Pith. sign in

REVIEW 4 major objections 8 minor 35 references

NOVA: Next-step Open-Vocabulary Autoregression for 3D Multi-Object Tracking in Autonomous Driving

T0 review · 4 major / 8 minor · reviewed 2026-07-15 · grok-4.5

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

Pith's one-line read NOVA recasts open-vocabulary 3D multi-object tracking as next-step sequence completion so a 0.5B language model can keep identities for unseen classes, raising novel AMOTA on nuScenes from 2.2% to 22.4%.

desk verdict Solid OV-3D association recipe with real novel-category gains on a 0.5B model; the big AMOTA jump is only partly isolated from the detector. read the letter →

arxiv 2603.06254 v2 pith:GBKSPFC5 submitted 2026-03-06 cs.CV cs.ROeess.IV

Kai Luo , Xu Wang , Rui Fan , Kailun Yang This is my paper
classification cs.CVcs.ROeess.IV
keywords open-vocabulary3Dmulti-objecttrackingautoregressiveassociationtrajectorysequencemodelinggeometryencoderhybridpromptinghardnegativeminingautonomousdrivingLLMfor
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

Real-world driving scenes keep introducing objects outside any fixed label list, so conventional 3D trackers fail when detectors suppress unknowns and association relies on closed-set distance rules. This paper claims that trajectories can instead be treated as short spatio-temporal semantic sequences and that a small language model can decide matches by next-token prediction over that context. Geometry is injected as continuous embeddings rather than brittle number strings, novel class names are masked as Unknown during training so the model learns to fall back on motion and shape, and hard negatives force fine discrimination among nearby objects. On nuScenes, V2X-Seq-SPD, and KITTI the approach yields large gains on novel categories while remaining competitive on base classes, all with a 0.5B backbone. A sympathetic reader cares because open-world perception for planning needs stable identities for objects the system has never been told to expect.

What carries the argument

Next-step Open-Vocabulary Autoregression (NOVA): each track–detection pair is serialized into a prompt of recent boxes and a candidate; a lightweight LLM emits a Yes/No decision token whose probability becomes the association score. Geometry Encoder embeddings, Hybrid Prompting (base names kept, novel names replaced by Unknown), and hard negative mining supply the inductive bias that makes the decision reliable under open-vocabulary noise.

What would settle it

Replace the open-vocabulary detector with one that produces systematically mislocalized or empty novel boxes on the same nuScenes Novel split and re-run the identical NOVA association pipeline; if Novel AMOTA collapses back near the prior baseline, the claim that trajectory-conditioned next-token scoring is the primary source of the gain is falsified.

Watch

Extended reading notes

Core claim

The central claim is that online open-vocabulary 3D multi-object association is better solved as autoregressive next-token prediction over serialized trajectory context than as hand-crafted geometric or semantic cost matching. With a Geometry Encoder that maps 3D boxes into the model embedding space plus IoU-quality supervision, Hybrid Prompting that masks novel labels, and hard negative mining, a compact 0.5B model produces association scores that substantially improve novel-category tracking—most notably lifting nuScenes Novel AMOTA from 2.20% to 22.41%—while remaining detector-agnostic across the evaluated benchmarks.

Load-bearing premise

The method assumes upstream open-vocabulary 3D detectors already supply usable novel-class boxes, and that training with those labels masked as Unknown plus a short motion history is enough for the language model to recover correct identities under the same detection noise at test time.

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 / 8 minor

Summary. NOVA reformulates open-vocabulary 3D multi-object tracking (OV-3D-MOT) association as autoregressive next-token prediction over serialized trajectory context. A lightweight LLM (default Qwen2.5-0.5B) scores track–detection pairs by the probability of generating “Yes,” using a Geometry Encoder that injects continuous 3D box features via a <box> token (with IoU-quality auxiliary loss), Hybrid Prompting that keeps base class names but masks novel labels as “Unknown,” and hard-negative mining of spatially proximate identity-inconsistent candidates. Online inference builds a cost matrix from these scores and applies Hungarian matching with standard birth/death logic. Experiments on nuScenes, V2X-Seq-SPD, and KITTI report large novel-category gains versus Open3DTrack—most notably nuScenes Novel AMOTA 22.41% vs 2.20%—with ablations on LLM size, geometry encoding, prompting, sampling, history length, and epochs.

