Pith. sign in

REVIEW 4 major objections 4 minor 28 references

KDMOS:Knowledge Distillation for Motion Segmentation

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A distilled BEV student reaches 78.8% IoU on SemanticKITTI-MOS at 40 FPS, closing most of the gap to a slow point-cloud teacher.

desk verdict Useful incremental KD result for MOS, but the distillation loss's geometric and class-space alignment is underspecified enough that the core mechanism may not be what it claims. read the letter →

arxiv 2506.14130 v1 pith:A62KTYF6 submitted 2025-06-17 cs.CV cs.AIcs.RO

Chunyu Cao , Jintao Cheng , Zeyu Chen , Linfan Zhan , Rui Fan , Zhijian He , Xiaoyu Tang This is my paper · ORCID
classification cs.CVcs.AIcs.RO
keywords movingobjectsegmentationknowledgedistillationLiDARpointcloudbird's-eyeviewclassimbalancedecoupledreal-timeinferenceSemanticKITTI-MOS
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 tries to prove that motion segmentation for autonomous driving does not have to trade accuracy for speed. It trains a fast bird's-eye-view student by distilling the outputs of a powerful but slow non-projection point-cloud teacher, with a loss shaped for the extreme imbalance between moving and static points. In the paper's experiments, the student reaches 78.8% test IoU on SemanticKITTI-MOS at 40 FPS, just 1.3 points below the heavy teacher, while also reducing parameters relative to its BEV baseline. If true, this gives a practical recipe for deploying motion segmentation in real time without sacrificing most of the accuracy of the most accurate offline models.

What carries the argument

The central mechanism is Weighted Decoupled Class Distillation (WDCD), a logits-only loss that rewrites ordinary KL distillation as a binary target/non-target term plus a non-target distribution term, then further branches by semantic class. For moving points, both terms are distilled; for the abundant non-moving class, only the non-target term is kept, and the loss is divided by class-content weights so the imbalance does not drown out rare moving points. The second piece of machinery is DySample, a point-sampling dynamic upsampler that replaces fixed convolutional upsampling, shrinking the student to 4.08M parameters.

What would settle it

Compute WDCD on the validation set with the teacher logits attached to randomly chosen grid cells rather than the actual ones; if the student's validation IoU stays near 79.4%, then the alignment between teacher points and student cells is not doing the work, and the distillation signal could be arbitrary.

Watch

Extended reading notes

Core claim

KDMOS claims that a fast BEV-projection student can inherit most of the accuracy of a slow 4D point-cloud teacher through logits-only distillation, that is, by training on the teacher's final per-class scores. On SemanticKITTI-MOS, this student scores 79.4% validation IoU and 78.8% test IoU, which is 3.9 and 2.9 points above the MotionBEV baseline and only 2.9 and 1.3 points below the frozen MambaMOS teacher, while running at 40 FPS and shrinking parameters by 7.69%. The paper attributes the gain to a class-aware distillation loss that avoids letting the abundant non-moving class dominate the learning signal.

Load-bearing premise

The load-bearing assumption is that the teacher's per-point predictions can be meaningfully paired with the student's BEV cells for a per-point loss, yet the paper does not explain which point goes to which cell when multiple scans land in one cell.

Editorial extensions

If this is right

  • A BEV student trained with WDCD reaches 79.4% validation and 78.8% test IoU on SemanticKITTI-MOS, improving over MotionBEV by 2.9 to 3.9 points while staying real-time.
  • The student runs at 40 FPS with 4.08M parameters, a 7.69% parameter cut and faster inference than the 34 FPS MotionBEV baseline.
  • WDCD is logits-only, so it adds no parameters, and the paper shows it lifts other MOS baselines when appended to them.
  • Dropping the target-class term for non-moving points avoids a harmful distillation effect that otherwise costs 0.3% IoU, so the decoupling is a correctness move rather than just a speed trick.
  • On Apollo, the method reaches 68.2% validation IoU without domain adaptation, supporting the claim that the distilled student generalizes across environments.

