Pith. sign in

REVIEW 4 major objections 5 minor 39 references

STAR-VLM: Spatiotemporal Grounding Vision-Language Models for Motion and Velocity Estimation via Automotive Radar Supervision

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

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

Pith's one-line read STAR-VLM claims that a vision-language model supervised by projected automotive radar Doppler data can classify object motion and estimate per-pixel metric radial velocity, beating task-specific motion and scene-flow models on the authors'

desk verdict Radar Doppler as VLM supervision is a genuinely new idea with solid ablations, but the radar-relative vs camera-relative frame mismatch makes the headline velocity numbers less interpretable. read the letter →

arxiv 2608.01535 v1 pith:SB6GGGBE submitted 2026-08-02 cs.CV cs.RO

Pou-Chun Kung , Aryaman Rao , Utkrisht Sahai , Hemanth Murali , Yi Liu , Rui-Yu Lin , Katherine A. Skinner This is my paper
classification cs.CVcs.RO
keywords vision-languagemodelsautomotiveradarDopplersupervisionradialvelocityestimationmotionclassificationspatiotemporalreasoningautonomousdrivingmetric
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 seeks to establish that a general vision-language model can learn metric spatiotemporal reasoning—deciding whether a queried object is moving and how fast it approaches or recedes—from automotive radar Doppler measurements, without human annotation for its main training signal and without a task-specific architecture. Radar points are projected onto the camera image, filtered for occlusion, and converted into question-answer pairs that ask the model to report the motion state and radial velocity at a marked pixel. On the authors' STAR-Bench evaluation, the 4B model reaches motion classification accuracy 0.80/0.94 (human-annotation/radar labels) and radial velocity MAE 1.94/1.37 m/s, improving on general VLMs and, on these metrics, on task-specific baselines SegAnyMo (0.73/0.77) and Any4D (2.31/3.60). The broader claim is that radar is a scalable, physically grounded supervision source that can give VLMs the temporal counterpart of what LiDAR depth supervision gave spatial perception.

What carries the argument

Radar-as-supervisor: automobile-grade radar provides point-wise range and Doppler radial velocity; these returns are projected onto the camera image, filtered by a nearest-neighbor occlusion heuristic (30 pixels, 0.3 m depth gap), and turned into visual-question-answer training pairs with a rendered arrow plus textual pixel coordinates. Cross-task joint training with ego-velocity estimation is the complementary mechanism that lets the model compensate for ego motion and improves both motion classification and radial velocity output.

What would settle it

On held-out nuScenes frames, replace the radar-derived STAR-Bench labels with an independent reference—e.g., velocities from high-rate LiDAR tracking or manual object annotation—and recompute STAR-VLM's motion accuracy and radial velocity MAE on visible, unoccluded objects. If static surfaces are frequently labeled moving, or if the radar labels deviate from the independent reference by more than ~1.4 m/s, the reported gains are an artifact of label noise rather than VLM reasoning.

Watch

Extended reading notes

Core claim

Doppler radar can act as a free, physically consistent teacher for VLM spatiotemporal understanding. STAR-VLM projects radar returns with range and Doppler velocity into camera frames, filters occluded points, and builds QA pairs asking the model to output motion state and radial velocity at a marked pixel. Using a 4B Qwen3-VL fine-tuned with standard next-token SFT, no regression loss or architectural change, plus an ego-velocity estimation task that helps separate ego motion from object motion, the model reaches 0.80/0.94 motion accuracy and 1.94/1.37 m/s radial velocity MAE on STAR-Bench, surpassing task-specific methods. The authors take this as evidence that radar is a scalable supervis

Load-bearing premise

Radar Doppler returns, after projection to the image and the hand-designed occlusion filter, give correct radial-velocity and motion-state ground truth for the queried pixel; if radar points attach to the wrong image location or the filter systematically removes valid points, the model is trained and scored on false labels.

