Pith. sign in

REVIEW 3 major objections 6 minor 55 references

No Identity, no problem: Motion through detection for people tracking

T0 review · 3 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read People tracking can be trained from detection labels alone: the paper learns a 2D motion field by warping one detection heatmap onto the next and enforcing consistency, with no identity or motion annotations.

desk verdict Multi-view motion-supervision story is solid, but the 'no identity' claim is untested in single-view because OSNet re-ID features are baked into the input. read the letter →

arxiv 2411.16466 v1 pith:R4M6Q6QA submitted 2024-11-25 cs.CV cs.LG

classification cs.CVcs.LG
keywords multi-objecttrackingdetection-onlysupervisionmotionestimationheatmapwarpingdifferentiablereconstructionlowframeratemulti-view
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

This paper argues that motion estimation for people tracking does not need identity or motion annotations: the consistency between two detection heatmaps, linked by a predicted 2D displacement field, is itself enough supervision. The authors predict detection heatmaps at times $t$ and $t+1$ plus an offset map, warp the first heatmap by the offsets, and penalize disagreement with the second. This couples information across frames during training and is shown to produce motion offsets of near-fully-supervised quality, while improving single-view low-frame-rate MOTA by more than 10 points over ByteTrack and giving higher multi-view tracking numbers than the compared baselines on WILDTRACK. The method works with existing detectors and trackers, so the practical payoff is cheaper annotation for robust tracking in crowded scenes and low-frame-rate video.

What carries the argument

The load-bearing mechanism is the differentiable reconstruction from motion: $\hat{x}^{t+1}_j = \sum_{i \in G} x^t_i \, W\big(d(j,\, i + \delta^{t,t+1}_i)\big)$, with $W(l) = 1/(1 + e^{4\lambda_r l - 10})$ a sigmoidal distance weight. It turns the offset map into a predicted future heatmap, so the detection loss backpropagates into the motion predictor; $\lambda_r$ starts small for easy optimization and increases to sharpen the reconstruction. The forward/backward loss $L_{\mathrm{fb}}$ (reversing frame order and requiring $\delta^{t+1,t} \approx -\delta^{t,t+1}$) and the spatial-extent loss $L_{\mathrm{se}}$ (encouraging offset smoothness within detection peaks) are what break the symmetry of $L_{\mathrm{mot}}$'s multiple global minima.

What would settle it

Train the model with $L_{\mathrm{mot}}$ but without $L_{\mathrm{fb}}$ and $L_{\mathrm{se}}$ on data with known ground-truth displacements; if the predicted offsets keep near-zero $L_{\mathrm{mot}}$ while L1 error against true motion stays high, the core supervision alone does not recover motion.

Watch

Extended reading notes

Core claim

The central claim is that motion can be supervised through detection. For each pair of frames, the network outputs detection heatmaps $X^t$ and $X^{t+1}$ and a 2D offset map $\Delta^{t,t+1}$; a differentiable reconstruction module uses $\Delta$ to warp $X^t$ into a predicted $\hat{X}^{t+1}$, and the loss $L_{\mathrm{mot}} = \|\hat{X}^{t+1} - X^{t+1}_{\mathrm{gt}}\|^2$ pushes the offsets to explain the observed presence changes. Two regularizers, a forward/backward consistency $L_{\mathrm{fb}}$ and a spatial-extent $L_{\mathrm{se}}$, select the physically meaningful displacement among the many matchings that would zero $L_{\mathrm{mot}}$. On WILDTRACK the resulting offsets reach L1 error 0.58 versus 0.55 for a fully supervised counterpart and 1.06 for RAFT optical flow; on MOT17 at 2 FPS the modified ByteTrack improves MOTA from 59.1 to 65.5. The authors take this as evidence that the motion signal can be extracted from detection-level annotation alone.

Load-bearing premise

The consistency loss has many equally valid solutions, because any offset field that maps every detection at $t$ to some detection at $t+1$ gives zero loss, so the claim that the network learns true physical motion depends on the two regularizers and the training schedule selecting the right solution.

