Pith. sign in

REVIEW 5 major objections 5 minor 45 references

DIMM: Decoupled Multi-hierarchy Kalman Filter for 3D Object Tracking

T0 review · 5 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Fusing per-direction Kalman estimates with learned weights cuts 3D tracking error by 31–99 percent.

desk verdict A promising decoupled IMM synthesis undercut by a wrong normalization in the fusion equation and unsupported headline numbers. read the letter →

arxiv 2505.12340 v1 pith:QB3KWNVA submitted 2025-05-18 cs.CV

Jirong Zha , Yuxuan Fan , Kai Li , Han Li , Chen Gao , Xinlei Chen , Yong Li This is my paper · ORCID
classification cs.CV
keywords 3DobjecttrackingstateestimationinteractingmultiplemodelKalmanfilterreinforcementlearningTD3decoupledmulti-hierarchybankmaneuveringtarget
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

The paper introduces DIMM, a hybrid state-estimation framework for 3D object tracking when the target's motion dynamics are unknown and switch rapidly. It replaces the standard interacting multiple model (IMM) trick of weighting whole state vectors with a per-direction scheme: a bank of three linear Kalman filters — constant velocity, constant acceleration, and constant jerk — runs separately along x, y, and z, and a learned network assigns each model a separate fusion weight in each direction. The authors argue this expands the model-combination solution space from a hyperplane to a hypercube and replaces observation-likelihood weights with data-driven weights, and they report 31.61%–99.23% lower estimation error than seven baselines on driving, multi-model, and drone-trajectory datasets. If the claim holds, DIMM offers a practical hybrid estimator for highly maneuverable targets that needs only noisy position measurements.

What carries the argument

Two modules carry the argument. The decoupled multi-hierarchy filter bank (DHFB) runs three linear Kalman filters — CV, CA, and CJ — on each axis separately, so each filter produces a per-direction position estimate. The differentiable adaptive fusion network (DAFN) is an attention-based TD3 agent that consumes a window of noisy measurements plus the filter estimates and emits, for each model $i$, a diagonal transformation matrix $T^i_k = \mathrm{diag}(w^i_{k,x}, w^i_{k,y}, w^i_{k,z})$, with weights normalized by softmax to $[0,1]$. The fusion rule is $\hat p_k = \sum_i T^i_k \hat p^i_k$. The claimed geometric effect is stated as Proposition 1: vector-level weighting confines IMM's combination to a hyperplane, while per-axis weight matrices span a hypercube, enlarging the feasible solution space. A hierarchical reward — the difference between DIMM's localization error and that of a non-learning IMM benchmark — stabilizes the RL training signal.

What would settle it

Apply DIMM to a curated trajectory of coordinated constant-speed turns, where x-y acceleration is coupled and no constant-turn model is present in the bank; if DIMM's MSE does not beat an IMM that includes a constant-turn model on that trajectory, the per-axis decoupling claim is not supported.

Watch

Extended reading notes

Core claim

DIMM's central claim is that the accuracy ceiling of IMM-style tracking for maneuvering 3D targets comes from two fixable design choices: fusing estimates at the level of whole state vectors, and deriving combination weights from observation likelihood alone. The paper's solution is to decouple the motion models per spatial axis and to learn the fusion weights. Concretely, a multi-hierarchy filter bank runs Kalman filters for the constant-velocity, constant-acceleration, and constant-jerk models on each coordinate independently, and an attention-based TD3 reinforcement-learning network outputs a diagonal transformation matrix per model, with softmax-normalized entries in [0,1]. The fused position is the sum of model estimates weighted by these matrices. The paper claims this per-axis linear representation captures nonlinear 3D motion better than the constant-turn models used in conventional IMM, and that learned weights are more reliable than likelihood-based weights under measurement uncertainty; the reported consequence is a large reduction in MSE and MAE across all three evaluation datasets.

Load-bearing premise

The load-bearing premise is that the target's 3D motion is separable into three independent linear models — constant velocity, constant acceleration, and constant jerk — applied axis by axis, so that independent per-direction weights can represent coupled maneuvers such as turns and emergency stops.

