Toward a Low-Cost Perception System in Autonomous Vehicles: A Spectrum Learning Approach

Aidan Erickson; John M. Dolan; Mohammed Alsakabi; Ozan K. Tonguz

REVIEW 5 major objections 5 minor 27 references

Toward a Low-Cost Perception System in Autonomous Vehicles: A Spectrum Learning Approach

T0 review · 5 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read This paper shows that a camera used only during offline training can teach a 4D radar to produce sharper depth maps, beating the state-of-the-art detector by up to 52.59% in unidirectional chamfer distance.

desk verdict Clever spectral encoding idea, but the main quantitative claim is unverifiable as written: no inverse transform from the spectrum domain back to depth maps is specified. read the letter →

arxiv 2502.01940 v2 pith:NJF46OFD submitted 2025-02-04 cs.CV eess.IV

Mohammed Alsakabi , Aidan Erickson , John M. Dolan , Ozan K. Tonguz This is my paper · ORCID

classification cs.CVeess.IV

keywords autonomousvehicles4Dradardepthmapgenerationspatialspectrumestimationcamera-radarfusionsemanticsegmentationpositionalencodingcompletion

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that a low-cost camera can be used offline to train a 4D radar sensor to produce dense, sharp depth maps for autonomous vehicles. The method transforms the radar depth map and a semantic segmentation of the camera image into a shared 'spatial spectrum' subspace, using a nonlinear pixel positional encoding inspired by periodogram-averaging spectral estimation, then trains a residual network to predict the elementwise product of the two spectra from the radar spectrum alone. In the paper's experiments on two scenes of a public 4D radar dataset, the resulting depth maps reduce mean absolute error by 24.24%, relative error by 18.52%, unidirectional chamfer distance (a point-cloud distance metric) by 52.59%, and bidirectional chamfer distance by 10.41% compared with the state-of-the-art data-driven radar detector. The encoding also raises linear correlation and mutual information between the two modalities by factors of 3.88 and 76.69. If the claim holds, lidar could be removed from the sensor stack without giving up radar's all-weather robustness.

What carries the argument

The load-bearing object is the nonlinear spatial spectrum transform. Starting from $M$ segments with phase progression $x(m,\phi) = e^{-j\pi m\sin(\phi)}$ and $x(m,\theta) = e^{-j\pi m\sin(\theta)}$, the algorithm forms periodogram estimates $y(\phi_n)$ and $y(\theta_k)$, combines them into a joint 2D periodogram $Y(\phi_n,\theta_k)$, and accumulates $P(n,k) = \sum_{n,k} |Y(\phi_n,\theta_k) \circ I|$ over all pixels of the input image. The resolution of the resulting spectrum grows with the number of segments $M$, and spectral leakage introduces high-frequency components that help complete sparse depth and fit sharp camera contours. This transform is what lets the training network compare radar and camera data in a common subspace, rather than trying to map between raw depth and RGB pixel spaces directly.

What would settle it

Run the trained pipeline on a held-out scene of the same 4D radar dataset and compare the output depth maps to lidar ground truth; if the reported unidirectional chamfer distance (UCD) and mean absolute error (MAE) improvements over the state-of-the-art detector do not appear on that scene, the claimed generalization is not established.

Watch

Extended reading notes

Core claim

The central claim is that radar and camera images can be brought into one learnable subspace, the spatial spectrum, by estimating each image's 2D power spectrum over a common set of nonlinear spatial-frequency basis functions. For the radar, the preprocessing computes $P_{radar} = F(I_{radar}, M_{radar})$; for the camera, the RGB frame is first segmented semantically and then encoded as $P_{cam} = F(Seg, M_{cam})$, with $M_{cam} > M_{radar}$ so the camera side retains higher resolution. Training uses the objective $P_{radar} \circ P_{cam} = \mathrm{ResNet}(P_{radar})$ under L2 loss, so the network learns to reconstruct the elementwise product of radar and camera spectra from the radar input alone. The paper reports that this yields sharper object contours and lowers MAE, REL, UCD, and BCD relative to the state-of-the-art data-driven radar detector, and that after offline training the camera is no longer needed at runtime.