Editorial extensions

If this is right

  • A single general VLM can handle both motion classification and metric velocity estimation at a queried pixel, removing the need for task-specific segmentation or scene-flow heads.
  • Radar-derived supervision can replace human annotation for spatiotemporal training, since Doppler and ego-motion signals are obtained without labeling labor.
  • Jointly training with ego-velocity estimation improves both motion classification and radial velocity accuracy compared with training on those tasks alone.
  • The prompting design generalizes beyond marker-based queries: STAR-VLM retains performance on the TADBench format that refers to objects only by natural language.
  • The same training recipe yields better metric velocity accuracy than task-specific dense scene flow on the evaluation set.

Reading between the lines

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

  • Because radar Doppler is label-free and already deployed on production vehicles, the same projection-plus-filter supervision recipe could be applied to other multi-modal driving datasets or fleet data, scaling spatiotemporal VLM training without new annotation pipelines.
  • The occlusion filter is a blunt heuristic; a learned or depth-aware occlusion model would likely recover more valid radar points and could tighten supervision, a testable extension the paper leaves open.
  • Radial velocity is only one component of full 3D motion; combining this radar-supervised VLM with tangential velocity cues (optical flow or a second radar) is a natural next step toward full-velocity estimation, beyond the limitation the authors state.
  • The joint pixel-coordinate plus visual-marker prompt design may transfer to other point-level metric queries (depth, object size, distance) beyond velocity, since it solves the pixel-grounding problem generally.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. STAR-VLM proposes a radar-supervised training framework for vision-language models (VLMs) to perform spatiotemporal motion classification and metric radial velocity estimation at queried pixels. Using nuScenes front-camera images and front-radar measurements, the authors project radar points into the image, apply a hand-designed occlusion filter, and generate QA pairs from radar-derived Doppler/range labels (STAR-Bench-radar) and from human-annotated 3D boxes (STAR-Bench-anno). They fine-tune Qwen3-VL-4B with supervised fine-tuning, jointly training motion classification, radial velocity estimation, and ego-velocity estimation, using a combination of textual pixel coordinates and visual arrow markers as the pixel reference. They report improvements over off-the-shelf VLMs and over task-specific baselines SegAnyMo (motion classification) and Any4D (velocity estimation), plus results on the public TADBench temporal benchmark.

Significance. If the claims hold, the paper would demonstrate a scalable, annotation-free path to metric velocity reasoning in VLMs using widely deployed automotive radar, extending prior depth-supervision work (DepthLM) from spatial to temporal metric perception. The paper has several strengths: it uses a real-world, publicly available dataset; the internal ablation in Table II shows that joint ego-velocity training improves both motion classification and velocity prediction; and the STAR-Bench-anno evaluation provides an independent label source for the velocity task, mitigating the pure circularity of the radar-based benchmark. However, the headline claim of outperforming task-specific methods is weakened by a coordinate-frame ambiguity in the radar labels, by the self-referential nature of the radar benchmark, and by the lack of controlled comparison against task-specific baselines.