Editorial extensions

If this is right

  • Training a motion estimator no longer requires identity or motion labels; detection annotations, which are cheaper and more common, suffice.
  • Low-frame-rate tracking improves substantially: at 2 FPS on MOT17, MOTA rises from 59.1 to 65.5 over the ByteTrack baseline, and the gap widens as frame rate drops.
  • Learned offsets rival full supervision: L1 error 0.58 versus 0.55 on WILDTRACK, far ahead of RAFT optical flow at 1.06.
  • The motion field plugs into existing trackers, ByteTrack and muSSP, improving association without changing the detector or the tracking graph.
  • Multi-view tracking on WILDTRACK reaches MOTA 91.7, ahead of the compared methods, and the single-view approach transfers to MOT20 with similar low-FPS gains.

Reading between the lines

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

  • Because $L_{\mathrm{mot}}$ alone is satisfied by any detection-to-detection matching, the physical-motion claim rests on $L_{\mathrm{fb}}$ and $L_{\mathrm{se}}$; on scenes with repeated or symmetric motion patterns the offsets could degrade into a valid matching that is not true displacement.
  • The same consistency-by-warping recipe should transfer to other dense prediction tasks where temporal pairs exist but labels are scarce, such as keypoint tracking, instance segmentation, or generic scene flow.
  • A testable extension is to train the motion predictor jointly with the detector end-to-end, instead of using a frozen detector, and measure whether the consistency signal also sharpens detection in crowded scenes.
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

3 major / 6 minor

Summary. The paper proposes a tracking method in which a 2D motion offset map is learned without motion or identity annotations. The key idea is to predict detection heatmaps at two times together with an offset field, then warp the earlier heatmap by the offsets (Eq. 1-3) and require consistency with the later heatmap. The total loss (Eq. 4) combines this motion-consistency term with a detection loss, a forward-backward consistency loss (Eq. 6), and a spatial-extent regularization (Eq. 7). Two instantiations are presented: a single-view model built around YOLOX and a ResNet motion predictor, and a multi-view ground-plane model. Tracking is performed by modifying ByteTrack with learned motion and by using muSSP with a motion-based edge weight (Eq. 8). Experiments cover MOT17 (validation and test), WILDTRACK, MOT20, and MultiviewX, with ablations on loss components, frame intervals, and detection features.

Significance. If the claims hold, the method would be a practically valuable step toward reducing annotation cost in multi-object tracking: motion supervision emerges from detection heatmaps alone, and the multi-view results on WILDTRACK (MOTA 91.7 vs 91.3 for MVFlow) are competitive. The motion offsets are validated against identity-derived ground truth (Table 3: L1 0.58 vs 0.55 for fully supervised), the ablations cleanly separate the loss components (Table 4), and the nearest-neighbor and bipartite-matching baselines in Appendix A.2 show that the method is not merely performing trivial detection matching. However, the single-view 'no identity' claim is weakened by the use of OSNet re-ID features pretrained with identity labels as input to the motion predictor (Appendix A.4), and the low-FPS gains are obtained with training intervals matched to the test frame rate (Appendix A.5). These issues are addressable with additional experiments, so the contribution remains potentially sound.

major comments (3)
  1. [Appendix A.4 and Section 4.1.2] The single-view motion predictor receives 512-dimensional OSNet features per detection (Appendix A.4), and OSNet is pretrained with identity annotations. The main-text claim that motion is learned 'without identity or motion annotations' is therefore not established for the single-view experiments in Table 2 and Fig. 5: Table A.1 shows that adding these features improves MOTA/IDF1 at both 2 and 30 FPS, so the low-FPS improvement over ByteTrack may be partly due to the identity-pretrained representation rather than to the motion-consistency loss. Please report single-view results without these features (or with features from a detector-only backbone) and adjust the claim accordingly.
  2. [Section 3.3, Eq. (3); Section 4.4; Appendix A.2] The paper acknowledges in Section 4.4 that L_mot has multiple global minima: any offset map sending each detection at t to some detection at t+1 achieves zero loss. The nearest-neighbor and bipartite-matching baselines in Appendix A.2 are designed to realize such solutions, but they are evaluated only through downstream tracking MOTA/IDF1, not against the ground-truth motion metrics of Table 3. Because the claim that the network learns true motion (Table 3: L1 0.58 vs supervised 0.55) is load-bearing, please report L1/angle/norm errors for these two baselines as well, or otherwise demonstrate quantitatively that the learned offsets track physical motion rather than a valid matching.
  3. [Appendix A.5, Table A.1; Section 4.1.2] The training frame interval is selected to match the target test frame rate ('for all our experiments, we select the training sampling interval based on the desired target frame rate'). This makes the low-FPS gains in Fig. 5 and Table 2 partly a result of train/test interval matching, not of a single motion model that generalizes across frame rates. Please report results for a model trained once with a fixed interval and evaluated across all FPS values, or explicitly frame the per-FPS retraining as part of the method and discuss the associated compute cost.
minor comments (6)
  1. [Section 4.1.2] The text says detections are converted to 'gaussian heatmaps of 512 channels, first channel is a gaussian mask, while the remaining 512 channels...' which implies 513 channels in total; the next sentence says the first layer is modified to 513 input channels. Please correct this inconsistency.
  2. [Table A.4] The table heading says 'MOT17 val dataset' but the caption and content refer to the test set; fix the label.
  3. [Fig. 5] The y-axis begins at 0.5, which visually inflates the gap between the curves; consider starting at 0 or indicating an axis break.
  4. [Table 2 and Fig. 5] The single-view results are reported without variance or number of runs; given the small MOTA differences at 30 FPS (76.6 vs 77.0), please report multiple seeds or otherwise characterize uncertainty.
  5. [Throughout] Minor typos: 'Detetection features' in Table A.1; 'Bytetrack' in several captions; 'the remaining 512 channels are use to provide' (grammar) in Section 4.1.2.
  6. [Fig. 3 caption] The caption says a location 2 pixels away has weight 1 when lambda_r = 0.8, but inserting lambda_r = 0.8 and l = 2 into W(l) = 1/(1 + e^{4*lambda_r*l - 10}) gives approximately 0.97, not exactly 1; rephrase to 'close to 1'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: motion supervision is a proxy consistency loss whose outputs are validated against external ground-truth offsets.