Load-bearing premise

The pipeline rests on the unstated assumption that improving accuracy in the transformed spectrum space automatically improves the final depth maps, because the paper does not describe the inverse transform that turns the network's spectrum output back into depth coordinates.

Editorial extensions

If this is right

After offline training, the 4D radar depth generator runs without the camera, so a production system could keep radar's all-weather robustness while shedding lidar cost.
Denser, contour-accurate radar depth maps should improve downstream perception, tracking, and rendering modules that consume point clouds.
The large gains in linear correlation and mutual information after encoding suggest the spatial spectrum is a promising common representation for other sensor-modality pairs, not just radar and camera.
The ablation results tie performance to the resolution parameters $M_{radar} \approx 20$ and $M_{cam} \ge 180$, so each new sensor pairing would need its own resolution search.

Reading between the lines

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

If the spectrum transform is modality-agnostic, the same encoding could sharpen lidar or sonar depth maps using camera supervision offline, extending the approach beyond automotive radar.
Because the camera is removed at runtime, the method's practical value depends on weather robustness: a natural test is measuring depth-map accuracy in rain, fog, or darkness, where camera-based supervision would be degraded or absent.
The paper's argument that the segmentation benchmark matters more than the segmentation model implies that transferring to a new city or object set requires re-training the segmentation stage on a matching urban benchmark.
Reproducing the quantitative claims requires the authors to state the missing inverse spectrum transform; a follow-up with a differentiable, explicitly invertible encoding would make the whole pipeline verifiable.

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.

Desk Editor's Note

read the letter

Quick take: the non-linear phase-progression basis and the spectrum-domain training target are genuinely new relative to the cited radar-camera fusion work, and the cross-modal correlation/mutual-information gains are plausible. But the headline SOTA numbers cannot be checked, because the inverse spectral-to-depth transform is never given. That is load-bearing, not a footnote.

What the paper does well: the pixel positional encoding via a non-linear Bartlett-style basis is a real contribution, and using the element-wise product P_radar * P_cam as a training target is not something I have seen in the prior work they cite. The ablation study is a good-faith attempt, and the qualitative results in Fig. 5 do suggest sharper depth maps. The framing of the deployment scheme — camera supervises radar offline, radar runs solo online — is sensible and worth exploring.

Where it falls apart: Section V.B says "spectrum images are first transformed into 3D point clouds" for UCD/BCD, but no such inverse transform appears in Section IV, V, or the appendix. Eq. 8 is dimensionally off: the RHS sums over n and k, yielding a scalar, while P is declared an N-by-K matrix. Without the inverse mapping, a reader cannot reproduce Table III, verify the claimed 24.24% MAE / 52.59% UCD reductions, or even tell whether Fig. 5 shows raw ResNet output or a post-processed product. The stress-test note is on point.

The abstract also disagrees with the main text (it says 27.95% UCD and swaps the MAE/UCD percentages), which does not inspire confidence in the reporting. On top of that, no train/test split is described; all 5680 frames from scenes 2 and 6 are used, and the M_radar/M_cam hyperparameters are selected using the same evaluation frames via the Fig. 7 analysis. The paper also does not compare against the radar-camera fusion methods [12]–[16] that it lists in related work, which makes the "SOTA" claim weaker than it should be.

Bottom line: the core ideas deserve a serious referee round, but this version is not publishable as is. Major revision is needed: specify the inverse transform, fix the equation, report a real train/test split, and add at least one strong baseline from the cited fusion literature. I would not desk-reject it, but I would not accept it either.

Referee Report

5 major / 5 minor

Summary. The paper proposes a camera-radar fusion pipeline for generating denser 4D radar depth maps. RGB images are passed through a semantic segmentation network, and both the radar depth map and the segmentation map are transformed by a Bartlett-inspired nonlinear spatial-spectrum encoding. A ResNet is trained to map the radar spectrum to the element-wise product of the radar and camera spectra, with L2 loss. After offline training, the authors claim the radar-only model can produce sharper depth maps that outperform the SOTA detector on RaDelft, reporting reductions in MAE, REL, UCD, and BCD. The paper also reports increased Pearson correlation and mutual information between the encoded modalities and presents an ablation study over the number of spectral segments M.