Reading between the lines

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

  • The moving-versus-non-moving decoupling should transfer to any segmentation task with extreme class imbalance, since the dominating easy class can be given only the non-target distillation term.
  • A cheap way to stress-test KDMOS is to inspect the teacher-to-BEV projection: counting how many teacher points share one student cell, and whether their labels agree, would tell whether the point-wise KL is computed on clean correspondences.
  • Because only logits are transferred, the teacher's learned temporal coupling enters the student only through final probabilities; a feature-level motion distillation could close the remaining 1.3-point gap, at the cost of architectural alignment.
  • The method's value in deployment is exactly the student's 40 FPS, since the teacher is needed only at training time; this is a direct consequence of choosing a logits-based scheme.
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 / 4 minor

Summary. The paper proposes KDMOS, a knowledge-distillation framework for LiDAR moving-object segmentation. A frozen MambaMOS teacher distills logits into a BEV-based MotionBEV student via a weighted decoupled class distillation (WDCD) loss, together with a DySample-based upsampling module. The authors report 79.4% validation and 78.8% test IoU on SemanticKITTI-MOS, 40 FPS inference, and a 7.69% parameter reduction over MotionBEV, plus generalization results on Apollo. The algebraic derivation in Eqs. (7)-(8) is correct, and the ablations in Tables III-V are internally consistent, but several implementation details required to compute the distillation loss are missing.

Significance. If the reported numbers are reproducible, the contribution is practically significant: it would be the first logits-based KD for MOS, achieving near-MambaMOS accuracy at real-time speed with a simple, parameter-free-at-inference loss. The derivation in Eqs. (7)-(8) is correct and the ablation tables support the modular claims. However, the current manuscript does not specify how teacher point logits are aligned with student BEV outputs or how a binary teacher provides four-class logits, so the central mechanism cannot be verified from the text alone.

major comments (4)
  1. [Section III-C, Eq. (8)] The distillation loss is written as a per-point KL between teacher probability p_T and student p_S, but the teacher operates on unordered 4D point clouds while the student outputs a BEV grid. The manuscript never states how teacher point logits are associated with student BEV cells, how student BEV logits are converted back to per-point predictions, or how collisions (multiple points per cell, multiple scans per cell) are resolved. Without this alignment, Eq. (8) is not computable from the provided text, so the ablation gains in Table III cannot be independently reproduced or verified.
  2. [Section III-C, Eq. (5)] Eq. (5) defines four class probabilities (unlabeled, static, movable, moving), but the teacher is described in Section III-B as MambaMOS, a binary moving-object segmentation model for SemanticKITTI-MOS. The paper does not specify how the frozen teacher supplies four logits per point. If the teacher provides only binary moving/non-moving logits, then the non-target distribution \hat p has only one effective class and the NCKD term in Eq. (8) is degenerate; the claimed benefit of non-target class distillation would then not be attributable to the stated mechanism. Please state the teacher output space and, if necessary, the mapping from binary scores to the four classes.
  3. [Section IV-A, Tables IV and V] All quantitative results are single training runs; no standard deviations or number of seeds are reported. The reported gains are small (e.g., +1.3 IoU in Table III, +2.9 in validation), so without variance estimates or repeated runs it is unclear whether the differences are significant. Additionally, the hyperparameters alpha=0.25, gamma=0.25, beta, and window sizes N, Q1, Q2 are fixed without sensitivity analysis; a figure or table showing IoU as a function of these parameters would strengthen the claim that the chosen settings are not cherry-picked.
  4. [Section III-C, Eq. (10)] The definition of W_i is ambiguous: 'Content[label]' is not given a mathematical expression, and WDCD = DCD / W_i implies inverse-frequency weighting, whereas the text describes 'label-assigned weighting' without specifying the direction. Because this term is central to the proposed loss, please define Content[label] explicitly and state whether division (rather than multiplication) by this ratio is intended; otherwise the implementation cannot be reproduced.
minor comments (4)
  1. [Table II] The table title 'CROSSVALIOU' should read 'Cross-validation performance' or similar; please fix the typo.
  2. [Section III-B] The item numbered '1)' at the start of Section III-B is a formatting fragment; please move it into a proper enumerated list.
  3. [Table VI] The last row 'KDMOS40 254.08 15' should be formatted as 'KDMOS 40 25 4.08 15' so that the FPS, ms, params, and size columns are legible.
  4. [Section III-C, Eq. (10)] The relationship between 'Content[label]' and the per-frame class frequencies should be stated with an explicit formula, since the current text only says 'ratio of points from different categories in the ith frame'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: WDCD is a label-weighted DKD loss and the reported gains are empirical, not constructed from the claimed outputs.