full rationale

The central derivation chain is not circular. The motion offsets are learned by minimizing the reconstruction consistency loss L_mot of Eq. (3), which compares a heatmap warped by the predicted offset against the detection heatmap at time t+1, whose supervision comes from ground-truth detection maps through L_det in Eq. (5). The offset field is not defined to be the minimizer of the tracking metric; it is an intermediate prediction validated against external ground-truth displacements on WILDTRACK (Table 3) and MultiviewX (Table A.7). The acknowledged multiple-global-minima issue in Section 4.4 and Appendix A.2 is an identifiability limitation, not a circular reduction: the paper supplies regularizers L_fb and L_se and compares against nearest-neighbor and bipartite-matching baselines that also achieve L_mot = 0, showing the learned offsets are not merely an arbitrary matching. The single-view use of OSNet features pretrained with identity labels (Appendix A.4) is a legitimate confound for the 'no identity' claim, because those features are inputs to the motion predictor and Table A.1 shows they improve tracking; however, this is an external-validity or ablation concern, not a circularity, since the motion supervision still comes from detection consistency and the predicted offsets are still scored against ground-truth motion. Self-citations to Engilberge et al. (2023) are architectural and baseline references, not load-bearing uniqueness results or fitted parameters, so they do not make the derivation circular. The method is therefore self-contained in the sense required for a circularity finding; no step reduces by construction to its own input.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The method does not introduce new physical entities. It relies on standard domain assumptions (flat ground, calibration, Gaussian heatmap representation), on the reconstruction model being a faithful differentiable proxy for how detection heatmaps evolve, and on auxiliary losses resolving the degeneracy of L_mot. Free parameters are mostly loss weights and reconstruction schedule; association costs sigma_t, sigma_d, and sigma_m are not fully specified.

free parameters (6)
  • lambda_r = 0.8 initial, +0.08 per epoch, max 5
    Controls decay in Eq. 2 and therefore reconstruction accuracy vs differentiability; scheduled by hand.
  • lambda_fb = 0.05
    Weight of forward/backward consistency loss Eq. 6.
  • lambda_se = 1
    Weight of spatial extent loss Eq. 7.
  • muSSP edge weights sigma_t, sigma_d, sigma_m = not reported in paper
    Weights for temporal, spatial, and motion distances in Eq. 8; affect WILDTRACK tracking results.
  • sliding window size = 59 pixels
    Caps reconstructed motion range at 29.5 pixels; larger than any ground-truth motion but a free choice.
  • training frame interval = [1,2) to [8,13) depending on target FPS
    Selected per evaluation frame rate (Appendix A.5); matching the training interval to the test frame rate improves results.
assumptions (5)
  • domain assumption Ground is flat and camera homographies map image points to a common 2D ground plane.
    Section 3.1; standard in multi-view tracking; generalizable but assumed for MOT17 and WILDTRACK.
  • domain assumption Detection heatmaps are Gaussian peaks and can be warped by displacement fields via Eq. 1.
    Section 3.2; the reconstruction is a differentiable proxy for true person motion.
  • ad hoc to paper Auxiliary losses L_fb and L_se break the multiple-minima degeneracy of L_mot and select true motion.
    Section 4.4 and Appendix A.2; empirically supported by ablations but not proven.
  • domain assumption Frozen YOLOX detections plus optional OSNet features are a sufficient input representation for motion learning.
    Section 3.4.1 and Appendix A.4.
  • domain assumption Ground-plane homographies for MOT17 from Dendorfer et al. are accurate enough for training and association.
    Section 4.1.1 and appendix.