Significance. If the central claims held, the idea of learning in a common spectral subspace shared by radar and camera would be an interesting step toward low-cost, lidar-free depth perception for autonomous vehicles. The paper uses a public dataset, compares against a recent SOTA detector, and includes ablations over the key parameter M, which are positive features. However, the significance cannot currently be assessed because the evaluation pipeline is incomplete: no inverse transform from the spectral domain to depth maps or point clouds is specified, no train/test split is described, and hyperparameters appear to be selected on the same frames used for final evaluation. These are load-bearing gaps, not presentation issues, because Table III and the claimed SOTA improvements rest entirely on the missing components.

major comments (5)

[Section V.B and Eq. (11)] The inverse transform from the learned spatial spectrum to depth maps or point clouds is never defined. The network is trained to predict P_radar ∘ P_cam in the spectral domain, but MAE, REL, UCD, and BCD in Table III are computed on depth maps and point clouds. The sentence in Section V.B that "spectrum images are first transformed into 3D point clouds" is the only description, and no such transform appears in Section IV or anywhere else. Without an explicit inverse of F(·) or H(·), the quantitative results in Table III cannot be reproduced, and the claimed 24.24%, 18.52%, 52.59%, and 10.41% improvements are unverifiable from the manuscript.
[Section IV.A, Eq. (8)] The forward transform is not well-defined as written. Equation (8) states P(n,k) = Σ_{n=0}^{N-1} Σ_{k=0}^{K-1} |Y(ϕ_n,θ_k) ∘ I|, but the right-hand side sums over n and k, so it is a scalar independent of the left-hand side indices, making P(n,k) inconsistent with the declared N×K matrix P. Additionally, the dimensions of I and of Y are not specified, so it is unclear how an input image is encoded into the spectrum. This ambiguity undermines the definition of the core preprocessing step used throughout the paper.
[Section V (experimental setup)] No train/test split is described. The paper states that scenes 2 and 6 of RaDelft include 5680 frames and that these are used for the experiments, but it does not state how many frames are used for training, validation, and testing. If the metrics in Table III are computed on the same frames used for training, they measure fitting rather than generalization. The paper must specify the split and report metrics on a held-out test set for the claims to be meaningful.
[Section VI.A and Table III] The hyperparameters M_radar and M_cam appear to be selected on the same frames later used for evaluation. Figure 7 is used to choose M_radar = 20 and M_cam = 200 by maximizing Pearson correlation and mutual information, and Table III is then reported on the same dataset. This is a form of test-data leakage that can inflate the reported improvements. The paper should select hyperparameters on a validation split and clearly separate that selection from the final evaluation.
[Section IV.D, Eq. (11)] The training objective contains a potentially trivial component: the target P_radar ∘ P_cam includes P_radar as a multiplicative factor, and the network input is P_radar itself. A network can reduce part of the L2 loss by approximating the identity on P_radar, regardless of the camera information. The paper does not isolate how much of the reported depth-map improvement comes from the camera term versus this self-predictive component. An ablation that trains with P_radar as both input and target, or with the camera term removed, would be needed to support the claim that cross-modal learning is responsible for the gains.

minor comments (5)

[Abstract and Introduction] The headline improvement numbers are inconsistent. The abstract at the top of the paper says "outperforms the state-of-the-art (SOTA) by 27.95% in terms of Unidirectional Chamfer Distance (UCD)", while the full-text abstract and Introduction report different pairings; Table III lists 52.59% UCD and 24.24% MAE. These numbers should be reconciled.
[Section V.A] The paper says experiments were performed for M ∈ {10, 20, 50, 70, 200}, but Section VI.A says the ablation uses M ∈ {10, 12, 14, ..., 200}. Please clarify which values were actually used and how Figure 7 was generated.
[Section VI.C] The spectrum-representation ablation in Figure 8 is qualitative only. A quantitative comparison, using the same metrics as Table III, would be needed to support the claim that the spectrum step is essential.
[Section V.B] Training details are missing: the paper does not specify the optimizer, learning rate schedule, batch size, loss weighting, or hardware. The GitHub link is useful, but the manuscript should contain enough detail for independent reproduction.
[Throughout] There are numerous typographical and notation inconsistencies, including "V oD" instead of "VoD", "A Vs" spacing, "SoTA" vs "SOTA", and the undefined symbol "F(.)" in Figure 1. A careful editing pass is needed.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the core result is a supervised spectrum-domain training pipeline, not a derivation that reduces to its own inputs; the main issues are an omitted inverse transform and test-set-based hyperparameter selection, which are reproducibility and leakage concerns rather than circularity.