major comments (4)
  1. [Sec. III-A, Sec. III-B2, Table I] The radar Doppler measurement is the projection of the 3D velocity onto the radar–target ray, not the camera–target ray. The paper's prompt explicitly asks for radial velocity 'relative to the camera,' and STAR-Bench-radar labels are radar Doppler measurements projected onto the image plane, but no coordinate correction or conversion is described anywhere in Sec. III. For objects with substantial tangential motion, the radar-relative and camera-relative radial velocities can differ considerably. Therefore the 'Radar' columns in Table I may reflect learning the radar-specific projection rather than camera-relative metric velocity, undermining the comparison to Any4D and the claim of camera-relative velocity estimation. The authors must either derive and apply the conversion using the known radar-camera extrinsics and ego-motion, or reframe the benchmark and prompts as radar-relative radia
  2. [Sec. III-A, Sec. IV-C, Table I] STAR-Bench-radar is built from the same radar label pipeline used to generate training data; the reported Radar-MAE of 1.37 is therefore partly a measure of self-consistency between training and evaluation label generation, not an independent test of metric velocity estimation. The independent grounding comes only from STAR-Bench-anno, where labels are derived from 3D annotations. The paper should present the annotation-based results as the primary evidence for metric velocity, and treat the radar-based results as a sensor-reproduction check. Additionally, the occlusion filter parameters (N=30 pixels, tau=0.3 m) and the box-occlusion IoU threshold are chosen without sensitivity analysis; if these choices mislabel occluded or non-occluded points, the bias affects both training and evaluation in the same direction.
  3. [Sec. III-A, Sec. III-B2, Table I] The motion-state labels in STAR-Bench-radar are derived from Doppler radial velocity, so an object moving tangentially to the radar ray can have near-zero Doppler and be labeled 'static.' This conflates radial motion with full motion. The comparison to SegAnyMo, which is designed to segment full 3D/2D motion from video, is therefore not measuring the same physical quantity: the high 'Radar' accuracy of 0.94 may reflect that both STAR-VLM and the radar labels use the same radial criterion. The authors should define the threshold and direction convention for 'moving' and 'static,' report motion classification against the annotation-derived full-velocity labels as the primary metric, and discuss the discrepancy between radar- and annotation-derived motion labels.
  4. [Sec. IV-B, Sec. IV-D, Table I] The claim of state-of-the-art performance 'outperforming even task-specific methods' is not supported by a controlled comparison. STAR-VLM is fine-tuned on 34K nuScenes images, while SegAnyMo and Any4D are used as pre-trained, off-the-shelf models without nuScenes fine-tuning. The reported gains over these task-specific baselines may be due to the in-domain training data rather than to the radar supervision or the VLM architecture. The authors should either fine-tune or adapt the task-specific baselines on the same training split, or clearly state that the comparison is against zero-shot task-specific models and temper the 'state-of-the-art' wording accordingly.
minor comments (5)
  1. [Table I] The table header has inconsistent formatting ('VLMS Size' and the layout of the 'Cosmos-Reason 1 7B' entry). Please clean up the table for readability.
  2. [Sec. III-C1] The sentence 'no regression or regularization loss is needed in pure vision models' is confusing. It should be 'in vision-language models' or similar, and it would help to clarify what loss is used for the velocity text output.
  3. [Figure 6] The caption and text refer to 'depth estimation accuracy' but give no quantitative scale or metric. Please specify the metric (e.g., relative error, RMSE) and the evaluation set used to produce the figure.
  4. [Sec. IV-A] The implementation details state 34K images and ~900K radar points, but not the number of QA pairs used for training. Reporting the number of motion, velocity, and ego-velocity questions would help reproducibility.
  5. [References] Reference [31] is incomplete; author and venue information should be completed. Also, several arXiv references lack version or DOI information; consider adding them.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity; the only author self-citation is non-load-bearing.

full rationale

We found no circular step. STAR-VLM's central claim is an empirical supervised-learning result: a VLM is fine-tuned on radar-derived labels from the nuScenes trainval split and evaluated on the disjoint mini split plus the independent TADBench benchmark. The model's predictions are not definitionally equal to the radar labels; the held-out evaluation and the STAR-Bench-anno (human 3D annotation) results provide external checks. The comparison against SegAnyMo and Any4D is a benchmark comparison, not a reduction. The paper does not fit a parameter to the evaluation set, does not define its output quantity in terms of its input, and does not invoke a self-authored uniqueness theorem. The only author self-citation is [18] (CRKD) in the related-work survey, which is not load-bearing. The acknowledged limitation ('The current method is limited to relative radial velocity estimation') and the potential coordinate-frame mismatch between radar Doppler and camera-relative radial velocity are correctness/validity concerns, not circularity.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claims rest on sensor-derived labels (radar Doppler, ego-pose) being reliable and on the VLM being able to absorb metric supervision through text SFT. Several data-generation thresholds (N=30, tau=0.3, IoU=0.3, query height 0.3 m) are hand-chosen and unablated. No new physical entities are introduced.