Significance. OV-3D-MOT is still nascent; moving association from hand-crafted geometric costs toward a learned, trajectory-conditioned policy is a timely and useful direction for open-world driving perception. If the reported novel-category gains hold under controlled isolation of the association module, the work would be a solid empirical contribution: multi-benchmark evaluation, systematic ablations, a compact 0.5B backbone with reported FPS, and promised code. The Geometry Encoder + IoU quality head and Hybrid Prompting are concrete, transferable design choices even if the “LLM commonsense” framing is secondary to the binary association scorer. The paper does not claim machine-checked proofs or parameter-free theory; its value is systems evidence for generative association under open-vocabulary detections.

major comments (4)
  1. [Table I, §IV.B, Fig. 2] Table I / §IV.B: The central attribution of the +20.21 Novel AMOTA gain (2.20→22.41 on nuScenes) to autoregressive association is not fully isolated. Although both methods are listed under Find n’ Propagate on nuScenes, Open3DTrack is defined (§II, Fig. 2) as 2D-OV semantics on closed-set 3D proposals with conventional cost matching. The manuscript does not report (i) a standard non-LLM association baseline (e.g., 3D IoU/distance/Mahalanobis + Hungarian, or AB3DMOT-style) on exactly the same OV boxes NOVA receives, nor (ii) NOVA’s scorer on the proposal stream Open3DTrack originally used. Without that swap, detector/proposal quality and association formulation remain partially confounded, and Hybrid Prompting / L=3 ablations (Tables IV, VI) only validate design choices inside NOVA.
  2. [Table I, §IV.B, Abstract] Table I (V2X-Seq-SPD rows): Novel AMOTA is largely negative for NOVA (−141.78 and −129.85) while the text emphasizes sAMOTA gains and “superior performance.” Negative AMOTA typically indicates performance below the untracked baseline under the official averaging; this is load-bearing for the open-vocabulary claim and is not explained in §IV.B or the conclusion. Please (a) state when AMOTA vs sAMOTA is the primary metric and why, (b) discuss what drives Novel AMOTA collapse (recall, confidence calibration, score thresholds, ID switches), and (c) temper abstract/intro claims of broad superiority where Novel AMOTA fails.
  3. [Abstract, §I, §III.B Hybrid Prompting, Eq. (2), Table II] Abstract / §I / §III.B: The narrative that the model “utilizes the hierarchical structure of language space” and “high-level commonsense reasoning” overstates the implemented mechanism. Hybrid Prompting masks novel names as “Unknown,” the decision is a binary Yes/No token (Eq. 2), and Table II shows a 0.5B model outperforming larger LLMs—consistent with learning geometric/motion discrimination more than deep linguistic category reasoning. Please align claims with the actual task (trajectory-conditioned match scoring) and clarify what, if any, language priors remain after novel-label masking; otherwise the contribution is better framed as learned geometric association with optional base-class text, not open-vocabulary linguistic reasoning.
  4. [§IV.B, Table I KITTI] §IV.A–B / Table I (KITTI): Open3DTrack is omitted on KITTI as “degenerate,” so the only multi-method comparison is incomplete on a third benchmark. Given that the paper’s strongest claim is cross-dataset OV-3D-MOT superiority, either provide at least one conventional association baseline on the same Find n’ Propagate (+ GroundingDINO/YOLO-World) detections for KITTI, or clearly restrict the SOTA claim to settings where the baseline is reported and non-degenerate.
minor comments (8)
  1. [Title, headers, Fig. 1] Throughout: the title/name is inconsistently rendered as “NOV A” (space) vs “NOVA”; unify for searchability and typesetting.
  2. [§III.B Geometry Encoder, Eq. (3)] Eq. (3): f_raw includes vol=lwh which is deterministically redundant with (l,w,h); a one-line justification or ablation note would help.
  3. [§III.B, Eq. (4)] Eq. (4): λ_reg is never specified numerically; please report the value used and sensitivity if any.
  4. [Table I nuScenes] Table I: AMOTP for NOVA Base on nuScenes rises to 100.09 from 63.70 (worse if lower-is-better); briefly discuss the precision trade-off when claiming overall superiority.
  5. [Fig. 4] Fig. 4: failure callouts are informative but dense; consider marking which objects are Novel vs Base and ensuring IDs are legible in print.
  6. [§IV.A Datasets and Metrics] §IV.A: Base/Novel splits differ across datasets (e.g., Pedestrian is Base on nuScenes/V2X but Novel on KITTI); a short rationale table would reduce confusion when comparing Novel metrics across rows.
  7. [§II Autoregressive Modeling] Related work: AR-MOT and recent OV tracking citations are present; a clearer sentence on how binary Yes/No association differs from full sequence-generation MOT would help position the method.
  8. [§IV.A Implementation Details] Implementation: LoRA rank, learning rate, batch construction (pairs per step), and track age K are underspecified relative to reproducibility claims; add to §IV.A or appendix.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: empirical tracking system with learned association scores evaluated on held-out standard MOT metrics.

