric depth in less than a second. arXiv preprint arXiv:2410.02073, 2024. 1, 2, 7, 4

  4. [4]

    MVUDA: Unsupervised Domain Adaptation for Multi-view Pedestrian Detection

    Erik Brorsson, Lennart Svensson, Kristofer Bengtsson, and Knut ˚Akesson. Mvuda: Unsupervised domain adapta- tion for multi-view pedestrian detection. arXiv preprint arXiv:2412.04117, 2024. 3

  5. [5]

    La- bel efficient lifelong multi-view broiler detection

    Thorsten Cardoen, Sam Leroux, and Pieter Simoens. La- bel efficient lifelong multi-view broiler detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5393–5402, 2024. 3

  6. [6]

    Wild- track: A multi-camera hd dataset for dense unscripted pedes- trian detection

    Tatjana Chavdarova, Pierre Baqu ´e, St ´ephane Bouquet, An- drii Maksai, Cijo Jose, Timur Bagautdinov, Louis Lettry, Pascal Fua, Luc Van Gool, and Franc ¸ois Fleuret. Wild- track: A multi-camera hd dataset for dense unscripted pedes- trian detection. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 5030– 5039, 2018. 7

  7. [7]

    High-quality surface recon- struction using gaussian surfels

    Pinxuan Dai, Jiamin Xu, Wenxiang Xie, Xinguo Liu, Huamin Wang, and Weiwei Xu. High-quality surface recon- struction using gaussian surfels. In ACM SIGGRAPH 2024 Conference Papers, pages 1–11, 2024. 5

  8. [8]

    Cosseggaussians: Compact and swift scene segmenting 3d gaussians with dual feature fusion

    Bin Dou, Tianyu Zhang, Yongjia Ma, Zhaohui Wang, and Zejian Yuan. Cosseggaussians: Compact and swift scene segmenting 3d gaussians with dual feature fusion. arXiv preprint arXiv:2401.05925, 2024. 3

Show all 65 references
  1. [9]

    Multi-view tracking using weakly supervised human motion prediction

    Martin Engilberge, Weizhe Liu, and Pascal Fua. Multi-view tracking using weakly supervised human motion prediction. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 1582–1592, 2023. 3

  2. [10]

    Density-based spatial clustering of applications with noise

    Martin Ester, Hans-Peter Kriegel, J ¨org Sander, and Xiaowei Xu. Density-based spatial clustering of applications with noise. In Int. Conf. knowledge discovery and data mining,

  3. [11]

    Multicamera people tracking with a probabilistic occupancy map

    Francois Fleuret, Jerome Berclaz, Richard Lengagne, and Pascal Fua. Multicamera people tracking with a probabilistic occupancy map. IEEE transactions on pattern analysis and machine intelligence, 30(2):267–282, 2007. 3

  4. [12]

    Multicamera people tracking with a probabilistic occupancy map

    Francois Fleuret, Jerome Berclaz, Richard Lengagne, and Pascal Fua. Multicamera people tracking with a probabilistic occupancy map. IEEE transactions on pattern analysis and machine intelligence, 30(2):267–282, 2007. 7

  5. [13]

    Multiview detection with shadow transformer (and view-coherent data augmentation)

    Yunzhong Hou and Liang Zheng. Multiview detection with shadow transformer (and view-coherent data augmentation). In Proceedings of the 29th ACM International Conference on Multimedia, pages 1673–1682, 2021. 2, 3, 8

  6. [14]

    Multi- view detection with feature perspective transformation

    Yunzhong Hou, Liang Zheng, and Stephen Gould. Multi- view detection with feature perspective transformation. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part VII 16, pages 1–18. Springer, 2020. 2, 3, 7, 8

  7. [15]

    Metric3d v2: A versatile monocular geomet- ric foundation model for zero-shot metric depth and surface normal estimation

    Mu Hu, Wei Yin, Chi Zhang, Zhipeng Cai, Xiaoxiao Long, Hao Chen, Kaixuan Wang, Gang Yu, Chunhua Shen, and Shaojie Shen. Metric3d v2: A versatile monocular geomet- ric foundation model for zero-shot metric depth and surface normal estimation. arXiv preprint arXiv:2404.15506, 20...

  8. [16]

    Semantic anything in 3d gaussians

    Xu Hu, Yuxi Wang, Lue Fan, Junsong Fan, Junran Peng, Zhen Lei, Qing Li, and Zhaoxiang Zhang. Semantic anything in 3d gaussians. arXiv preprint arXiv:2401.17857, 2024. 2, 3

  9. [17]

    Repurpos- ing diffusion-based image generators for monocular depth estimation

    Bingxin Ke, Anton Obukhov, Shengyu Huang, Nando Met- zger, Rodrigo Caye Daudt, and Konrad Schindler. Repurpos- ing diffusion-based image generators for monocular depth estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024...