Editorial extensions

If this is right

  • Per-axis model weighting becomes the natural design for IMM-style trackers, since the paper proves vector-level combination sits on a lower-dimensional solution space.
  • Learning-based fusion weights can supplement or replace observation-likelihood weights; the transformation matrix stays interpretable by showing which model dominates in each direction.
  • A purely linear multi-order filter bank can replace a nonlinear constant-turn model for representing highly nonlinear 3D motion, at lower computational cost.
  • The hierarchical reward (comparing against a non-learning benchmark at every step) provides a reusable way to reduce reward variance in RL-aided filtering.
  • The reported inference time of 22 ms for a batch of 256 on one GPU indicates the accuracy gain is compatible with real-time tracking workloads.

Reading between the lines

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

  • The per-axis decoupling suggests a boundary condition worth testing: trajectories with strong cross-axis kinematic coupling, such as coordinated turns, may require coupling terms across axes; a natural extension is to learn block-diagonal or sparse coupling matrices alongside the per-axis weights.
  • Because the fusion weights are learned on particular trajectory statistics, one can probe generalization by measuring how DIMM's margin over IMM degrades when test acceleration distributions shift away from the training distribution.
  • The reward is defined relative to a fixed non-learning IMM benchmark; replacing that benchmark with a stronger learned filter would show whether the policy's advantage comes from the decoupled representation itself or partly from the choice of reference estimator.
  • The reported improvement range of 31.61%–99.23% is dataset- and metric-dependent, so a practical deployment should re-measure the gain on the target system's own motion statistics rather than assume a single improvement factor.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper proposes DIMM, a hybrid 3D object tracking method that combines a decoupled multi-hierarchy Kalman filter bank (CV, CA, and CJ linear models applied separately in x, y, and z) with a differentiable adaptive fusion network based on attention-augmented TD3. The network outputs per-model, per-direction transformation matrices intended to fuse the model estimates, and the authors argue this extends the IMM combination solution space from a hyperplane to a hypercube while making the weights more reliable than observation-likelihood-based weights. The paper reports MSE and MAE results on OKF, Multi-model, and Flightmare datasets, claiming improvements of 31.61% to 99.23% over seven baselines.

Significance. If the core mechanism were correct, DIMM would be a practically useful hybrid estimator for highly maneuverable 3D targets, and the decoupled per-axis fusion idea is a reasonable direction to explore. The evaluation spans several datasets, including a realistic Flightmare quadrotor trajectory set, and the paper also reports inference efficiency. However, the significance is currently undermined by a mathematical error in the central fusion formula, missing statistical support for the headline numbers, and an apparently per-dataset selection of the action-space size. No code or appendix is provided, so the deferred proofs and experimental details cannot be checked.