full rationale

NOVA is a systems/ML paper: it trains an LLM association scorer and reports AMOTA/sAMOTA/MOTA on nuScenes, V2X-Seq-SPD, and KITTI under fixed Base/Novel splits. The association score pi,j = Pr(Yes|P(Ti,bj_t)) is a learned model output, not a quantity defined from the evaluation metrics or from parameters fitted to those same metrics. Geometry Encoder, Hybrid Prompting, hard negatives, history length L, and epoch count are design choices and hyperparameters ablated on the same benchmarks; none of the reported gains reduce by construction to the training objective or to a self-cited uniqueness theorem. Citations (Open3DTrack, Find n' Propagate, Qwen, etc.) are external baselines or tools, not load-bearing self-proofs. Concerns about detector confounds or attribution of gains to association vs. proposals are experimental-isolation issues, not circular derivation. Derivation chain is self-contained against external benchmarks; steps empty by design.

Assumptions & free parameters 6 free parameters · 5 assumptions · 3 invented entities

As a tracking-methods paper, the load-bearing content is mostly engineering assumptions and fitted training/inference knobs rather than physical axioms. The central claim rests on the tracking-by-detection setup, the usability of particular open-vocabulary 3D detectors, the validity of yes-probability costs under Hungarian matching, and several hand-chosen training choices (history length, epochs, novel masking, hard-negative locality). Invented modules are the geometry embedding path and hybrid prompting policy; they are algorithmic constructs with only in-paper ablation support.

free parameters (6)
  • inference history length L
    Default L=3 chosen by ablation; performance peaks there and drops at L=5, so reported gains depend on this hand-selected context length.
  • training epochs
    Default 8 epochs selected because novel sAMOTA peaks there; longer training hurts novel generalization in their table.
  • IoU-quality loss weight λ_reg
    Balances decision-token CE and IoU MSE; value not fully specified but required for the auxiliary head that supports geometry learning.
  • track age threshold K
    Controls how long unmatched tracks survive occlusion; directly affects birth/death lifecycle and MOT metrics.
  • association 3D IoU evaluation threshold 0.25
    Unified matching threshold used for all categories/metrics; changes would rescale reported (s)AMOTA/MOTA.
  • hard-negative spatial selection / positive jitter magnitude
    Mining and augmentation policy is decisive in ablations but only qualitatively specified; these sampling radii/noise scales are free design choices.
assumptions (5)
  • domain assumption Online tracking-by-detection with one-to-one bipartite matching is an adequate formulation for OV-3D-MOT association.
    Problem formulation and inference (§III.A, §III.C) assume detections are given and Hungarian assignment on 1−p costs yields identities.
  • domain assumption Open-vocabulary 3D detectors used (Find n’ Propagate ± GroundingDINO/YOLO-World) produce novel-class boxes usable enough that association, not proposal generation, is the main bottleneck.
    All experiments freeze/reuse these detectors; NOVA does not redesign detection (§III intro, §IV.A).
  • ad hoc to paper Masking novel class names as “Unknown” during training teaches class-agnostic association without discarding necessary identity cues.
    Hybrid Prompting is a paper-specific training policy justified by ablation Table IV, not a standard MOT law.
  • ad hoc to paper Probability of generating the token “Yes” is a calibrated enough match affinity for cost-based assignment across base and novel classes.
    Association score definition p_i,j = Pr(Yes|prompt) is taken as the sole learned cost (§III.B).
  • domain assumption Standard MOT metrics at fixed IoU, with author-defined Base/Novel splits, measure open-vocabulary tracking success.
    Evaluation protocol in §IV.A; split definitions are dataset-specific conventions used to support the central claim.
invented entities (3)
  • Geometry Encoder with <box> embedding injection
    purpose: Map continuous 3D box features into the LLM embedding space so association is not limited to brittle numeric text tokens.
    Introduced in §III.B; supported only by in-paper ablations (Table III), not independent external evidence.
  • Hybrid Prompting (base names kept, novel names → Unknown)
    purpose: Reduce semantic overfitting to base labels and train under novel-label uncertainty.
    Core paper-specific mechanism; gains shown in Table IV only within this study.
  • NOVA autoregressive association policy
    purpose: Replace hand-crafted similarity costs with LLM next-token match decisions over trajectory context.
    The overall method entity; evaluated only in this paper’s experiments against Open3DTrack-style baselines.