full rationale

The core derivation in Eqs. (5)-(9) is the standard DKD decomposition of the KL divergence into target and non-target components, grounded in the external reference [16]; the subsequent moving/non-moving decoupling and label-frequency weighting in Eqs. (9)-(10) are explicit loss-design choices, not definitions that presuppose the IoU result. The class weights W_i are computed from ground-truth label frequencies, so the validation and test IoU numbers are empirical training outcomes rather than fitted quantities re-reported as predictions. MambaMOS is used as a frozen pretrained teacher with externally reported benchmark performance, and the student (MotionBEV) is also an external baseline; although some baselines and the teacher come from overlapping author groups, none is invoked as a uniqueness theorem or used to forbid alternatives, so this is normal self-citation rather than load-bearing circularity. The paper does leave unspecified how teacher point logits are aligned to the student's BEV grid and how four-class probabilities are obtained from a binary MOS teacher, but that is a reproducibility and correctness gap, not a circular reduction of the claimed result to its own inputs.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

KDMOS does not introduce new physical entities; it combines existing components (DKD loss, DySample upsampling, BEV student, MambaMOS teacher). The central claim depends on three tuned hyperparameters (alpha, beta, gamma) and on assumptions inherited from prior work (z-range, four-category split, BEV projection adequacy). No machine-checked proofs or formal verification are provided.

free parameters (4)
  • alpha (DySample offset factor) = 0.25
    Controls the sampling offset in the DySample upsampling module; fixed to 0.25 without sensitivity analysis.
  • gamma (distillation loss weight) = 0.25
    Balances the segmentation loss and the WDCD distillation loss in Eq. 11; fixed to 0.25 without sensitivity analysis.
  • beta (NCKD weighting coefficient) = Not reported
    Weights the non-target-class KL loss in Eq. 9; the paper does not report its value or a sensitivity study.
  • Temporal window configuration (N, Q1, Q2) = Inherited from MotionBEV
    The number of input frames and the two time windows are taken from MotionBEV [8]; the central claim depends on these choices, which are not re-validated in this paper.
assumptions (4)
  • standard math The KL divergence decomposition in Eq. 8 is valid.
    The rearrangement of the KL divergence into target and non-target components follows from probability definitions and is correct.
  • domain assumption The four-category semantic split (unlabeled, static, movable, moving) is sufficient for MOS and the teacher's logits over these categories carry the knowledge to transfer.
    The teacher's output space is discretized into these four classes; if this decomposition misses relevant motion information, the distillation target is incomplete.
  • domain assumption The BEV projection preserves enough information for MOS and is compatible with the teacher's 3D representation for logits matching.
    The student operates on a 2D BEV grid while the teacher processes 4D point clouds; the paper assumes the two representations can be aligned for point-wise distillation, which is not explicitly validated.
  • ad hoc to paper The class-frequency ratio Content[label] is a correct per-frame weighting for balancing the distillation loss.
    The weighting W_i in Eq. 10 is a design choice based on label frequencies; it is not derived from first principles or justified by a comparison to other weighting schemes.

how reviews work

0 comments
Cite this review

Pith. "Pith review of KDMOS:Knowledge Distillation for Motion Segmentation." pith.science (2026). https://pith.science/paper/A62KTYF6