how reviews work

0 comments
Cite this review

Pith. "Pith review of No Identity, no problem: Motion through detection for people tracking." pith.science (2026). https://pith.science/paper/R4M6Q6QA

@misc{pith2026241116466,
  author       = {Pith},
  title        = {Pith review of: No Identity, no problem: Motion through detection for people tracking},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/R4M6Q6QA}},
  note         = {Machine review of arXiv:2411.16466}
}
read the original abstract

Tracking-by-detection has become the de facto standard approach to people tracking. To increase robustness, some approaches incorporate re-identification using appearance models and regressing motion offset, which requires costly identity annotations. In this paper, we propose exploiting motion clues while providing supervision only for the detections, which is much easier to do. Our algorithm predicts detection heatmaps at two different times, along with a 2D motion estimate between the two images. It then warps one heatmap using the motion estimate and enforces consistency with the other one. This provides the required supervisory signal on the motion without the need for any motion annotations. In this manner, we couple the information obtained from different images during training and increase accuracy, especially in crowded scenes and when using low frame-rate sequences. We show that our approach delivers state-of-the-art results for single- and multi-view multi-target tracking on the MOT17 and WILDTRACK datasets.

Figures

Figures reproduced from arXiv: 2411.16466 by the authors.

Figure 1
Figure 1. Predicting human motion. Left: We use muSSP Wang et al. (2019) to link detections at different frame rates. We plot the MOTA and IDF1 metrics as a function of the frame rate. Below 3FPS, the degradation becomes severe. Right: Our model estimates a detection heatmap at time t and predicts the motion of objects between t and t + 1. The offsets are used to warp the heatmap into a prediction at time t + 1 and we enforce… view at source ↗
Figure 2
Figure 2. Details of the proposed differentiable reconstruction from motion Given a detection map at time t and an offset map capturing motion of objects between time t and time t + 1 we reconstruct the detection map at time t+ 1. Each reconstructed pixel is a weighted sum of the detection of the previous time step, the weights are derived from the distance between the reconstructed location and the expected position of the p… view at source ↗
Figure 3
Figure 3. Reconstruction weight function During the reconstruction from detection and offset, the contributions of previous detections are reweighted using the function plotted above. With a value of λr = 0.8 a location 2 pixels away from the reconstructed location (distance accounting for motion offset) has a weight of 1 and is fully added to the reconstruction at that location. By varying λr we control the trade-off between… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Network Architectures. Our approach to motion prediction is flexible and usable in conjunction with various detectors/trackers. Trainable components have a blue background, while static modules have a white background. Top Left: Single-view setup where the detector’s s…
Figure 5
Figure 5. Figure 5: Tracking results in low FPS scenario. In low frame rate scenarios our model outperforms the ByteTrack baseline in term of MOTA. The lower the frame rate the higher the performance gap, showing the benefit of the pro￾posed approach. when the input/output resolutions inc…
Figure 6
Figure 6. Figure 6: Visualizing the Motion Offsets. We visualize the detection and the motion offset predicted by our model on the left and the ground truth on the right. The green squares mark the position of the person at time t and the blue square the position at time t + 1. The green …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

55 extracted references · 53 canonical work pages

  1. [1]

    Andriluka, S

    M. Andriluka, S. Roth, and B. Schiele. People-Tracking-By-Detection and People-Detection-By- Tracking. InConference on Computer Vision and Pattern Recognition, June 2008

  2. [2]

    Baqué, F

    P. Baqué, F. Fleuret, and P. Fua. Deep Occlusion Reasoning for Multi-Camera Multi-Target Detection. In International Conference on Computer Vision, 2017

  3. [3]