how reviews work

0 comments
Cite this review

Pith. "Pith review of NOVA: Next-step Open-Vocabulary Autoregression for 3D Multi-Object Tracking in Autonomous Driving." pith.science (2026). https://pith.science/paper/GBKSPFC5

@misc{pith2026260306254,
  author       = {Pith},
  title        = {Pith review of: NOVA: Next-step Open-Vocabulary Autoregression for 3D Multi-Object Tracking in Autonomous Driving},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GBKSPFC5}},
  note         = {Machine review of arXiv:2603.06254}
}
read the original abstract

Generalizing across unknown targets is critical for open-world perception, yet existing 3D Multi-Object Tracking (3D MOT) pipelines remain limited by closed-set assumptions and ``semantic-blind'' heuristics. To address this, we propose Next-step Open-Vocabulary Autoregression (NOVA), an autoregressive association formulation that shifts the data association stage from fragmented distance-based matching toward trajectory-conditioned spatio-semantic modeling. NOVA reformulates 3D trajectories as structured spatio-temporal semantic sequences, enabling the simultaneous encoding of physical motion continuity and deep linguistic priors. By leveraging the autoregressive capabilities of Large Language Models (LLMs), we transform the tracking task into a principled process of next-step sequence completion. This mechanism allows the model to explicitly utilize the hierarchical structure of language space to resolve fine-grained semantic ambiguities and maintain identity consistency across complex long-range sequences through high-level commonsense reasoning. Extensive experiments on nuScenes, V2X-Seq-SPD, and KITTI demonstrate the superior performance of NOVA. Notably, on the nuScenes dataset, NOVA achieves an AMOTA of 22.41% for Novel categories, yielding a significant 20.21% absolute improvement over the baseline. These gains are realized through a compact 0.5B autoregressive model. Code will be available at https://github.com/xifen523/NOVA.

Figures

Figures reproduced from arXiv: 2603.06254 by the authors.

Figure 1
Figure 1. Overview and performance of our proposed Next-step Open-Vocabulary Autoregression (NOVA) framework for 3D [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Comparison of 3D MOT paradigms under open [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The pipeline of the proposed NOVA framework. (1) Open-Vocabulary 3D Detection: Multi-modal inputs are processed to generate detections Dt with open-vocabulary labels. (2) Serialization & Hybrid Prompting: A Geometry Encoder projects raw box features fraw into embeddings Egeo. These are interleaved with text using a Hybrid Prompting strategy that explicitly masks novel class labels (e.g., Unknown) to enforce geometri… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison of OV-3D-MOT across datasets. Four representative scenes from different autonomous￾driving benchmarks, each visualized over four consecutive frames (t=1∼4). For each scene, we compare Open3DTrack [1] (top) with NOVA (bottom). Callouts and arrows …

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

35 extracted references · 4 linked inside Pith

  1. [1]

    Open3DTrack: Towards open-vocabulary 3D multi- object tracking,

    A. Ishaqet al., “Open3DTrack: Towards open-vocabulary 3D multi- object tracking,” inProc. ICRA, 2025, pp. 787–793

    2025

  2. [2]

    nuScenes: A multimodal dataset for autonomous driving,

    H. Caesaret al., “nuScenes: A multimodal dataset for autonomous driving,” inProc. CVPR, 2020, pp. 11 618–11 628

    2020

  3. [3]

    V2X-Seq: A large-scale sequential dataset for vehicle- infrastructure cooperative perception and forecasting,

    H. Yuet al., “V2X-Seq: A large-scale sequential dataset for vehicle- infrastructure cooperative perception and forecasting,” inProc. CVPR, 2023, pp. 5486–5495

    2023

  4. [4]

    A survey of autonomous driving: Common practices and emerging technologies,

    E. Yurtsever, J. Lambert, A. Carballo, and K. Takeda, “A survey of autonomous driving: Common practices and emerging technologies,” IEEE Access, vol. 8, pp. 58 443–58 469, 2020

    2020

  5. [5]