@misc{pith2026250614130,
  author       = {Pith},
  title        = {Pith review of: KDMOS:Knowledge Distillation for Motion Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/A62KTYF6}},
  note         = {Machine review of arXiv:2506.14130}
}
read the original abstract

Motion Object Segmentation (MOS) is crucial for autonomous driving, as it enhances localization, path planning, map construction, scene flow estimation, and future state prediction. While existing methods achieve strong performance, balancing accuracy and real-time inference remains a challenge. To address this, we propose a logits-based knowledge distillation framework for MOS, aiming to improve accuracy while maintaining real-time efficiency. Specifically, we adopt a Bird's Eye View (BEV) projection-based model as the student and a non-projection model as the teacher. To handle the severe imbalance between moving and non-moving classes, we decouple them and apply tailored distillation strategies, allowing the teacher model to better learn key motion-related features. This approach significantly reduces false positives and false negatives. Additionally, we introduce dynamic upsampling, optimize the network architecture, and achieve a 7.69% reduction in parameter count, mitigating overfitting. Our method achieves a notable IoU of 78.8% on the hidden test set of the SemanticKITTI-MOS dataset and delivers competitive results on the Apollo dataset. The KDMOS implementation is available at https://github.com/SCNU-RISLAB/KDMOS.

Figures

Figures reproduced from arXiv: 2506.14130 by the authors.

Figure 1
Figure 1. The performance, inference speed, and parameter size of different [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The KDMOS framework comprises three main components: the teacher model, the student model, and knowledge distillation. During training, [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Structure of the Dysample module. The input feature and original [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: The structure of WDCD, where p T t and p S t represent the teacher’s and student’s probabilities for the target class, respectively. that of moving points. During training, the high accuracy of non-moving classes significantly reduces the effectiveness of KL(b T |b S),…
Figure 5
Figure 5. Figure 5: The performance of the proposed WDCD module on other MOS [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Qualitative results of LiDAR-MOS on the SemanticKITTI-MOS validation set using different methods. Green circles indicate false negatives, [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 23 canonical work pages

  1. [2]

    MF-MOS: A Motion-Focused Model for Moving Object Segmentation

    J. Cheng, K. Zeng, Z. Huang, X. Tang, J. Wu, C. Zhang, X. Chen, and R. Fan, “Mf-mos: A motion-focused model for moving object segmentation,”arXiv preprint arXiv:2401.17023, 2024

    work page Pith review arXiv 2024

  2. [3]

    Slim: Self-supervised lidar scene flow and motion seg- mentation,

    S. A. Baur, D. J. Emmerichs, F. Moosmann, P. Pinggera, B. Ommer, and A. Geiger, “Slim: Self-supervised lidar scene flow and motion seg- mentation,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 13 126–13 136

    work page 2021

  3. [4]

    Ssf-mos: Semantic scene flow assisted moving object segmentation for autonomous vehicles,

    T. Song, Y . Liu, Z. Yao, and X. Wu, “Ssf-mos: Semantic scene flow assisted moving object segmentation for autonomous vehicles,”IEEE Transactions on Instrumentation and Measurement, 2024

    work page 2024

  4. [5]

    Self-supervised learning of lidar segmentation for autonomous indoor navigation,

    H. Thomas, B. Agro, M. Gridseth, J. Zhang, and T. D. Barfoot, “Self-supervised learning of lidar segmentation for autonomous indoor navigation,” in2021 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2021, pp. 14 047–14 053

    work page 2021

  5. [6]

    Moving object segmentation in 3d lidar data: A learning-based approach exploiting sequential data,

    X. Chen, S. Li, B. Mersch, L. Wiesmann, J. Gall, J. Behley, and C. Stachniss, “Moving object segmentation in 3d lidar data: A learning-based approach exploiting sequential data,”IEEE Robotics and Automation Letters, vol. 6, no. 4, pp. 6529–6536, 2021

    work page 2021

  6. [7]