    Benfold and I

    B. Benfold and I. Reid. Stable Multi-Target Tracking in Real-Time Surveillance Video. InConference on Computer Vision and Pattern Recognition, 2011

  4. [4]

    BenShitrit, J

    H. BenShitrit, J. Berclaz, F. Fleuret, and P. Fua. Multi-Commodity Network Flow for Tracking Multiple People. IEEE Transactions on Pattern Analysis and Machine Intelligence, 36(8):1614–1627, 2014

  5. [5]

    Berclaz, F

    J. Berclaz, F. Fleuret, and P. Fua. Robust People Tracking with Global Trajectory Optimization. In Conference on Computer Vision and Pattern Recognition, 2006

  6. [6]

    Berclaz, F

    J. Berclaz, F. Fleuret, E. Türetken, and P. Fua. Multiple Object Tracking Using K-Shortest Paths Optimization. IEEE Transactions on Pattern Analysis and Machine Intelligence, 33(11):1806–1819, 2011

  7. [7]

    Bergmann, T

    P. Bergmann, T. Meinhardt, and L. Leal-Taixe. Tracking Without Bells and Whistles. InConference on Computer Vision and Pattern Recognition, 2019

  8. [8]

    Bewley, Z

    A. Bewley, Z. Ge, L. Ott, F. Ramos, and B. Upcroft. Simple Online and Realtime Tracking. In International Conference on Image Processing, 2016

Show all 55 references
  1. [9]

    Breitenstein, F

    M.D. Breitenstein, F. Reichlin, B. Leibe, E. Koller-Meier, and L.Van Gool. Robust Tracking-By- Detection Using a Detector Confidence Particle Filter. InInternationalConference on Computer Vision, pp. 1515–1522, 2009

  2. [10]

    Chavdarova, P

    T. Chavdarova, P. Baqué, S. Bouquet, A. Maksai, C. Jose, L. Lettry, P. Fua, L. Van Gool, and F. Fleuret. The Wildtrack Multi-Camera Person Dataset. In Conference on Computer Vision and Pattern Recognition, 2018

  3. [11]

    MMDetection: Open MMLab detection toolbox and benchmark

    Kai Chen, Jiaqi Wang, Jiangmiao Pang, Yuhang Cao, Yu Xiong, Xiaoxiao Li, Shuyang Sun, Wansen Feng, Ziwei Liu, Jiarui Xu, Zheng Zhang, Dazhi Cheng, Chenchen Zhu, Tianheng Cheng, Qijie Zhao, Buyu Li, Xin Lu, Rui Zhu, Yue Wu, Jifeng Dai, Jingdong Wang, Jianping Shi, Wanli Ouyang,...

  4. [12]

    Cheng, M.-X

    C.-C. Cheng, M.-X. Qiu, C.-K. Chiang, and S.-H. Lai. Rest: A Reconfigurable Spatial-Temporal Graph Model for Multi-Camera Multi-Object Tracking. InInternationalConference on Computer Vision, 2023

  5. [13]

    Ciaparrone, F

    G. Ciaparrone, F. L. Sánchez, S. Tabik, L. Troiano, R. Tagliaferri, and F. Herrera. Deep Learning in Video Multi-Object Tracking: A Survey.Neurocomputing, 2020

  6. [14]

    Dendorfer, V

    P. Dendorfer, V. Yugay, A. Osep, and L. Leal-Taixé. Quo Vadis: Is Trajectory Forecasting the Key Towards Long-Term Multi-Object Tracking?, 2022

  7. [15]

    Dosovitskiy, P

    A. Dosovitskiy, P. Fischer, E. Ilg, P. Hausser, C. Hazırbas, V. Golkov, P. Smagt, D. Cremers, and T. Brox. Flownet: Learning Optical Flow with Convolutional Networks. InInternational Conference on Computer Vision, 2015

  8. [16]

    Engilberge, W

    M. Engilberge, W. Liu, and P. Fua. Multi-View Tracking Using Weakly Supervised Human Motion Prediction. In IEEE Winter Conference on Applications of Computer Vision, 2023

  9. [17]

    Engilberge, H

    M. Engilberge, H. Shi, Z. Wang, and P. Fua. Two-Level Data Augmentation for Calibrated Multi-View Detection. In IEEE Winter Conference on Applications of Computer Vision, 2023. 13

  10. [18]