major comments (5)
  1. [§4.4.3, Eq. (5)-(6)] Equation (5) normalizes the weights over the three directions for each fixed model i, since the denominator sums over j in {x,y,z}. However, Equation (6) requires, for each direction j, that the three model weights sum to one so that the fusion is a convex combination. As written, sum_{i in {cv,ca,cj}} w^i_{k,j} is not generally 1; for example, with a^{cv}_{k,x}=1 and all other actions zero, the x-axis weights sum to roughly 0.71 and the y/z-axis weights to roughly 1.09. The fused estimate in Eq. (6) is therefore not a convex combination, the diagonal matrices T^i_k can scale the estimate arbitrarily, and the claimed expansion from a hyperplane to a hypercube is not realized by the given formula. The denominator should sum over models i for each direction j, or an alternative correct normalization should be supplied; all reported results need to be re-evaluated under the corrected fusion rule.
  2. [§5.2, Table 1] Table 1 reports only point estimates averaged over 100 randomized trials, with no standard deviations, confidence intervals, or significance tests. Several differences between DIMM and the best baseline are modest (e.g., Flightmare MSE 1.4934 vs. 1.6920 for DANSE), and the abstract's 31.61%–99.23% improvement range is not directly traceable to any stated baseline/dataset pair. Please report error bars and specify the exact comparison underlying each claimed percentage.
  3. [§5.6, Table 3] No single action-space bound is best across all three datasets: the (−2,2) bound is best for OKF, while (−5,5) is best for Multi-model and Flightmare. The paper does not state which action-space row was used for the DIMM results in Table 1, so the headline results may be the best of several fitted configurations. Please state the selection protocol (e.g., a validation split) and report Table 1 numbers for a fixed, a priori action-space choice.
  4. [§4.1.1, Proposition 1 and Appendix] The proof of Proposition 1, the motion-model equations, the IMM details, the DIMM pseudocode, and the Lorenz-experiment results are all deferred to an appendix that is absent from the submitted text. Since Proposition 1 is the theoretical basis for the central hypercube claim, that claim is currently unverified in the reviewed manuscript; these materials should be included in a revision.
  5. [§4.3] The DHFB replaces the CT model with an axis-decoupled bank of CV/CA/CJ linear models, yet the paper asserts that this bank captures highly nonlinear maneuvers better than CT-based IMM without providing a proof or a targeted ablation on coupled maneuvers. Because axis-decoupled linear models do not explicitly represent rotational coupling (e.g., a constant turn), the authors should provide an analysis or an experiment showing that the decoupled linear bank is sufficient; otherwise the L1 motivation is only partially supported.
minor comments (5)
  1. [§5.2.1, Table 1] The IMM entry for the OKF dataset is a dash ('-'), but the text refers to a 'UKF-based IMM' failing; please clarify which IMM variant is reported and why the entry is missing.
  2. [Figure 8 captions] The captions call T^i_k a 'transition matrix' in two places, while the main text defines it as a 'transformation matrix'.
  3. [Author affiliations] The affiliation contains the typo 'Graudate School'; it should read 'Graduate School'.
  4. [§4.4.3, Eq. (5)] The notation ||a_{k,j}||_inf is not defined; please state explicitly that the infinity norm is taken over the model index within the three-dimensional vector a_{k,j}.
  5. [§4.1.1, Proposition 1] The phrase 'hyperplane to hypercube' is used loosely: the hypercube is for the per-direction weight matrix, not the estimate vector itself; consider clarifying the exact feasible set in the proposition statement.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the reported accuracy gains are empirical and not definitionally forced by the reward, fusion equations, or self-citations.

full rationale

Walking the derivation chain, DIMM's central claim (Table 1) is an empirical comparison, not an identity. The fused position estimate in Eq. (6) is a learned weighted sum of per-model Kalman estimates; the weights come from a policy network and are not defined in terms of the ground-truth or the reported error. The hierarchical reward in Eq. (8) uses the IMM estimate only as a training baseline: the reward equals the difference between DIMM error and IMM error, but p̂_k,AdaFuse-TD3 is not constructed from p̂_k,IMM, so outperforming IMM is optimized, not guaranteed by definition. The action-space ranges in Table 3 are hyperparameter choices reported across settings; they are not fitted parameters that make the Table 1 numbers equal to their own inputs. Proposition 1's proof is deferred to the appendix, which is an omitted proof rather than circularity. The only overlapping-author references ([43], [44]) support standard application context and a discrete-time system model and are not load-bearing for the claimed improvement. The possible normalization defect in Eq. (5) is a correctness/validity concern, not a circular reduction, because it does not make the output equivalent to an input by construction. No equation or cited result reduces the central claim to its own assumptions.

Assumptions & free parameters 4 free parameters · 5 assumptions · 2 invented entities

The central claim rests on empirical results from three datasets, with no code, no released data, and several unreported hyperparameters. The main assumptions are per-axis linear decoupling of motion, Gaussian noise, and ground-truth availability during training. The learned modules introduce no independently testable entities. The action-space selection and benchmark-relative reward are fitted choices that shape the reported improvements.