  10. [18]

    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., 42(4):139–1,

  11. [19]

    Yolov11: An overview of the key architectural enhancements

    Rahima Khanam and Muhammad Hussain. Yolov11: An overview of the key architectural enhancements. arXiv preprint arXiv:2410.17725, 2024. 7, 8

  12. [20]

    Segment anything

    Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer White- head, Alexander C Berg, Wan-Yen Lo, et al. Segment anything. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 4015–4026, 2023. 3, 5

  13. [21]

    Pointpillars: Fast encoders for object detection from point clouds

    Alex H Lang, Sourabh V ora, Holger Caesar, Lubing Zhou, Jiong Yang, and Oscar Beijbom. Pointpillars: Fast encoders for object detection from point clouds. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12697–12705, 2019. 2

  14. [22]

    Multi-view target transformation for pedestrian detection

    Wei-Yu Lee, Ljubomir Jovanov, and Wilfried Philips. Multi-view target transformation for pedestrian detection. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 90–99, 2023. 3

  15. [23]

    Ground- ing image matching in 3d with mast3r

    Vincent Leroy, Yohann Cabon, and J´erˆome Revaud. Ground- ing image matching in 3d with mast3r. arXiv preprint arXiv:2406.09756, 2024. 1, 7, 4

  16. [24]

    Language-driven semantic seg- mentation

    Boyi Li, Kilian Q Weinberger, Serge Belongie, Vladlen Koltun, and Ren ´e Ranftl. Language-driven semantic seg- mentation. arXiv preprint arXiv:2201.03546, 2022. 3

  17. [25]

    Generalizable multi- camera 3d pedestrian detection

    Joao Paulo Lima, Rafael Roberto, Lucas Figueiredo, Fran- cisco Simoes, and Veronica Teichrieb. Generalizable multi- camera 3d pedestrian detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, pages 1232–1240, 2021. 2, 3, 8

  18. [26]

    Focal loss for dense object detection

    Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Doll ´ar. Focal loss for dense object detection. In Proceedings of the IEEE international conference on computer vision, pages 2980–2988, 2017. 2

  19. [27]

    Un- supervised multi-view pedestrian detection

    Mengyin Liu, Chao Zhu, Shiqi Ren, and Xu-Cheng Yin. Un- supervised multi-view pedestrian detection. In Proceedings of the 32nd ACM International Conference on Multimedia, pages 1034–1042, 2024. 1, 2, 3, 8

  20. [28]

    Semantic-driven multi- camera pedestrian detection

    Alejandro L ´opez-Cifuentes, Marcos Escudero-Vi˜nolo, Jes´us Besc´os, and Pablo Carballeira. Semantic-driven multi- camera pedestrian detection. Knowledge and Information Systems, 64(5):1211–1237, 2022. 2, 3, 8

  21. [29]

    Gaga: Group any gaussians via 3d-aware memory bank, 2024

    Weijie Lyu, Xueting Li, Abhijit Kundu, Yi-Hsuan Tsai, and Ming-Hsuan Yang. Gaga: Group any gaussians via 3d-aware memory bank, 2024. 2, 3

  22. [30]

    V oxelized 3d fea- ture aggregation for multiview detection

    Jiahao Ma, Jinguang Tong, Shan Wang, Wei Zhao, Liang Zheng, and Chuong Nguyen. V oxelized 3d fea- ture aggregation for multiview detection. arXiv preprint arXiv:2112.03471, 2021. 2, 3

  23. [31]

    Multiview detection with cardboard human mod- eling

    Jiahao Ma, Zicheng Duan, Liang Zheng, and Chuong Nguyen. Multiview detection with cardboard human mod- eling. arXiv preprint arXiv:2207.02013, 2022. 2, 3, 8

  24. [32]

    Pointnet: Deep learning on point sets for 3d classification and segmentation

    Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. Pointnet: Deep learning on point sets for 3d classification and segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 652–660,

  25. [33]

    3d random occlusion and multi-layer projection for deep multi-camera pedestrian localization

    Rui Qiu, Ming Xu, Yuyao Yan, Jeremy S Smith, and Xi Yang. 3d random occlusion and multi-layer projection for deep multi-camera pedestrian localization. arXiv preprint arXiv:2207.10895, 2022. 2, 3

  26. [34]

    Conditional random fields for multi-camera object detection

    Gemma Roig, Xavier Boix, Horesh Ben Shitrit, and Pas- cal Fua. Conditional random fields for multi-camera object detection. In 2011 International Conference on Computer Vision, pages 563–570. IEEE, 2011. 3

  27. [35]