full rationale

The paper's central quantitative claim is an empirical comparison of a ResNet trained in a spatial-spectrum domain against SOTA baselines on MAE, REL, UCD, and BCD. The training objective in Eq. 11, Pradar ∘ Pcam = ResNet(Pradar), defines a supervised target from the input radar spectrum and a camera-derived spectrum; this is a standard cross-modal distillation setup, not a self-definitional reduction. The evaluation, however, requires converting the ResNet output back to a depth map or point cloud, and the paper only states in Section V.B that 'the spectrum images are first transformed into 3D point clouds before the metrics are computed,' without specifying any inverse of Eqs. 6-8. This omission makes Table III unverifiable, but it does not make the claim equivalent to its inputs by construction; it is a correctness and reproducibility gap. The hyperparameters Mradar = 20 and Mcam = 200 are selected using the same scenes on which final metrics are reported (Section VI.A, Fig. 7), which is self-referential test-set tuning and a mild leakage concern, yet these choices are made on cross-modal correlation and mutual information, not on the target depth errors, so they do not force the reported improvements. There are no load-bearing self-citations: the cited radar detector [6] and semantic segmentation models are external prior work, not the present authors' results. Overall, the paper is not circular in the sense that a prediction is equivalent to a fitted input or to a self-citation chain; the honest finding is no significant circularity, with the inverse-transform omission and test-set tuning noted as correctness risks.

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

The main load-bearing choices are the spectral transform parameters (Mradar, Mcam, angle range), which are hand-picked on the evaluation frames, and the untested element-wise product fusion rule. The paper does not introduce new physical entities.

free parameters (6)

Mradar = 20
Number of spectrum segments for radar encoding; selected via ablation on the test frames (Figure 7c).
Mcam = 200
Number of segments for camera encoding; selected similarly; paper states any value except 120 gives best values.
Angle range Phi and Theta = (-70, 70) degrees
Chosen to limit spectral leakage; no independent justification.
Training epochs = 10,000
No early stopping or regularization described; likely overfits the 5680 frames.
Logarithmic compression = natural logarithm
Applied to input data for scale; no ablation.
ResNet101 = ResNet101
Generator architecture; no comparison with other networks.

assumptions (5)

standard math Bartlett's periodogram averaging is a valid spectral estimator for 2D image data
Invoked in Section III.B and IV.A.
ad hoc to paper The non-linear basis e^{-jπm sin(φ)} provides higher resolution than the standard Fourier basis
Claimed in Section IV.A without proof or theoretical justification.
domain assumption Semantic segmentation from DeepLab v3 on Cityscapes is an adequate proxy for camera scene structure
Used throughout; no analysis of segmentation error impact.
ad hoc to paper Element-wise multiplication of spectra captures the mutuality of radar and camera information
Eq. (11) states the target without derivation or ablation of alternative fusion rules.
domain assumption RaDelft scenes 2 and 6 generalize to other complex vehicular environments
Only these two scenes are used; no cross-dataset validation.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Toward a Low-Cost Perception System in Autonomous Vehicles: A Spectrum Learning Approach." pith.science (2026). https://pith.science/paper/NJF46OFD