    Felzenszwalb, R.B

    P.F. Felzenszwalb, R.B. Girshick, D. McAllester, and D. Ramanan. Object Detection with Discrimina- tively Trained Part Based Models.IEEE Transactions on Pattern Analysis and Machine Intelligence, 32(9):1627–1645, 2010

  11. [19]

    Fleuret, J

    F. Fleuret, J. Berclaz, R. Lengagne, and P. Fua. Multi-Camera People Tracking with a Probabilistic Occupancy Map. IEEE Transactions on Pattern Analysis and Machine Intelligence, 30(2):267–282, February 2008

  12. [20]

    YOLOX: Exceeding YOLO Series in 2021, 2021

    Zheng Ge, Songtao Liu, Feng Wang, Zeming Li, and Jian Sun. YOLOX: Exceeding YOLO Series in 2021, 2021

  13. [21]

    Girshick

    R. Girshick. Fast R-CNN. InInternational Conference on Computer Vision, 2015

  14. [22]

    K. He, X. Zhang, S. Ren, and J. Sun. Deep Residual Learning for Image Recognition. InConference on Computer Vision and Pattern Recognition, pp. 770–778, 2016

  15. [23]

    Hou and L

    Y. Hou and L. Zheng. Multiview Detection with Shadow Transformer (And View-Coherent Data Aug- mentation). In Proceedings of the 29th ACM International Conference on Multimedia, pp. 1673–1682, 2021

  16. [24]

    Y. Hou, L. Zheng, and S. Gould. Multiview Detection with Feature Perspective Transformation. In European Conference on Computer Vision, pp. 1–18, 2020

  17. [25]

    Ioffe and C

    S. Ioffe and C. Szegedy. Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift. InInternational Conference on Machine Learning, 2015

  18. [26]

    Jiang, S

    H. Jiang, S. Fels, and J.J. Little. A Linear Programming Approach for Multiple Object Tracking. In Conference on Computer Vision and Pattern Recognition, pp. 1–8, June 2007

  19. [27]

    Kasturi, D

    R. Kasturi, D. Goldgof, P. Soundararajan, V. Manohar, J. Garofolo, M. Boonstra, V. Korzhova, and J. Zhang. Framework for Performance Evaluation of Face, Text, and Vehicle Detection and Tracking in Video: Data, Metrics, and Protocol.IEEE Transactionson Pattern Analysis and Mach...

  20. [28]

    D. P. Kingma and J. Ba. Adam: A Method for Stochastic Optimization. InInternational Conference on Learning Representations, 2015

  21. [29]

    P. Lenz, A. Geiger, and R. Urtasun. Followme: Efficient Online Min-Cost Flow Tracking with Bounded Memory and Computation. InInternational Conference on Computer Vision, pp. 4364–4372, December 2015

  22. [30]

    W. Liu, M. Salzmann, and P. Fua. Estimating People Flows to Better Count Them in Crowded Scenes. In European Conference on Computer Vision, 2020

  23. [31]

    W. Liu, N. Durasov, and P. Fua. Leveraging Self-Supervision for Cross-Domain Crowd Counting. In Conference on Computer Vision and Pattern Recognition, 2022

  24. [32]

    Milan, L

    A. Milan, L. Leal-Taixe, I. Reid, S. Roth, and K. Schindler. Mot16: A Benchmark for Multi-Object Tracking. InarXiv Preprint, 2016

  25. [33]

    Nair and G

    V. Nair and G. E. Hinton. Rectified Linear Units Improve Restricted Boltzmann Machines. In International Conference on Machine Learning, 2010

  26. [34]

    Paszke, S

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Kopf, E. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, and S. Chintala. Pytorch: An Imperative Style, High-Per...

  27. [35]

    Pirsiavash, D

    H. Pirsiavash, D. Ramanan, and C. Fowlkes. Globally-Optimal Greedy Algorithms for Tracking a Variable Number of Objects. InConference on Computer Vision and Pattern Recognition, pp. 1201– 1208, June 2011. 14

  28. [36]

    S. Ren, K. He, R. Girshick, and J. Sun. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. InAdvances in Neural Information Processing Systems, 2015

  29. [37]

    Ristani and C

    E. Ristani and C. Tomasi. Features for Multi-Target Multi-Camera Tracking and Re-Identification. In Conference on Computer Vision and Pattern Recognition, 2018

  30. [38]

    Ristani, F

    E. Ristani, F. Solera, R. Zou, R. Cucchiara, and C. Tomasi. Performance Measures and a Data Set for Multi-Target, Multi-Camera Tracking. InEuropean Conference on Computer Vision, 2016