free parameters (4)
  • Occlusion filter pixel threshold N = 30
    Hand-chosen pixel distance threshold for removing radar points occluded in the camera view (Sec. III-A). No ablation is provided for this value.
  • Occlusion filter depth margin tau = 0.3 m
    Hand-chosen depth margin to determine occlusion (Sec. III-A). Unablated.
  • Bounding box occlusion IoU threshold = 0.3
    Hand-chosen IoU threshold to remove occluded 3D box annotations for STAR-Bench-anno (Sec. III-A). Unablated.
  • Query point height for 3D boxes = 0.3 m
    Height at which the bounding box center is used as the query point (Sec. III-A). Unablated.
assumptions (5)
  • domain assumption Automotive radar Doppler measurements provide accurate ground-truth radial velocity relative to the radar sensor.
    The entire supervision pipeline treats radar radial velocity as label-free ground truth (Sec. III-A, III-C). Radar noise, calibration error, and multipath effects are not modeled.
  • domain assumption Ego-velocity can be reliably obtained from ego-pose estimation without human annotation.
    Cross-task training uses ego-velocity supervision derived from ego-pose (Sec. III-C2). The accuracy of this ego-velocity signal is not evaluated.
  • domain assumption Projecting radar points into the camera image using known extrinsics and intrinsics yields correct pixel correspondences for QA supervision.
    Training and evaluation labels assume accurate sensor synchronization and calibration (Sec. III-A).
  • domain assumption A VLM can learn metric radial velocity and motion state from text-formatted regression targets via supervised fine-tuning.
    The method relies on next-token prediction on numeric answers to teach metric reasoning (Sec. III-C1).
  • domain assumption nuScenes human annotations and camera-radar synchronization are reliable enough for benchmark construction.
    STAR-Bench-anno uses ground-truth 3D boxes and STAR-Bench-radar uses radar sweeps from nuScenes (Sec. III-A).

how reviews work

0 comments
Cite this review

Pith. "Pith review of STAR-VLM: Spatiotemporal Grounding Vision-Language Models for Motion and Velocity Estimation via Automotive Radar Supervision." pith.science (2026). https://pith.science/paper/SB6GGGBE

@misc{pith2026260801535,
  author       = {Pith},
  title        = {Pith review of: STAR-VLM: Spatiotemporal Grounding Vision-Language Models for Motion and Velocity Estimation via Automotive Radar Supervision},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SB6GGGBE}},
  note         = {Machine review of arXiv:2608.01535}
}
read the original abstract

Vision-language models (VLMs) are emerging as a key component of embodied intelligence, with growing applications in auto-labeling and end-to-end autonomous driving. However, existing approaches for improving spatiotemporal reasoning in VLMs often rely on complex preprocessing pipelines, expensive human annotations, or synthetic data, which limit scalability and introduce potential sim-to-real gaps. Moreover, although these methods have improved spatiotemporal understanding, they still lack strong metric reasoning capabilities for dynamic scenes, such as estimating object motion in real-world units. Prior work has explored LiDAR-based metric depth supervision to enhance spatial perception, but it does not directly address temporal reasoning. We introduce STAR-VLM, an automotive radar-supervised framework that enhances spatiotemporal VLMs with motion reasoning and metric velocity estimation for autonomous driving. Automotive radar is a low-cost and widely deployed sensor that provides complementary spatiotemporal supervision through range and Doppler measurements. By leveraging these measurements as label-free ground truth during training, STAR-VLM improves the metric spatiotemporal reasoning ability of VLMs. Through experiments on driving scenarios, we show that STAR-VLM achieves state-of-the-art performance on both motion classification and metric velocity estimation, outperforming even task-specific methods designed for each task. These results highlight automotive radar as a scalable and cost-effective source of supervision for building metric-aware spatiotemporal VLMs for real-world autonomous driving.

Figures

Figures reproduced from arXiv: 2608.01535 by the authors.