@misc{pith2026250201940,
  author       = {Pith},
  title        = {Pith review of: Toward a Low-Cost Perception System in Autonomous Vehicles: A Spectrum Learning Approach},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NJF46OFD}},
  note         = {Machine review of arXiv:2502.01940}
}

read the original abstract

We present a cost-effective new approach for generating denser depth maps for Autonomous Driving (AD) and Autonomous Vehicles (AVs) by integrating the images obtained from deep neural network (DNN) 4D radar detectors with conventional camera RGB images. Our approach introduces a novel pixel positional encoding algorithm inspired by Bartlett's spatial spectrum estimation technique. This algorithm transforms both radar depth maps and RGB images into a unified pixel image subspace called the Spatial Spectrum, facilitating effective learning based on their similarities and differences. Our method effectively leverages high-resolution camera images to train radar depth map generative models, addressing the limitations of conventional radar detectors in complex vehicular environments, thus sharpening the radar output. We develop spectrum estimation algorithms tailored for radar depth maps and RGB images, a comprehensive training framework for data-driven generative models, and a camera-radar deployment scheme for AV operation. Our results demonstrate that our approach also outperforms the state-of-the-art (SOTA) by 27.95% in terms of Unidirectional Chamfer Distance (UCD).

Figures

Figures reproduced from arXiv: 2502.01940 by the authors.

**Figure 2.** The pipeline for our proposed method: (a) The offline network training scheme is divided into four modules: the predefined radar [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Results for M = 10, 70 and 200, and Φ = Θ = (−70, 70) against the original RGB scene on the left. shows the deployment pipeline for real-time operation, noting that the 4D radar and camera work independently but synchronously. A. Pixel Positional Encoding and Spectrum Estimation This encoding method aims to facilitate the transformation of radar and camera images into the spatial spectrum. Fast implementation of th… view at source ↗

Figures from the paper (6 more)

**Figure 4.** Figure 4: (a) Correlation and (b) mutual information between several depth map pairs; and (c) UCD per frame. ’SoTA’ refers to depth map [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

**Figure 5.** Figure 5: Results from the training module for four example frames. From left to right for each example frame: Scene in RGB, ground truth, [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

**Figure 6.** Figure 6: Error distributions: (a) MAE; (b) REL; (c) UCD; and (d) BCD. [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]

**Figure 7.** Figure 7: Results for different pairs of Mradar and Mcam: (a) Pearson correlation coefficient; (b) mutual information; (c) element-wise multiplication of (a) and (b) showing best values for Mradar and Mcam. Example Frame #1 Example Frame #2 Example Frame #3 Example Frame #4 [PI…

**Figure 8.** Figure 8: Results for ablation. Scene images left to right: camera RGB image, ground truth for training, spectrum estimation ablation result, [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]

**Figure 9.** Figure 9: Average MSE and SSIM between RGB images semantic [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 21 canonical work pages

[12]

Radar-camera pixel depth association for depth completion,

Y . Long, D. Morris, X. Liu, M. Castro, P. Chakravarty, and P. Narayanan, “Radar-camera pixel depth association for depth completion,” in Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 12 507–12 516

2021
[16]

RadarOcc: Robust 3D Occupancy Prediction with 4D Imaging Radar

F. Ding, X. Wen, Y . Zhu, Y . Li, and C. X. Lu, “Radarocc: Ro- bust 3d occupancy prediction with 4d imaging radar,” arXiv preprint arXiv:2405.14014, 2024

work page Pith review arXiv 2024
[1]

Multi-sensor fusion in automated driving: A survey,

Z. Wang, Y . Wu, and Q. Niu, “Multi-sensor fusion in automated driving: A survey,” Ieee Access , vol. 8, pp. 2847–2868, 2019

work page 2019
[2]

Principles of modern radar,

M. A. Richards, J. Scheer, W. A. Holm, and W. L. Melvin, “Principles of modern radar,” in Principles of Modern Radar . Citeseer, 2010, ch. 16

work page 2010
[3]

A comprehensive survey of depth completion approaches,

M. A. U. Khan, D. Nazir, A. Pagani, H. Mokayed, M. Liwicki, D. Stricker, and M. Z. Afzal, “A comprehensive survey of depth completion approaches,” Sensors, vol. 22, no. 18, p. 6969, 2022

work page 2022
[4]

Deep radar detector,

D. Brodeski, I. Bilik, and R. Giryes, “Deep radar detector,” in 2019 IEEE Radar Conference (RadarConf) . IEEE, 2019, pp. 1–6

work page 2019
[5]