    AB3DMOT: A baseline for 3D multi-object tracking and new evaluation metrics,

    X. Weng, J. Wang, D. Held, and K. Kitani, “AB3DMOT: A baseline for 3D multi-object tracking and new evaluation metrics,”arXiv preprint arXiv:2008.08063, 2020

    arXiv 2008

  6. [6]

    Towards long-tailed 3D detection,

    N. Peri, A. Dave, D. Ramanan, and S. Kong, “Towards long-tailed 3D detection,” inProc. CoRL, vol. 205, 2023, pp. 1904–1915

    2023

  7. [7]

    Towards open world recognition,

    A. Bendale and T. Boult, “Towards open world recognition,” inProc. CVPR, 2015, pp. 1893–1902

    2015

  8. [8]

    Are we ready for autonomous driving? The KITTI vision benchmark suite,

    A. Geiger, P. Lenz, and R. Urtasun, “Are we ready for autonomous driving? The KITTI vision benchmark suite,” inProc. CVPR, 2012, pp. 3354–3361

    2012

Show all 35 references
  1. [9]

    Learning transferable visual models from natural language supervision,

    A. Radfordet al., “Learning transferable visual models from natural language supervision,” inProc. ICML, vol. 139, 2021, pp. 8748–8763

    2021

  2. [10]

    Grounded language-image pre-training,

    L. H. Liet al., “Grounded language-image pre-training,” inProc. CVPR, 2022, pp. 10 955–10 965

    2022

  3. [11]

    OVTrack: Open-vocabulary multiple object tracking,

    S. Li, T. Fischer, L. Ke, H. Ding, M. Danelljan, and F. Yu, “OVTrack: Open-vocabulary multiple object tracking,” inProc. CVPR, 2023, pp. 5567–5577

    2023

  4. [12]

    VOVTrack: Exploring the potentiality in videos for open-vocabulary object tracking,

    Z. Qian, R. Han, J. Hou, L. Song, and W. Feng, “VOVTrack: Exploring the potentiality in videos for open-vocabulary object tracking,”arXiv preprint arXiv:2410.08529, 2024

    2024 arXiv

  5. [13]

    Center-based 3D object detection and tracking,

    T. Yin, X. Zhou, and P. Krähenbühl, “Center-based 3D object detection and tracking,” inProc. CVPR, 2021, pp. 11 784–11 793

    2021

  6. [14]

    EagerMOT: 3D multi-object tracking via sensor fusion,

    A. Kim, A. Osep, and L. Leal-Taixé, “EagerMOT: 3D multi-object tracking via sensor fusion,” inProc. ICRA, 2021, pp. 11 315–11 321

    2021

  7. [15]

    MCTrack: A unified 3D multi-object tracking framework for autonomous driving,

    X. Wanget al., “MCTrack: A unified 3D multi-object tracking framework for autonomous driving,” inProc. IROS, 2025, pp. 4551– 4558

    2025

  8. [16]

    PointPillars: Fast encoders for object detection from point clouds,

    A. H. Lang, S. V ora, H. Caesar, L. Zhou, J. Yang, and O. Beijbom, “PointPillars: Fast encoders for object detection from point clouds,” in Proc. CVPR, 2019, pp. 12 697–12 705

    2019

  9. [17]

    PV-RCNN: Point-voxel feature set abstraction for 3D object detection,

    S. Shiet al., “PV-RCNN: Point-voxel feature set abstraction for 3D object detection,” inProc. CVPR, 2020, pp. 10 526–10 535

    2020

  10. [18]

    Grounding DINO: Marrying DINO with grounded pre-training for open-set object detection,

    S. Liuet al., “Grounding DINO: Marrying DINO with grounded pre-training for open-set object detection,” inProc. ECCV, 2024, pp. 38–55

    2024

  11. [19]

    YOLO- World: Real-time open-vocabulary object detection,

    T. Cheng, L. Song, Y . Ge, W. Liu, X. Wang, and Y . Shan, “YOLO- World: Real-time open-vocabulary object detection,” inProc. CVPR, 2024, pp. 16 901–16 911

    2024

  12. [20]

    PointCLIP: Point cloud understanding by CLIP,

    R. Zhanget al., “PointCLIP: Point cloud understanding by CLIP,” in Proc. CVPR, 2022, pp. 8542–8552

    2022

  13. [21]

    OpenScene: 3D scene understanding with open vocabularies,

    S. Peng, K. Genova, C. M. Jiang, A. Tagliasacchi, M. Pollefeys, and T. A. Funkhouser, “OpenScene: 3D scene understanding with open vocabularies,” inProc. CVPR, 2023, pp. 815–824