free parameters (4)
  • Action space bounds for AdaFuse-TD3 = Unreported; Table 3 tests five ranges, with best range differing per dataset
    The action space is a hyperparameter, and selecting different bounds per dataset after seeing ablation results is a fitted choice that affects reported accuracy.
  • TD3 and network hyperparameters = Unreported
    No learning rate, discount factor, batch size, network depth, or training iterations are given; these affect the learned weights and final accuracy.
  • KF noise covariances and IMM transition probabilities = Unreported
    Process and measurement noise covariances for the filter bank and the IMM baseline are not specified, so the method and baseline are not exactly reproducible.
  • Measurement window length l = Unreported
    The state s_k includes a length-l measurement window, and l is not specified; zero-padding for k < l is defined but the actual window size is missing.
assumptions (5)
  • domain assumption 3D motion can be decoupled into independent per-axis linear models (CV, CA, CJ).
    Sec 4.3 builds the filter bank on per-axis linear models; maneuvers like turns couple axes, so independence is assumed rather than proven.
  • domain assumption Gaussian process and measurement noise.
    Eq (1) assumes Gaussian noises, even though the paper motivates the learned weights by non-Gaussian noise cases.
  • domain assumption Ground truth positions are available during training for the hierarchical reward.
    Eq (8) uses p_k, the true position, in the reward; many tracking applications lack ground truth at training time.
  • ad hoc to paper IMM is a suitable benchmark for the hierarchical reward.
    The reward in Eq (8) measures improvement over a single IMM implementation; if that implementation is weak, the reported gains are inflated.
  • standard math Standard probability and linear algebra for Proposition 1.
    The hyperplane-to-hypercube argument relies on elementary convex geometry; the proof is deferred to the appendix.
invented entities (2)
  • Decoupled multi-hierarchical filter bank (DHFB)
    purpose: Runs CV, CA, and CJ Kalman filters to produce per-model estimates for per-axis fusion.
    Internal algorithmic module with no falsifiable handle outside this paper.
  • AdaFuse-TD3 (DAFN)
    purpose: Generates per-model per-direction importance weights from measurements.
    Learned artifact; its claims are supported only by the datasets reported in this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DIMM: Decoupled Multi-hierarchy Kalman Filter for 3D Object Tracking." pith.science (2026). https://pith.science/paper/QB3KWNVA

@misc{pith2026250512340,
  author       = {Pith},
  title        = {Pith review of: DIMM: Decoupled Multi-hierarchy Kalman Filter for 3D Object Tracking},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QB3KWNVA}},
  note         = {Machine review of arXiv:2505.12340}
}
read the original abstract

State estimation is challenging for 3D object tracking with high maneuverability, as the target's state transition function changes rapidly, irregularly, and is unknown to the estimator. Existing work based on interacting multiple model (IMM) achieves more accurate estimation than single-filter approaches through model combination, aligning appropriate models for different motion modes of the target object over time. However, two limitations of conventional IMM remain unsolved. First, the solution space of the model combination is constrained as the target's diverse kinematic properties in different directions are ignored. Second, the model combination weights calculated by the observation likelihood are not accurate enough due to the measurement uncertainty. In this paper, we propose a novel framework, DIMM, to effectively combine estimates from different motion models in each direction, thus increasing the 3D object tracking accuracy. First, DIMM extends the model combination solution space of conventional IMM from a hyperplane to a hypercube by designing a 3D-decoupled multi-hierarchy filter bank, which describes the target's motion with various-order linear models. Second, DIMM generates more reliable combination weight matrices through a differentiable adaptive fusion network for importance allocation rather than solely relying on the observation likelihood; it contains an attention-based twin delayed deep deterministic policy gradient (TD3) method with a hierarchical reward. Experiments demonstrate that DIMM significantly improves the tracking accuracy of existing state estimation methods by 31.61%~99.23%.

Figures

Figures reproduced from arXiv: 2505.12340 by the authors.

Figure 1
Figure 1. Illustration of 3D object tracking with unknown dy￾namics. We aim to improve the estimation accuracy by determin￾ing what type of motion model to employ and how to select and fuse the models independently of each dimension. • Planar solution space constraint (L1). The traditional IMM algorithm uses direct weighting on the filters’ 3D state estimate vectors, limiting the solution space of model combination as the obj… view at source ↗
Figure 2
Figure 2. Relationship between three kinds of methods. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Extend the combination solution space by converting the [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Overview of DIMM. Some critical technical contributions are highlighted in red. mation accuracy. According to the state vector considered, the CV, CA, and CJ model correspond to the first-order, second-order, and third-order motion model, respectively4 . Specifically, …
Figure 5
Figure 5. Figure 5: Network structure of the DAFN module. • DANSE [12] (Data-driven Nonlinear State Estimation) is the state-of-the-art model-free method. • LSTM-IMM [7] is an IMM method based on LSTM. • XGBoost-IMM [22] is a XGBoost-based IMM method. • OKF [14] is an optimized KF with pa…
Figure 6
Figure 6. Figure 6: Examples of comparison between the actual and esti [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Example of comparison between the actual and esti [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Examples of the transformation matrix of each motion [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 43 canonical work pages