Figure 1
Figure 1. We propose STAR-VLM, a simple yet effective method that leverages [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Automotive radar provides relative radial velocity, indicating whether [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figure 4
Figure 4. Overview of STAR-VLM. We project radar points onto the camera image to obtain supervision labels and construct a QA dataset from radar [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗
Figures from the paper (5 more)
Figure 5
Figure 5. Figure 5: Projection of radar points into image space with and without the proposed occlusion filter. Point color represents the distance of each radar point. [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 6
Figure 6. Figure 6: Impact of different input pixel references. The numbers indicate [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: Illustration of the effect of ego-velocity joint training on motion [PITH_FULL_IMAGE:figures/full_fig_p005_7.png]
Figure 8
Figure 8. Figure 8: Although SegAnyMo produces dense motion masks, it does not [PITH_FULL_IMAGE:figures/full_fig_p006_8.png]
Figure 9
Figure 9. Figure 9: Any4D produces dense scene flow, but the metric velocity estimation [PITH_FULL_IMAGE:figures/full_fig_p006_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

39 extracted references · 25 canonical work pages

  1. [1]

    Simlingo: Vision-only closed-loop autonomous driving with language- action alignment,

    K. Renz, L. Chen, E. Arani, and O. Sinavski, “Simlingo: Vision-only closed-loop autonomous driving with language- action alignment,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025, pp. 11 993–12 003

    work page 2025

  2. [2]

    Autovla: A vision-language-action model for end-to-end autonomous driving with adaptive reasoning and reinforcement fine-tuning,

    Z. Zhou,et al., “Autovla: A vision-language-action model for end-to-end autonomous driving with adaptive reasoning and reinforcement fine-tuning,”arXiv preprint arXiv:2506.13757, 2025

    arXiv 2025

  3. [3]

    Spatialvlm: Endowing vision-language mod- els with spatial reasoning capabilities,

    B. Chen,et al., “Spatialvlm: Endowing vision-language mod- els with spatial reasoning capabilities,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recog- nition (CVPR), 2024, pp. 14 455–14 465

    work page 2024

  4. [4]

    Robospatial: Teaching spatial understanding to 2d and 3d vision-language models for robotics,

    C. H. Song, V . Blukis, J. Tremblay, S. Tyree, Y . Su, and S. Birchfield, “Robospatial: Teaching spatial understanding to 2d and 3d vision-language models for robotics,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025, pp. 15 768–15 780

    work page 2025

  5. [5]

    Vlm4d: Towards spatiotemporal awareness in vision language models,

    S. Zhou,et al., “Vlm4d: Towards spatiotemporal awareness in vision language models,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2025, pp. 8600–8612

    work page 2025

  6. [6]

    Depthlm: Metric depth from vision language models,

    Z. Cai,et al., “Depthlm: Metric depth from vision language models,”arXiv preprint arXiv:2509.25413, 2025

    arXiv 2025

  7. [7]

    Rt-2: Vision-language-action models transfer web knowledge to robotic control,

    A. Brohan,et al., “Rt-2: Vision-language-action models transfer web knowledge to robotic control,” inConference on Robot Learning (CoRL), 2023

    work page 2023

  8. [8]

    Drivelm: Driving with graph visual question an- swering,

    C. Sima,et al., “Drivelm: Driving with graph visual question an- swering,” inEuropean Conference on Computer Vision (ECCV), 2024

    work page 2024

Show all 39 references
  1. [9]

    Drivevlm: The convergence of autonomous driving and large vision-language models,

    X. Tian,et al., “Drivevlm: The convergence of autonomous driving and large vision-language models,”arXiv preprint arXiv:2402.12289, 2024

    2024 arXiv

  2. [10]

    Drivemlm: Aligning multi-modal large lan- guage models with behavioral planning states for autonomous driving,

    W. Wang,et al., “Drivemlm: Aligning multi-modal large lan- guage models with behavioral planning states for autonomous driving,”arXiv preprint arXiv:2312.09245, 2023

    2023

  3. [11]

    Drivemllm: A benchmark for spatial under- standing with multimodal large language models in autonomous driving,