    Structure-from-motion revisited

    Johannes Lutz Sch ¨onberger and Jan-Michael Frahm. Structure-from-motion revisited. In Conference on Computer Vision and Pattern Recognition (CVPR), 2016. 2, 4

  28. [36]

    Flashsplat: 2d to 3d gaussian splatting segmentation solved optimally

    Qiuhong Shen, Xingyi Yang, and Xinchao Wang. Flashsplat: 2d to 3d gaussian splatting segmentation solved optimally. European Conference of Computer Vision, 2024. 2

  29. [37]

    Stacked homography transformations for multi-view pedestrian detection

    Liangchen Song, Jialian Wu, Ming Yang, Qian Zhang, Yuan Li, and Junsong Yuan. Stacked homography transformations for multi-view pedestrian detection. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 6049–6057, 2021. 2, 3, 8

  30. [38]

    Selfpose3d: Self-supervised multi-person multi-view 3d pose estimation

    Vinkle Srivastav, Keqi Chen, and Nicolas Padoy. Selfpose3d: Self-supervised multi-person multi-view 3d pose estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2502–2512, 2024. 3

  31. [39]

    The clear 2006 evaluation

    Rainer Stiefelhagen, Keni Bernardin, Rachel Bowers, John Garofolo, Djamel Mostefa, and Padmanabhan Soundarara- jan. The clear 2006 evaluation. In Multimodal Technologies for Perception of Humans: First International Evaluation Workshop on Classification of Events, Activities a...

  32. [40]

    Earlybird: Early-fusion for multi- view tracking in the bird’s eye view

    Torben Teepe, Philipp Wolters, Johannes Gilg, Fabian Her- zog, and Gerhard Rigoll. Earlybird: Early-fusion for multi- view tracking in the bird’s eye view. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 102–111, 2024. 3

  33. [41]

    Bringing generalization to deep multi-view pedestrian detection

    Jeet V ora, Swetanjal Dutta, Kanishk Jain, Shyamgopal Karthik, and Vineet Gandhi. Bringing generalization to deep multi-view pedestrian detection. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (W ACV)Workshops, pages 110–119, 2023. 3

  34. [42]

    3d-muppet: 3d multi-pigeon pose es- timation and tracking

    Urs Waldmann, Alex Hoi Hang Chan, Hemal Naik, M ´at´e Nagy, Iain D Couzin, Oliver Deussen, Bastian Goldluecke, and Fumihiro Kano. 3d-muppet: 3d multi-pigeon pose es- timation and tracking. International Journal of Computer Vision, pages 1–18, 2024. 3

  35. [43]

    Yolov9: Learning what you want to learn us- ing programmable gradient information

    Chien-Yao Wang, I-Hau Yeh, and Hong-Yuan Mark Liao. Yolov9: Learning what you want to learn us- ing programmable gradient information. arXiv preprint arXiv:2402.13616, 2024. 8

  36. [44]

    SparseGS: Real-time 360° sparse view synthesis using Gaussian splatting

    Haolin Xiong. SparseGS: Real-time 360° sparse view synthesis using Gaussian splatting. University of California, Los Angeles, 2024. 4

  37. [45]

    Mvpgs: Excavating multi-view priors for gaussian splatting from sparse input views

    Wangze Xu, Huachen Gao, Shihe Shen, Rui Peng, Jianbo Jiao, and Ronggang Wang. Mvpgs: Excavating multi-view priors for gaussian splatting from sparse input views. In European Conference on Computer Vision, pages 203–220. Springer, 2024. 4

  38. [46]

    Embodiedsam: Online segment any 3d thing in real time

    Xiuwei Xu, Huangxing Chen, Linqing Zhao, Ziwei Wang, Jie Zhou, and Jiwen Lu. Embodiedsam: Online segment any 3d thing in real time. arXiv preprint arXiv:2408.11811,

  39. [47]

    Multi-view people tracking via hierarchical trajectory com- position

    Yuanlu Xu, Xiaobai Liu, Yang Liu, and Song-Chun Zhu. Multi-view people tracking via hierarchical trajectory com- position. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4256–4265,

  40. [48]

    Multicamera pedestrian detection using logic minimization

    Yuyao Yan, Ming Xu, Jeremy S Smith, Mo Shen, and Jin Xi. Multicamera pedestrian detection using logic minimization. Pattern Recognition, 112:107703, 2021. 3

  41. [49]

    Depth anything: Unleashing the power of large-scale unlabeled data

    Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, and Hengshuang Zhao. Depth anything: Unleashing the power of large-scale unlabeled data. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10371–10381, 2024. 2

  42. [50]