  1. [1]

    Cubature Kalman filters.IEEE Transactions on Automatic Control, 54(6): 1254–1269, 2009

    Ienkaran Arasaratnam and Simon Haykin. Cubature Kalman filters.IEEE Transactions on Automatic Control, 54(6): 1254–1269, 2009. 2

    work page 2009

  2. [2]

    Recurrent Kalman networks: Factorized inference in high-dimensional deep feature spaces

    Philipp Becker, Harit Pandya, Gregor Gebhardt, Cheng Zhao, C James Taylor, and Gerhard Neumann. Recurrent Kalman networks: Factorized inference in high-dimensional deep feature spaces. InInternational Conference on Machine Learning (ICML), pages 544–552, 2019. 3, 6, 8

    work page 2019

  3. [3]

    XGBoost: A scalable tree boosting system

    Tianqi Chen and Carlos Guestrin. XGBoost: A scalable tree boosting system. InProceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 785–794, 2016. 3

    work page 2016

  4. [4]

    Split-KalmanNet: A robust model-based deep learning approach for state estimation.IEEE Transac- tions on Vehicular Technology, 72(9):12326–12331, 2023

    Geon Choi, Jeonghun Park, Nir Shlezinger, Yonina C Eldar, and Namyoon Lee. Split-KalmanNet: A robust model-based deep learning approach for state estimation.IEEE Transac- tions on Vehicular Technology, 72(9):12326–12331, 2023. 3

    work page 2023

  5. [5]

    Long short-term memory Kalman filters: Recurrent neural estimators for pose regu- larization

    Huseyin Coskun, Felix Achilles, Robert DiPietro, Nassir Navab, and Federico Tombari. Long short-term memory Kalman filters: Recurrent neural estimators for pose regu- larization. InProceedings of the IEEE International Confer- ence on Computer Vision (ICCV), pages 5524–5532, 2017. 3

    work page 2017

  6. [6]

    Normalizing Kalman filters for multi- variate time series analysis.Advances in Neural Information Processing Systems (NeurIPS), 33:2995–3007, 2020

    Emmanuel de B ´ezenac, Syama Sundar Rangapuram, Kon- stantinos Benidis, Michael Bohlke-Schneider, Richard Kurle, Lorenzo Stella, Hilaf Hasson, Patrick Gallinari, and Tim Januschowski. Normalizing Kalman filters for multi- variate time series analysis.Advances in Neural Information Processing Systems (NeurIPS), 33:2995–3007, 2020. 3

    work page 2020

  7. [7]

    Improved IMM al- gorithm based on RNNs

    Lichuan Deng, Da Li, and Ruifang Li. Improved IMM al- gorithm based on RNNs. InJournal of Physics: Conference Series, page 012055, 2020. 3, 7, 8

    work page 2020

  8. [8]

    Self-supervised 6D object pose estimation for robot manipulation

    Xinke Deng, Yu Xiang, Arsalan Mousavian, Clemens Epp- ner, Timothy Bretl, and Dieter Fox. Self-supervised 6D object pose estimation for robot manipulation. In2020 IEEE International Conference on Robotics and Automation (ICRA), pages 3665–3671, 2020. 1

    work page 2020

Show all 45 references
  1. [9]

    A disentangled recognition and nonlinear dynam- ics model for unsupervised learning.Advances in Neural Information Processing Systems (NeurIPS), 30, 2017

    Marco Fraccaro, Simon Kamronn, Ulrich Paquet, and Ole Winther. A disentangled recognition and nonlinear dynam- ics model for unsupervised learning.Advances in Neural Information Processing Systems (NeurIPS), 30, 2017. 3

    2017

  2. [10]