    X. Guo,et al., “Drivemllm: A benchmark for spatial under- standing with multimodal large language models in autonomous driving,”arXiv preprint arXiv:2411.13112, 2024

    2024 arXiv

  4. [12]

    Futuresightdrive: Thinking visually with spatio-temporal cot for autonomous driving,

    S. Zeng,et al., “Futuresightdrive: Thinking visually with spatio-temporal cot for autonomous driving,”arXiv preprint arXiv:2505.17685, 2025

    2025 arXiv

  5. [13]

    Stsbench: A spatio-temporal scenario benchmark for multi-modal large language models in autonomous driving,

    C. Fruhwirth-Reisinger, D. Mali ´c, W. Lin, D. Schinagl, S. Schulter, and H. Possegger, “Stsbench: A spatio-temporal scenario benchmark for multi-modal large language models in autonomous driving,”arXiv preprint arXiv:2506.06218, 2025

    2025 arXiv

  6. [14]

    From segments to scenes: Temporal un- derstanding in autonomous driving via vision-language model,

    K. Cannons,et al., “From segments to scenes: Temporal un- derstanding in autonomous driving via vision-language model,” arXiv preprint arXiv:2512.05277, 2025

    2025 arXiv

  7. [15]

    Are vlms ready for autonomous driving? an em- pirical study from the reliability, data, and metric perspectives,

    S. Xie,et al., “Are vlms ready for autonomous driving? an em- pirical study from the reliability, data, and metric perspectives,” inProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2025, pp. 6585–6597

    2025

  8. [16]

    Centerfusion: Center-based radar and camera fusion for 3d object detection,

    R. Nabati and H. Qi, “Centerfusion: Center-based radar and camera fusion for 3d object detection,” inProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2021

    2021

  9. [17]

    Rcbevdet: Radar-camera fusion in bird’s eye view for 3d object detection,

    Z. Lin,et al., “Rcbevdet: Radar-camera fusion in bird’s eye view for 3d object detection,” inProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR), 2024

    2024

  10. [18]

    Crkd: Enhanced camera- radar object detection with cross-modality knowledge distilla- tion,

    L. Zhao, J. Song, and K. A. Skinner, “Crkd: Enhanced camera- radar object detection with cross-modality knowledge distilla- tion,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024

    2024

  11. [19]

    Depth estimation from monocular images and sparse radar data,

    J.-T. Lin, D. Dai, and L. Van Gool, “Depth estimation from monocular images and sparse radar data,” inProceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2020

    2020

  12. [20]

    Tacodepth: Towards efficient radar-camera depth estimation with one-stage fusion,

    Y . Wang,et al., “Tacodepth: Towards efficient radar-camera depth estimation with one-stage fusion,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025

    2025

  13. [21]

    Depth estimation based on mmwave radar and camera fusion with attention mechanisms and multi-scale features for autonomous driving vehicles,

    Z. Zhu, F. Wu, W. Sun, Q. Wu, F. Liang, and W. Zhang, “Depth estimation based on mmwave radar and camera fusion with attention mechanisms and multi-scale features for autonomous driving vehicles,”Electronics, vol. 14, no. 2, p. 300, 2025

    2025

  14. [22]

    Full-velocity radar returns by radar-camera fusion,

    Y . Long, D. Morris, X. Liu, M. Castro, P. Chakravarty, and P. Narayanan, “Full-velocity radar returns by radar-camera fusion,” inProceedings of the IEEE/CVF International Con- ference on Computer Vision (ICCV), 2021

    2021

  15. [23]

    Pow4r: Point-wise full-velocity esti- mation using 4d radar-camera fusion beyond radial limitations,

    H. Jeon and K.-S. Kim, “Pow4r: Point-wise full-velocity esti- mation using 4d radar-camera fusion beyond radial limitations,” IEEE Robotics and Automation Letters, 2025

    2025

  16. [24]