A novel radar point cloud gen- eration method for robot environment perception,

Y . Cheng, J. Su, M. Jiang, and Y . Liu, “A novel radar point cloud gen- eration method for robot environment perception,” IEEE Transactions on Robotics , vol. 38, no. 6, pp. 3754–3773, 2022

work page 2022
[6]

See Further Than CFAR: a Data-Driven Radar Detector Trained by Lidar

I. Roldan, A. Palffy, J. F. Kooij, D. M. Gavrila, F. Fioranelli, and A. Yarovoy, “See further than cfar: a data-driven radar detector trained by lidar,” arXiv preprint arXiv:2402.12970 , 2024

work page Pith review arXiv 2024

Show all 27 references

[7]

Rethinking atrous convolution for semantic image segmentation,

L.-C. Chen, G. Papandreou, F. Schroff, and H. Adam, “Rethinking atrous convolution for semantic image segmentation,” arXiv preprint arXiv:1706.05587, 2017

2017 arXiv
[8]

Smoothing periodograms from time-series with contin- uous spectra,

M. S. Bartlett, “Smoothing periodograms from time-series with contin- uous spectra,” Nature, vol. 161, no. 4096, pp. 686–687, 1948

1948
[9]

Deep learning for spectrum sensing,

J. Gao, X. Yi, C. Zhong, X. Chen, and Z. Zhang, “Deep learning for spectrum sensing,” IEEE Wireless Communications Letters, vol. 8, no. 6, pp. 1727–1730, 2019

2019
[10]

Multi- class road user detection with 3+1d radar in the view-of-delft dataset,

A. Palffy, E. Pool, S. Baratam, J. F. P. Kooij, and D. M. Gavrila, “Multi- class road user detection with 3+1d radar in the view-of-delft dataset,” IEEE Robotics and Automation Letters , vol. 7, no. 2, pp. 4961–4968, 2022

2022
[11]

K-radar: 4d radar object detection for autonomous driving in various weather conditions,

D.-H. Paek, S.-H. Kong, and K. T. Wijaya, “K-radar: 4d radar object detection for autonomous driving in various weather conditions,” Ad- vances in Neural Information Processing Systems , vol. 35, pp. 3819– 3829, 2022

2022
[13]

Depth estimation from camera image and mmwave radar point cloud,

A. D. Singh, Y . Ba, A. Sarker, H. Zhang, A. Kadambi, S. Soatto, M. Srivastava, and A. Wong, “Depth estimation from camera image and mmwave radar point cloud,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023, pp. 9275–9285

2023
[14]

Radarcam-depth: Radar-camera fusion for depth estimation with learned metric scale,

H. Li, Y . Ma, Y . Gu, K. Hu, Y . Liu, and X. Zuo, “Radarcam-depth: Radar-camera fusion for depth estimation with learned metric scale,” arXiv preprint arXiv:2401.04325 , 2024

2024 arXiv
[15]

Semantic-guided depth com- pletion from monocular images and 4d radar data,

Z. Li, Y . Song, F. Ai, C. Song, and Z. Xu, “Semantic-guided depth com- pletion from monocular images and 4d radar data,” IEEE Transactions on Intelligent V ehicles, 2024

2024
[17]

Detection and localization of targets using millimeter wave radars: An experimental study,

A. Dixit and et al., “Detection and localization of targets using millimeter wave radars: An experimental study,” in 2021 IEEE International Con- ference on Electronics, Computing and Communication Technologies (CONECCT), 2021, pp. 1–6

2021
[18]

M. B. Priestley, Spectral analysis and time series . Academic press London, 1981, vol. 890

1981
[19]

Overview frequency princi- ple/spectral bias in deep learning,

Z.-Q. J. Xu, Y . Zhang, and T. Luo, “Overview frequency princi- ple/spectral bias in deep learning,” Communications on Applied Mathe- matics and Computation , pp. 1–38, 2024

2024
[20]

S. U. Pillai, Array signal processing . Springer Science & Business Media, 2012

2012
[21]

Spatial-spectrum estimation in a location sector,

K. M. Buckley and X.-L. Xu, “Spatial-spectrum estimation in a location sector,” IEEE transactions on acoustics, speech, and signal processing , vol. 38, no. 11, pp. 1842–1852, 1990