    Depth any- thing v2

    Lihe Yang, Bingyi Kang, Zilong Huang, Zhen Zhao, Xiao- gang Xu, Jiashi Feng, and Hengshuang Zhao. Depth any- thing v2. arXiv:2406.09414, 2024. 1, 7, 3, 4

  43. [51]

    Gaussian grouping: Segment and edit anything in 3d scenes

    Mingqiao Ye, Martin Danelljan, Fisher Yu, and Lei Ke. Gaussian grouping: Segment and edit anything in 3d scenes. arXiv preprint arXiv:2312.00732, 2023. 2, 3

  44. [52]

    Feature 3dgs: Supercharg- ing 3d gaussian splatting to enable distilled feature fields

    Shijie Zhou, Haoran Chang, Sicheng Jiang, Zhiwen Fan, Ze- hao Zhu, Dejia Xu, Pradyumna Chari, Suya You, Zhangyang Wang, and Achuta Kadambi. Feature 3dgs: Supercharg- ing 3d gaussian splatting to enable distilled feature fields. In Proceedings of the IEEE/CVF Conference on Comp...

  45. [53]

    V oxelnet: End-to-end learning for point cloud based 3d object detection

    Yin Zhou and Oncel Tuzel. V oxelnet: End-to-end learning for point cloud based 3d object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4490–4499, 2018. 2

  46. [54]

    Multi-camera people detection and tracking,

    Chuting Zhu. Multi-camera people detection and tracking,

  47. [55]

    Fsgs: Real-time few-shot view synthesis using gaussian splatting

    Zehao Zhu, Zhiwen Fan, Yifan Jiang, and Zhangyang Wang. Fsgs: Real-time few-shot view synthesis using gaussian splatting. In European conference on computer vision, pages 145–163. Springer, 2024. 4 DCHM: Depth-Consistent Human Modeling for Multiview Detection Supplementary Mat...

  48. [56]

    Methodology

    Supervised localization We conducted additional experiments to explore if our depth-consistent human modeling can enhance label-based methods. Methodology. For supervised localization, we adopt the Method Label-based Accuracy (MODA) Speed(F P S) UMPD [27] ✗ 76.6 1 .0 Ours ✗ 84...

  49. [57]

    Comparison

    for position regression, defined as: Lreg = −α(1 − p)γ log(p), (10) where α and γ are hyperparameters specified in [26]. Comparison. We replace label-free clustering with label- based aggregation to explore the potential of our frame- work under supervised settings. Quantitati...

  50. [58]

    The main functions, such as matching and clustering, are listed in Algorithm 1 and 2

    Unsupervised localization Our multiview unsupervised localization is summarized in Algorithm 3. The main functions, such as matching and clustering, are listed in Algorithm 1 and 2

  51. [59]

    Multi-view Detection Compensation We formalize our multiview detection compensation ap- proach in Algorithm 4

  52. [60]

    Integrating pro- posed human modeling with supervised method outper- forms existing label-based approaches, achieving both the highest accuracy and computational speed

    Computational cost We provide an additional comparison of accuracy (MODA) and computational efficiency in Table 5. Integrating pro- posed human modeling with supervised method outper- forms existing label-based approaches, achieving both the highest accuracy and computational speed

  53. [61]

    Although the ground range differs in the MultiviewX dataset, the calculation method remains consistent across both datasets

    Ground depth calculation This section explains the process of calculating ground depth in the Wildtrack dataset. Although the ground range differs in the MultiviewX dataset, the calculation method remains consistent across both datasets. The ground plane is specified in a worl...

  54. [62]

    We define us as the homogenous pixel coordinate in source view vs, Ks is the source view camera intrinsic matrix

    GS depth filtering details This section provides additional details about the GS depth filtering process. We define us as the homogenous pixel coordinate in source view vs, Ks is the source view camera intrinsic matrix. We unproject each pixel us from source view vs into a 3D ...

  55. [63]

    Discussion Sparse view setup in multi-view detection differs from existing sparse view reconstruction in several aspects

  56. [64]

    Performance (%) of supervised methods on Wildtrack and MultiviewX.Our-Sup: our human modeling + supervised localization

    Large-scale scene: unlike traditional indoor or object- level settings, we tackle crowded surveillance (e.g., the Wildtrack dataset) covering a 30 m × 40 m plaza with over Method Wildtrack MultiviewX MODA MODP Precision Recall MODA MODP Precision Recall MVDet [14] 88.7 73 .6 9...

  57. [65]

    While multiview detection compensation improves monoc- ular detection, its performance is highly dependent on the accuracy of depth estimation

    Limitations and future works Impact of depth estimation on detection compensation. While multiview detection compensation improves monoc- ular detection, its performance is highly dependent on the accuracy of depth estimation. Error in depth estimation can result in imprecise ...

Pith tools

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