  31. [39]

    G. Shu, A. Dehghan, O. Oreifej, E. Hand, and M. Shah. Part-Based Multiple-Person Tracking with Partial Occlusion Handling. InConference on Computer Vision and Pattern Recognition, 2012

  32. [40]

    A. W. M. Smeulders, D. M. Chu, R. Cucchiara, S. Calderara, A. Dehghan, and M. Shah. Visual Tracking: An Experimental Survey.IEEE Transactions on Pattern Analysis and Machine Intelligence, 36(7):1442–1468, July 2014

  33. [41]

    L. Song, J. Wu, M. Yang, Q. Zhang, Y. Li, and J. Yuan. Stacked Homography Transformations for Multi-View Pedestrian Detection. InConference on Computer Vision and Pattern Recognition, 2021

  34. [42]

    S. Tang, B. Andres, M. Andriluka, and B. Schiele. Subgraph Decomposition for Multi-Target Tracking. In Conference on Computer Vision and Pattern Recognition, pp. 5033–5041, 2015

  35. [43]

    Teed and J

    Z. Teed and J. Deng. RAFT: Recurrent All-Pairs Field Transforms for Optical Flow. InEuropean Conference on Computer Vision, 2020

  36. [44]

    Teepe, P

    T. Teepe, P. Wolters, J. Gilg, F. Herzog, and G. Rigoll. Earlybird: Early-Fusion for Multi-View Tracking in the Bird’s Eye View. InConference on Computer Vision and Pattern Recognition, 2024

  37. [45]

    R.Y. Tsai. A Versatile Cameras Calibration Technique for High Accuracy 3D Machine Vision Metrology Using Off-The-Shelf TV Cameras and Lenses.Journal of Robotics and Automation, 3(4):323–344, 1987

  38. [46]

    Mots: Multi-Object Tracking and Segmentation

    Paul Voigtlaender, Michael Krause, Aljosa Osep, Jonathon Luiten, Berin Balachandar Gnana Sekar, Andreas Geiger, and Bastian Leibe. Mots: Multi-Object Tracking and Segmentation. InConference on Computer Vision and Pattern Recognition, 2019

  39. [47]

    C. Wang, Y. Wang, Y. Wang, C.T. Wu, and G. Yu. muSSP: Efficient Min-Cost Flow Algorithm for Multi-Object Tracking. InAdvances in Neural Information Processing Systems, pp. 423–432, 2019

  40. [48]

    Wojke, A

    N. Wojke, A. Bewley, and D. Paulus. Simple Online and Realtime Tracking with a Deep Association Metric. In International Conference on Image Processing, 2017

  41. [49]

    Y. Xu, X. Liu, Y. Liu, and S.C. Zhu. Multi-View People Tracking via Hierarchical Trajectory Compo- sition. In Conference on Computer Vision and Pattern Recognition, pp. 4256–4265, 2016

  42. [50]

    Y. Xu, A. Osep, Y. Ban, R. Horaud, L. Leal-Taixe, and X. Alameda-Pineda. How to Train Your Deep Multi-Object Tracker. InConference on Computer Vision and Pattern Recognition, 2020

  43. [51]

    Yilmaz, O

    A. Yilmaz, O. Javed, and M. Shah. Object Tracking: A Survey. ACM Computing Surveys, 38(4), December 2006

  44. [52]

    ByteTrack: Multi-Object Tracking by Associating Every Detection Box, 2022

    Yifu Zhang, Peize Sun, Yi Jiang, Dongdong Yu, Fucheng Weng, Zehuan Yuan, Ping Luo, Wenyu Liu, and Xinggang Wang. ByteTrack: Multi-Object Tracking by Associating Every Detection Box, 2022

  45. [53]

    K. Zhou, Y. Yang, A. Cavallaro, and T. Xiang. Omni-Scale Feature Learning for Person Re- Identification. In International Conference on Computer Vision, 2019

  46. [54]

    X. Zhou, D. Wang, and P. Krähenbühl. Objects as Points. InarXiv Preprint, 2019

  47. [55]

    X. Zhou, V. Koltun, and P. Krähenbühl. Tracking Objects as Points. In European Conference on Computer Vision, 2020. 15 Appendix Appendix is organized as follows: Appendix A provides a detailed evaluation of the predicted displacement and additionnal analysis of the single-view...

Pith tools

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