    Efficient spatial-temporal information fusion for lidar-based 3d moving object segmentation,

    J. Sun, Y . Dai, X. Zhang, J. Xu, R. Ai, W. Gu, and X. Chen, “Efficient spatial-temporal information fusion for lidar-based 3d moving object segmentation,” in2022 IEEE/RSJ International Conference on Intel- ligent Robots and Systems (IROS). IEEE, 2022, pp. 11 456–11 463

    work page 2022

  7. [8]

    Motionbev: Attention- aware online lidar moving object segmentation with bird’s eye view based appearance and motion features,

    B. Zhou, J. Xie, Y . Pan, J. Wu, and C. Lu, “Motionbev: Attention- aware online lidar moving object segmentation with bird’s eye view based appearance and motion features,”IEEE Robotics and Automa- tion Letters, 2023

    work page 2023

  8. [9]

    Receding moving object segmentation in 3d lidar data using sparse 4d convolutions,

    B. Mersch, X. Chen, I. Vizzo, L. Nunes, J. Behley, and C. Stachniss, “Receding moving object segmentation in 3d lidar data using sparse 4d convolutions,”IEEE Robotics and Automation Letters, vol. 7, no. 3, pp. 7503–7510, 2022

    work page 2022

Show all 28 references
  1. [10]

    Insmos: Instance-aware moving object segmentation in lidar data,

    N. Wang, C. Shi, R. Guo, H. Lu, Z. Zheng, and X. Chen, “Insmos: Instance-aware moving object segmentation in lidar data,” in2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2023, pp. 7598–7605

    2023

  2. [11]

    Mambamos: Lidar-based 3d moving object segmentation with motion-aware state space model,

    K. Zeng, H. Shi, J. Lin, S. Li, J. Cheng, K. Wang, Z. Li, and K. Yang, “Mambamos: Lidar-based 3d moving object segmentation with motion-aware state space model,” inACM International Confer- ence on Multimedia (MM), 2024

    2024

  3. [12]

    Distilling the knowledge in a neural network,

    G. Hinton, O. Vinyals, and J. Dean, “Distilling the knowledge in a neural network,”arXiv preprint arXiv:1503.02531, 2015

    2015 arXiv

  4. [13]

    Point-to-voxel knowl- edge distillation for lidar semantic segmentation,

    Y . Hou, X. Zhu, Y . Ma, C. C. Loy, and Y . Li, “Point-to-voxel knowl- edge distillation for lidar semantic segmentation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 8479–8488

    2022

  5. [14]

    Knowledge distillation from 3d to bird’s-eye-view for lidar semantic segmenta- tion,

    F. Jiang, H. Gao, S. Qiu, H. Zhang, R. Wan, and J. Pu, “Knowledge distillation from 3d to bird’s-eye-view for lidar semantic segmenta- tion,” in2023 IEEE International Conference on Multimedia and Expo (ICME). IEEE, 2023, pp. 402–407

    2023

  6. [15]

    2dpass: 2d priors assisted semantic segmentation on lidar point clouds,

    X. Yan, J. Gao, C. Zheng, C. Zheng, R. Zhang, S. Cui, and Z. Li, “2dpass: 2d priors assisted semantic segmentation on lidar point clouds,” inEuropean Conference on Computer Vision. Springer, 2022, pp. 677–695

    2022

  7. [16]

    Decoupled knowledge distillation,

    B. Zhao, Q. Cui, R. Song, Y . Qiu, and J. Liang, “Decoupled knowledge distillation,” inProceedings of the IEEE/CVF Conference on computer vision and pattern recognition, 2022, pp. 11 953–11 962

    2022

  8. [17]

    L3-net: Towards learning based lidar localization for autonomous driving,

    W. Lu, Y . Zhou, G. Wan, S. Hou, and S. Song, “L3-net: Towards learning based lidar localization for autonomous driving,” inProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, pp. 6389–6398

    2019

  9. [18]