    Addressing function approximation error in actor-critic methods

    Scott Fujimoto, Herke Hoof, and David Meger. Addressing function approximation error in actor-critic methods. InIn- ternational Conference on Machine Learning (ICML), pages 1587–1596, 2018. 5

    2018

  3. [11]

    RL-AKF: An adaptive Kalman filter navigation algorithm based on re- inforcement learning for ground vehicles.Remote Sensing, 12(11):1704, 2020

    Xile Gao, Haiyong Luo, Bokun Ning, Fang Zhao, Linfeng Bao, Yilin Gong, Yimin Xiao, and Jinguang Jiang. RL-AKF: An adaptive Kalman filter navigation algorithm based on re- inforcement learning for ground vehicles.Remote Sensing, 12(11):1704, 2020. 6

    2020

  4. [12]

    DANSE: Data-driven non-linear state estimation of model- free process in unsupervised learning setup.IEEE Transac- tions on Signal Processing, 2024

    Anubhab Ghosh, Antoine Honor ´e, and Saikat Chatterjee. DANSE: Data-driven non-linear state estimation of model- free process in unsupervised learning setup.IEEE Transac- tions on Signal Processing, 2024. 3, 6, 7, 8

    2024

  5. [13]

    Dynamical variational autoencoders: A comprehensive review.arXiv preprint arXiv:2008.12595, 2020

    Laurent Girin, Simon Leglaive, Xiaoyu Bie, Julien Diard, Thomas Hueber, and Xavier Alameda-Pineda. Dynamical variational autoencoders: A comprehensive review.arXiv preprint arXiv:2008.12595, 2020. 3

    2008 arXiv

  6. [14]

    Opti- mization or architecture: How to hack Kalman filtering.Ad- vances in Neural Information Processing Systems (NeurIPS), 36, 2024

    Ido Greenberg, Netanel Yannay, and Shie Mannor. Opti- mization or architecture: How to hack Kalman filtering.Ad- vances in Neural Information Processing Systems (NeurIPS), 36, 2024. 3, 6, 7, 8

    2024

  7. [15]

    Long short-term memory.Neural Computation MIT-Press, 1997

    S Hochreiter. Long short-term memory.Neural Computation MIT-Press, 1997. 3, 6

    1997

  8. [16]

    Multiple pedestrian tracking from monocular videos in an interacting multiple model framework.IEEE Transactions on Image Processing, 27(3):1361–1375, 2017

    Zhengqiang Jiang and Du Q Huynh. Multiple pedestrian tracking from monocular videos in an interacting multiple model framework.IEEE Transactions on Image Processing, 27(3):1361–1375, 2017. 1

    2017

  9. [17]

    Design and comparison of mode-set adaptive IMM algorithms for ma- neuvering target tracking.IEEE Transactions on Aerospace and Electronic Systems, 35(1):343–350, 1999

    VP Jilkov, DS Angelova, and TZ A Semerdjiev. Design and comparison of mode-set adaptive IMM algorithms for ma- neuvering target tracking.IEEE Transactions on Aerospace and Electronic Systems, 35(1):343–350, 1999. 2

    1999

  10. [18]

    The new trend of state esti- mation: From model-driven to hybrid-driven methods.Sen- sors, 21(6):2085, 2021

    Xue-Bo Jin, Ruben Jonhson Robert Jeremiah, Ting-Li Su, Yu-Ting Bai, and Jian-Lei Kong. The new trend of state esti- mation: From model-driven to hybrid-driven methods.Sen- sors, 21(6):2085, 2021. 2

    2021

  11. [19]

    Sampled-Data State Estimation for LSTM.IEEE Transactions on Neural Networks and Learn- ing Systems, 2024

    Yongsik Jin and SM Lee. Sampled-Data State Estimation for LSTM.IEEE Transactions on Neural Networks and Learn- ing Systems, 2024. 3

    2024

  12. [20]

    A new approach to linear filtering and prediction problems

    Rudolph Emil Kalman. A new approach to linear filtering and prediction problems. 1960. 2, 6, 8

    1960

  13. [21]