    Dogflow: Self-supervised lidar scene flow via cross-modal doppler guidance,

    A. Khoche, Q. Zhang, Y . Cai, S. Sharif Mansouri, and P. Jens- felt, “Dogflow: Self-supervised lidar scene flow via cross-modal doppler guidance,”arXiv preprint arXiv:2508.18506, 2025

    2025 arXiv

  17. [25]

    Raliflow: Scene flow esti- mation with 4d radar and lidar point clouds,

    J. Fu, Z. Xiang, and N. Zhao, “Raliflow: Scene flow esti- mation with 4d radar and lidar point clouds,”arXiv preprint arXiv:2512.10376, 2025

    2025

  18. [26]

    R4dyn: Exploring radar for self-supervised monocular depth estimation of dynamic scenes,

    J. Kunz, Y .-T. Ye, Q. Chen,et al., “R4dyn: Exploring radar for self-supervised monocular depth estimation of dynamic scenes,” arXiv preprint arXiv:2108.04814, 2021

    2021 arXiv

  19. [27]

    Radar as a teacher: Weakly supervised vehicle detection using radar labels,

    G. P. Meyer, V . Kurtz, C. Kanan, and D. Anguelov, “Radar as a teacher: Weakly supervised vehicle detection using radar labels,” inProceedings of the IEEE International Conference on Robotics and Automation (ICRA), 2020

    2020

  20. [28]

    Motion segmentation from a moving monocular camera,

    Y . Huang and J. Zelek, “Motion segmentation from a moving monocular camera,”arXiv preprint arXiv:2309.13772, 2023

    2023 arXiv

  21. [29]

    On moving object segmenta- tion from monocular video with transformers,

    C. Homeyer and C. Schn ¨orr, “On moving object segmenta- tion from monocular video with transformers,”arXiv preprint arXiv:2411.19141, 2024

    2024 arXiv

  22. [30]

    Segment any motion in videos,

    N. Huang,et al., “Segment any motion in videos,”arXiv preprint arXiv:2503.22268, 2025

    2025 arXiv

  23. [31]

    Camera-based vehicle velocity estimation from monocular video,

    M. Kampelm ¨uhler, M. G. M ¨uller, and C. Feichtenhofer, “Camera-based vehicle velocity estimation from monocular video,”arXiv preprint arXiv:1802.07094, 2018

    2018 arXiv

  24. [32]

    Kinematic 3d object detection in monocular video,

    G. Brazil, G. Pons-Moll, X. Liu, and B. Schiele, “Kinematic 3d object detection in monocular video,” inEuropean Conference on Computer Vision (ECCV), 2020

    2020

  25. [33]

    Self-supervised monocular scene flow estimation,

    J. Hur and S. Roth, “Self-supervised monocular scene flow estimation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020

    2020

  26. [34]

    Any4d: Unified feed-forward metric 4d reconstruction,

    J. Karhade,et al., “Any4d: Unified feed-forward metric 4d reconstruction,”arXiv preprint arXiv:2512.10935, 2025

    2025

  27. [35]

    Zero-shot monocular scene flow estimation in the wild,

    Y . Liang, A. Badki, H. Su, J. Tompkin, and O. Gallo, “Zero-shot monocular scene flow estimation in the wild,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025

    2025

  28. [36]

    nuscenes: A multimodal dataset for au- tonomous driving,

    H. Caesar,et al., “nuscenes: A multimodal dataset for au- tonomous driving,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020, pp. 11 621–11 631

    2020

  29. [37]

    Qwen3-vl technical report,

    S. Bai,et al., “Qwen3-vl technical report,”arXiv preprint arXiv:2511.21631, 2025

    2025 arXiv

  30. [38]

    Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution,

    P. Wanget al., “Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution,”arXiv preprint arXiv:2409.12191, 2024

    2024 arXiv

  31. [39]

    Cosmos-reason1: From physical common sense to embodied reasoning,

    NVIDIA,et al., “Cosmos-reason1: From physical common sense to embodied reasoning,”arXiv preprint arXiv:2503.15558, 2025

    2025 arXiv

Pith tools

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