    On the efficacy of knowledge distilla- tion,

    J. H. Cho and B. Hariharan, “On the efficacy of knowledge distilla- tion,” inProceedings of the IEEE/CVF international conference on computer vision, 2019, pp. 4794–4802

    2019

  10. [19]

    Improved knowledge distillation via teacher assis- tant,

    S. I. Mirzadeh, M. Farajtabar, A. Li, N. Levine, A. Matsukawa, and H. Ghasemzadeh, “Improved knowledge distillation via teacher assis- tant,” inProceedings of the AAAI conference on artificial intelligence, vol. 34, no. 04, 2020, pp. 5191–5198

    2020

  11. [20]

    Cv-mos: A cross-view model for motion segmentation,

    X. Tang, Z. Chen, J. Cheng, X. Chen, J. Wu, and B. Xue, “Cv-mos: A cross-view model for motion segmentation,”IEEE Transactions on Instrumentation and Measurement, 2024

    2024

  12. [21]

    Learning to upsample by learning to sample,

    W. Liu, H. Lu, H. Fu, and Z. Cao, “Learning to upsample by learning to sample,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 6027–6037

    2023

  13. [22]

    Limoseg: Real-time bird’s eye view based lidar motion segmentation,

    S. Mohapatra, M. Hodaei, S. Yogamani, S. Milz, H. Gotzig, M. Simon, H. Rashed, and P. Maeder, “Limoseg: Real-time bird’s eye view based lidar motion segmentation,”arXiv preprint arXiv:2111.04875, 2021

    2021 arXiv

  14. [23]

    Cylindrical and asymmetrical 3d convolution networks for lidar segmentation,

    X. Zhu, H. Zhou, T. Wang, F. Hong, Y . Ma, W. Li, H. Li, and D. Lin, “Cylindrical and asymmetrical 3d convolution networks for lidar segmentation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 9939–9948

    2021

  15. [24]

    Rvmos: Range-view moving object seg- mentation leveraged by semantic and motion features,

    J. Kim, J. Woo, and S. Im, “Rvmos: Range-view moving object seg- mentation leveraged by semantic and motion features,”IEEE Robotics and Automation Letters, vol. 7, no. 3, pp. 8044–8051, 2022

    2022

  16. [25]

    Semantickitti: A dataset for semantic scene un- derstanding of lidar sequences,

    J. Behley, M. Garbade, A. Milioto, J. Quenzel, S. Behnke, C. Stach- niss, and J. Gall, “Semantickitti: A dataset for semantic scene un- derstanding of lidar sequences,” inProceedings of the IEEE/CVF international conference on computer vision, 2019, pp. 9297–9307

    2019

  17. [26]

    The pascal visual object classes (voc) challenge,

    M. Everingham, L. Van Gool, C. K. Williams, J. Winn, and A. Zisser- man, “The pascal visual object classes (voc) challenge,”International journal of computer vision, vol. 88, pp. 303–338, 2010

    2010

  18. [27]

    Automatic labeling to generate training data for online lidar-based moving object segmentation,

    X. Chen, B. Mersch, L. Nunes, R. Marcuzzi, I. Vizzo, J. Behley, and C. Stachniss, “Automatic labeling to generate training data for online lidar-based moving object segmentation,”IEEE Robotics and Automation Letters, vol. 7, no. 3, pp. 6107–6114, 2022

    2022

  19. [28]

    Logit standardization in knowledge distillation,

    S. Sun, W. Ren, J. Li, R. Wang, and X. Cao, “Logit standardization in knowledge distillation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 15 731–15 740

    2024

  20. [29]

    One-for-all: Bridge the gap between heterogeneous architectures in knowledge distillation,

    Z. Hao, J. Guo, K. Han, Y . Tang, H. Hu, Y . Wang, and C. Xu, “One-for-all: Bridge the gap between heterogeneous architectures in knowledge distillation,”Advances in Neural Information Processing Systems, vol. 36, 2024

    2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.