    Structured inference networks for nonlinear state space models

    Rahul Krishnan, Uri Shalit, and David Sontag. Structured inference networks for nonlinear state space models. InPro- ceedings of the AAAI Conference on Artificial Intelligence,

  14. [22]

    Improved IMM algorithm based on XGBoost

    Da Li, Pei Zhang, and Ruifang Li. Improved IMM algorithm based on XGBoost. InJournal of Physics: Conference Se- ries, page 032017, 2021. 7, 8

    2021

  15. [23]

    Time3D: End-to-end joint monocu- lar 3D object detection and tracking for autonomous driving

    Peixuan Li and Jieyu Jin. Time3D: End-to-end joint monocu- lar 3D object detection and tracking for autonomous driving. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 3885–3894,

  16. [24]

    Hierarchical model-based human motion track- ing via unscented Kalman filter

    GuoJun Liu, XiangLong Tang, JianHua Huang, JiaFeng Liu, and Da Sun. Hierarchical model-based human motion track- ing via unscented Kalman filter. InProceedings of the IEEE International Conference on Computer Vision (ICCV), pages 1–8. IEEE, 2007. 2

    2007

  17. [25]

    Exploring simple 3D multi-object tracking for autonomous driving

    Chenxu Luo, Xiaodong Yang, and Alan Yuille. Exploring simple 3D multi-object tracking for autonomous driving. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 10488–10497, 2021. 1

    2021

  18. [26]

    Interacting multiple model methods in target track- ing: a survey.IEEE Transactions on Aerospace and Elec- tronic Systems, 34(1):103–123, 1998

    Efim Mazor, Amir Averbuch, Yakov Bar-Shalom, and Joshua Dayan. Interacting multiple model methods in target track- ing: a survey.IEEE Transactions on Aerospace and Elec- tronic Systems, 34(1):103–123, 1998. 1, 3, 4, 6, 8

    1998

  19. [27]

    Recurrent neural net- works.Design and Applications, 5(64-67):2, 2001

    Larry R Medsker, Lakhmi Jain, et al. Recurrent neural net- works.Design and Applications, 5(64-67):2, 2001. 3

    2001

  20. [28]

    Modeling and estimation for tracking maneuvering targets

    Richard L Moose, Hugh F Vanlandingham, and DH McCabe. Modeling and estimation for tracking maneuvering targets. IEEE Transactions on Aerospace and Electronic Systems, (3):448–456, 1979. 1

    1979

  21. [29]

    P2b: Point-to-box network for 3D object tracking in point clouds

    Haozhe Qi, Chen Feng, Zhiguo Cao, Feng Zhao, and Yang Xiao. P2b: Point-to-box network for 3D object tracking in point clouds. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 6329–6338, 2020. 1

    2020

  22. [30]

    Optimization-based state es- timation: Current status and some new results.Journal of Process Control, 22(8):1439–1444, 2012

    James B Rawlings and Luo Ji. Optimization-based state es- timation: Current status and some new results.Journal of Process Control, 22(8):1439–1444, 2012. 2

    2012

  23. [31]

    Unsupervised learned Kalman filtering

    Guy Revach, Nir Shlezinger, Timur Locher, Xiaoyong Ni, Ruud JG van Sloun, and Yonina C Eldar. Unsupervised learned Kalman filtering. InProceedings of the European Signal Processing Conference (EUSIPCO), pages 1571– 1575, 2022. 3, 6

    2022

  24. [32]

    KalmanNet: Neural network aided Kalman filtering for partially known dynamics.IEEE Transactions on Signal Processing, 70: 1532–1547, 2022

    Guy Revach, Nir Shlezinger, Xiaoyong Ni, Adria Lopez Es- coriza, Ruud JG Van Sloun, and Yonina C Eldar. KalmanNet: Neural network aided Kalman filtering for partially known dynamics.IEEE Transactions on Signal Processing, 70: 1532–1547, 2022. 3, 6

    2022

  25. [33]

    Kalman and extended Kalman filters: Concept, derivation and properties.Institute for Systems and Robotics, 43(46):3736–3741, 2004

    Maria Isabel Ribeiro. Kalman and extended Kalman filters: Concept, derivation and properties.Institute for Systems and Robotics, 43(46):3736–3741, 2004. 2