    2023

  14. [22]

    ImOV3D: Learning open vocabulary point clouds 3D object detection from only 2D images,

    T. Yang, Y . Ju, and L. Yi, “ImOV3D: Learning open vocabulary point clouds 3D object detection from only 2D images,” inProc. NeurIPS, 2024

    2024

  15. [23]

    FM-OV3D: Foundation model-based cross-modal knowledge blending for open-vocabulary 3D detection,

    D. Zhanget al., “FM-OV3D: Foundation model-based cross-modal knowledge blending for open-vocabulary 3D detection,” inProc. AAAI, 2024, pp. 16 723–16 731

    2024

  16. [24]

    Find n’ propagate: Open-vocabulary 3D object detection in urban environments,

    D. Etchegaray, Z. Huang, T. Harada, and Y . Luo, “Find n’ propagate: Open-vocabulary 3D object detection in urban environments,” inProc. ECCV, 2024, pp. 133–151

    2024

  17. [25]

    CoDA: Collaborative novel box discovery and cross-modal alignment for open-vocabulary 3D object detection,

    Y . Cao, Z. Yihan, H. Xu, and D. Xu, “CoDA: Collaborative novel box discovery and cross-modal alignment for open-vocabulary 3D object detection,” inProc. NeurIPS, 2023, pp. 71 862–71 873

    2023

  18. [26]

    OVTR: End-to-end open-vocabulary multiple object tracking with transformer,

    J. Li, E. Yu, S. Chen, and W. Tao, “OVTR: End-to-end open-vocabulary multiple object tracking with transformer,” inProc. ICLR, 2025

    2025

  19. [27]

    Monocular multi-object 3D visual language tracking,

    H. Weiet al., “Monocular multi-object 3D visual language tracking,” IEEE Transactions on Image Processing, 2026

    2026

  20. [28]

    Class-balanced grouping and sampling for point cloud 3D object detection,

    B. Zhu, Z. Jiang, X. Zhou, Z. Li, and G. Yu, “Class-balanced grouping and sampling for point cloud 3D object detection,”arXiv preprint arXiv:1908.09492, 2019

    1908 arXiv

  21. [29]

    BEVFusion: Multi-task multi-sensor fusion with unified bird’s-eye view representation,

    Z. Liuet al., “BEVFusion: Multi-task multi-sensor fusion with unified bird’s-eye view representation,” inProc. ICRA, 2022, pp. 2774–2781

    2022

  22. [30]

    Recurrent autoregressive networks for online multi-object tracking,

    K. Fang, Y . Xiang, X. Li, and S. Savarese, “Recurrent autoregressive networks for online multi-object tracking,” inProc. WACV, 2018, pp. 466–475

    2018

  23. [31]

    TrackFormer: Multi-object tracking with transformers,

    T. Meinhardt, A. Kirillov, L. Leal-Taixé, and C. Feichtenhofer, “TrackFormer: Multi-object tracking with transformers,” inProc. CVPR, 2022, pp. 8834–8844

    2022

  24. [32]

    3DMOT- Former: Graph transformer for online 3D multi-object tracking,

    S. Ding, E. Rehder, L. Schneider, M. Cordts, and J. Gall, “3DMOT- Former: Graph transformer for online 3D multi-object tracking,” in Proc. ICCV, 2023, pp. 9750–9760

    2023

  25. [33]

    TrackFusion: Enhancing multi-object tracking with temporal trajectory modeling and frame-integrated detection,

    X. Zhanget al., “TrackFusion: Enhancing multi-object tracking with temporal trajectory modeling and frame-integrated detection,” inProc. ICASSP, 2025, pp. 1–5

    2025

  26. [34]

    AR-MOT: Autoregressive multi-object tracking,

    L. Jia, Y . Wu, B. Ran, Y . Wang, L. Wang, and H. Lu, “AR-MOT: Autoregressive multi-object tracking,”arXiv preprint arXiv:2601.01925, 2026

    2026

  27. [35]

    Tracking meets large multimodal models for driving scenario understanding,

    A. Ishaq, J. Lahoud, F. S. Khan, S. Khan, H. Cholakkal, and R. M. Anwer, “Tracking meets large multimodal models for driving scenario understanding,”arXiv preprint arXiv:2503.14498, 2025

    2025 arXiv

Pith tools

Reviewed July 15, 2026 · model on record in the stance chip above.