1990
[22]

Semantic segmentation with extended deeplabv3 architecture,

S. C. Yurtkulu, Y . H. S ¸ahin, and G. Unal, “Semantic segmentation with extended deeplabv3 architecture,” in 2019 27th Signal Processing and Communications Applications Conference (SIU) . IEEE, 2019, pp. 1–4

2019
[23]

The cityscapes dataset for semantic urban scene understanding,

M. Cordts, M. Omran, S. Ramos, T. Rehfeld, M. Enzweiler, R. Be- nenson, U. Franke, S. Roth, and B. Schiele, “The cityscapes dataset for semantic urban scene understanding,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2016, pp. 3213– 3223

2016
[24]

Pyramid scene parsing network,

H. Zhao, J. Shi, X. Qi, X. Wang, and J. Jia, “Pyramid scene parsing network,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2017, pp. 2881–2890

2017
[25]

Deep high-resolution repre- sentation learning for human pose estimation,

K. Sun, B. Xiao, D. Liu, and J. Wang, “Deep high-resolution repre- sentation learning for human pose estimation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2019, pp. 5693–5703

2019
[26]

The apolloscape dataset for autonomous driving,

X. Huang, X. Cheng, Q. Geng, B. Cao, D. Zhou, P. Wang, Y . Lin, and R. Yang, “The apolloscape dataset for autonomous driving,” in Proceedings of the IEEE conference on computer vision and pattern recognition workshops, 2018, pp. 954–960

2018
[27]

The pascal visual object classes (voc) challenge,

M. Everingham and et al., “The pascal visual object classes (voc) challenge,” International Journal of Computer Vision , vol. 88, no. 2, pp. 303–338, Jun. 2010

2010

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.

[1] [12]

Radar-camera pixel depth association for depth completion,

Y . Long, D. Morris, X. Liu, M. Castro, P. Chakravarty, and P. Narayanan, “Radar-camera pixel depth association for depth completion,” in Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 12 507–12 516

2021

[2] [16]

RadarOcc: Robust 3D Occupancy Prediction with 4D Imaging Radar

F. Ding, X. Wen, Y . Zhu, Y . Li, and C. X. Lu, “Radarocc: Ro- bust 3d occupancy prediction with 4d imaging radar,” arXiv preprint arXiv:2405.14014, 2024

work page Pith review arXiv 2024

[3] [1]

Multi-sensor fusion in automated driving: A survey,

Z. Wang, Y . Wu, and Q. Niu, “Multi-sensor fusion in automated driving: A survey,” Ieee Access , vol. 8, pp. 2847–2868, 2019

work page 2019

[4] [2]

Principles of modern radar,

M. A. Richards, J. Scheer, W. A. Holm, and W. L. Melvin, “Principles of modern radar,” in Principles of Modern Radar . Citeseer, 2010, ch. 16

work page 2010

[5] [3]

A comprehensive survey of depth completion approaches,

M. A. U. Khan, D. Nazir, A. Pagani, H. Mokayed, M. Liwicki, D. Stricker, and M. Z. Afzal, “A comprehensive survey of depth completion approaches,” Sensors, vol. 22, no. 18, p. 6969, 2022

work page 2022

[6] [4]

Deep radar detector,

D. Brodeski, I. Bilik, and R. Giryes, “Deep radar detector,” in 2019 IEEE Radar Conference (RadarConf) . IEEE, 2019, pp. 1–6

work page 2019

[7] [5]

A novel radar point cloud gen- eration method for robot environment perception,

Y . Cheng, J. Su, M. Jiang, and Y . Liu, “A novel radar point cloud gen- eration method for robot environment perception,” IEEE Transactions on Robotics , vol. 38, no. 6, pp. 3754–3773, 2022

work page 2022

[8] [6]

See Further Than CFAR: a Data-Driven Radar Detector Trained by Lidar

I. Roldan, A. Palffy, J. F. Kooij, D. M. Gavrila, F. Fioranelli, and A. Yarovoy, “See further than cfar: a data-driven radar detector trained by lidar,” arXiv preprint arXiv:2402.12970 , 2024

work page Pith review arXiv 2024