    2004

  26. [34]

    Algorithm for perfor- mance analysis of the imm algorithm.IEEE Transactions on Aerospace and Electronic Systems, 47(2):1114–1124, 2011

    Chze Eng Seah and Inseok Hwang. Algorithm for perfor- mance analysis of the imm algorithm.IEEE Transactions on Aerospace and Electronic Systems, 47(2):1114–1124, 2011. 7

    2011

  27. [35]

    Gaussian processes for machine learn- ing.International Journal of Neural Systems, 14(02):69– 106, 2004

    Matthias Seeger. Gaussian processes for machine learn- ing.International Journal of Neural Systems, 14(02):69– 106, 2004. 3

    2004

  28. [36]

    Incorporating Transformer and LSTM to Kalman Filter with EM algorithm for state estimation.arXiv preprint arXiv:2105.00250, 2021

    Zhuangwei Shi. Incorporating Transformer and LSTM to Kalman Filter with EM algorithm for state estimation.arXiv preprint arXiv:2105.00250, 2021. 3

    2021 arXiv

  29. [37]

    Model-based deep learning.Proceedings of the IEEE, 111(5):465–499, 2023

    Nir Shlezinger, Jay Whang, Yonina C Eldar, and Alexan- dros G Dimakis. Model-based deep learning.Proceedings of the IEEE, 111(5):465–499, 2023. 3

    2023

  30. [38]

    Re- inforcement learning compensated extended Kalman filter for attitude estimation

    Yujie Tang, Liang Hu, Qingrui Zhang, and Wei Pan. Re- inforcement learning compensated extended Kalman filter for attitude estimation. InProceedings of the International Conference on Intelligent Robots and Systems (IROS), pages 6854–6859. IEEE, 2021. 6

    2021

  31. [39]

    Attention is all you need.Advances in Neural Information Processing Systems (NeurIPS), 2017

    A Vaswani. Attention is all you need.Advances in Neural Information Processing Systems (NeurIPS), 2017. 6

    2017

  32. [40]

    Edge comput- ing enabled video segmentation for real-time traffic monitor- ing in internet of vehicles.Pattern Recognition, 121:108146,

    Shaohua Wan, Songtao Ding, and Chen Chen. Edge comput- ing enabled video segmentation for real-time traffic monitor- ing in internet of vehicles.Pattern Recognition, 121:108146,

  33. [41]

    Online selecting discriminative tracking features using particle filter

    Jianyu Wang, Xilin Chen, and Wen Gao. Online selecting discriminative tracking features using particle filter. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 1037–1042. IEEE,

  34. [42]

    Center- based 3D object detection and tracking

    Tianwei Yin, Xingyi Zhou, and Philipp Krahenbuhl. Center- based 3D object detection and tracking. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 11784–11793, 2021. 1

    2021

  35. [43]

    Privacy-preserving push-sum distributed cubature informa- tion filter for nonlinear target tracking with switching di- rected topologies.ISA Transactions, 136:16–30, 2023

    Jirong Zha, Liang Han, Xiwang Dong, and Zhang Ren. Privacy-preserving push-sum distributed cubature informa- tion filter for nonlinear target tracking with switching di- rected topologies.ISA Transactions, 136:16–30, 2023. 1, 3

    2023

  36. [44]

    Diffusion-based filter for fast and accurate collabora- tive tracking with low data transmission.Authorea Preprints,

    Jirong Zha, Nan Zhou, Zhenyu Liu, Tao Sun, and Xinlei Chen. Diffusion-based filter for fast and accurate collabora- tive tracking with low data transmission.Authorea Preprints,

  37. [45]

    Innovative interaction ap- proach in IMM filtering for vehicle motion models with un- equal states dimension.IEEE Transactions on Vehicular Technology, 71(4):3579–3594, 2022

    Jasmina Zuba ˇca, Michael Stolz, Richard Seeber, Markus Schratter, and Daniel Watzenig. Innovative interaction ap- proach in IMM filtering for vehicle motion models with un- equal states dimension.IEEE Transactions on Vehicular Technology, 71(4):3579–3594, 2022. 6

    2022

Pith tools

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