Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

RadarSplat: Radar Gaussian Splatting for High-Fidelity Data Synthesis and 3D Reconstruction of Autonomous Driving Scenes

T0 review · 4 major / 6 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 RadarSplat turns 2D noisy radar scans into accurate 3D reconstructions and re-renders scenes with realistic multipath.

desk verdict A solid first 3DGS for scanning radar; image synthesis is convincing, but the geometry evaluation is not fully independent of the method's own occupancy prior. read the letter →

arxiv 2506.01379 v1 pith:FWPTANDW submitted 2025-06-02 cs.CV

Pou-Chun Kung , Skanda Harisha , Ram Vasudevan , Aline Eid , Katherine A. Skinner This is my paper
classification cs.CV
keywords radarGaussiansplattingnovelviewsynthesis3Dreconstructionmultipathmodelingnoisedetectionoccupancyestimationautonomousdriving
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 introduces RadarSplat, a 3D Gaussian Splatting method that takes 2D noisy scanning-radar images and reconstructs a 3D scene while also synthesizing new radar views that include realistic noise. The authors argue that previous radar neural representations fail because they treat radar images as clean measurements; RadarSplat instead models a radar image as the sum of real target returns, multipath reflections, and noise, and learns to separate these components. On the Boreas dataset, the method reports a gain of +3.4 PSNR and more than 2.6 times the SSIM over the state of the art for image synthesis, together with a 40 percent reduction in RMSE and a 1.5 times improvement in reconstruction accuracy against LiDAR ground truth. If the approach holds, radar could serve as a weather-robust sensor for generating training data and reconstructing driving scenes without relying on optical sensors.

What carries the argument

The central object is the noise-aware Gaussian primitive: a 3D Gaussian that stores occupancy probability, noise probability, reflectivity encoded by spherical harmonics, and a power return ratio formed by combining occupancy and noise so that the two probabilities compete during training. This lets the model attribute each pixel's power either to a real surface or to multipath, saturation, and speckle noise. A multipath source map stores each detected ghost's location, activated view angle, range, reflection power, and attenuation rate, and is used to render multipath effects from novel views. The rendering stage combines elevation antenna-gain projection, a 1D azimuth convolution weighted by the azimuth antenna profile, and a Gaussian kernel that approximates spectral leakage, and the full pipeline is supervised with L1, SSIM, occupancy, size, and regularization losses.

What would settle it

Record a scene with a known large flat wall and a single strong reflector using a different scanning radar, run the method with the paper's fixed thresholds, and compare the rendered ghost position and strength against the measured multipath peak across several viewpoints; if the predicted ghost location or strength is consistently wrong, the multipath model and the claimed reconstruction improvement are falsified.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that a noisy sensor can be turned into a high-fidelity 3D representation by making noise an explicit, learnable part of the scene rather than something to be filtered out. Each 3D Gaussian carries an occupancy probability and a noise probability that together determine radar power return, so the renderer can output a clean occupancy image, a noise image, and a full noisy radar image. A preprocessing stage detects multipath and saturation beams through FFT peak and constant-ratio thresholds, fits a multipath source map, and builds a denoised occupancy map used as training supervision. The rendering pipeline projects Gaussians through the radar antenna gain pattern in elevation and azimuth and smooths along range with a Gaussian that approximates Hamming-window spectral leakage. The result on Boreas is reported as superior to Radar Fields in both synthesis and reconstruction, with the paper stating +3.4 PSNR, 2.6 times SSIM, a 40 percent lower RMSE, and 1.5 times the accuracy.

Load-bearing premise

The noise detection and denoising rely on FFT magnitude thresholds that are experimentally tuned on the Boreas radar, and the paper shows no evidence that those thresholds transfer to other sensors, so the reported gains could be specific to this dataset.

Editorial extensions

If this is right

  • Radar-only occupancy estimation reaches 0.91 accuracy at a 0.5 m threshold, versus 0.59 for Radar Fields, suggesting LiDAR-like mapping from a sensor that works in rain, fog, and snow.
  • Novel-view radar rendering reproduces multipath and noise rather than only clean images, so synthetic radar frames could augment perception training data in adverse weather.
  • Inverse rendering splits a radar image into real-target, multipath, and noise layers, making each component individually inspectable for debugging and analysis.
  • Rendering runs at about 4.5 frames per second on an A6000 GPU, putting interactive radar synthesis within reach for closed-loop simulation loops.
  • The proposed denoising and occupancy map also improves the baseline Radar Fields reconstruction when substituted in, indicating that the preprocessing alone is a step forward.

Reading between the lines

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

  • The FFT thresholds, including C_th of 0.21 and A_th of 0.3, are tuned on one Boreas radar; demonstrating that the noise decomposition works on a second scanning radar with untouched thresholds would substantially strengthen the claim that the gains are not dataset-specific.
  • The same occupancy-plus-noise Gaussian decomposition could transfer to other coherent imaging sensors such as sonar or through-wall radar, where multipath and receiver saturation dominate performance.
  • Because the method assumes static scenes, a natural extension is to attach a source velocity to each multipath source and re-render ghost targets that move with the scene, which would open up dynamic driving scenarios.
  • A strong test of the multipath model is to re-render a scene from a novel trajectory that passes close to a detected multipath source and compare the predicted ghost amplitude and position against fresh radar measurements.
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 / 6 minor

Summary. The paper proposes RadarSplat, a 3D Gaussian Splatting framework for scanning radar in autonomous driving. The method introduces radar-specific rendering components including elevation and azimuth antenna gain, spectral-leakage modeling, a noise detection and denoising pipeline for multipath and receiver saturation, and a decomposition of radar power returns into occupancy, reflectance, and noise probabilities. The authors evaluate on the Boreas dataset, reporting improved radar image synthesis (+3.4 PSNR, 2.6x SSIM) and improved geometric reconstruction (-40% RMSE, 1.5x Accuracy) relative to the Radar Fields baseline, and additionally provide radar inverse rendering capabilities.

Significance. If the reported results are robust, this is a meaningful advance: it is, to the best of my knowledge, the first Gaussian Splatting formulation for automotive scanning radar, and it explicitly models multipath and saturation noise that prior neural-radar methods ignore. The physically motivated rendering pipeline (antenna-gain projection, spectral-leakage approximation) and the inverse rendering decomposition are appealing for radar simulation and sensor-fusion training. The paper also provides thorough ablations and scene-separated results, and it identifies real limitations (static scenes, occlusion). The main weakness is that the geometry evaluation is coupled to the proposed radar-based occupancy map and to radar-derived filtering of the LiDAR ground truth, so the central quantitative claim needs additional independent validation.

major comments (4)
  1. [Sec. 3.4, Sec. 3.6 (Eq. 12), Supp. 9.1] The geometry evaluation is potentially circular. The occupancy map used as supervision (L_occ with the largest weight, lambda_3=5) is built from the proposed radar denoising pipeline (Sec. 3.4), and the LiDAR ground-truth map is filtered using a 'small radar power threshold' of 0.1 and a 1.8-degree elevation crop (Supp. 9.1). Both the training signal and the evaluation oracle are therefore derived from the same radar measurements and thresholds. The reported -40% RMSE and 1.5x Accuracy improvements may partly reflect self-consistency between the model's occupancy output and the thresholded radar-derived ground truth rather than true geometric accuracy. The paper should report geometry results against an unfiltered LiDAR map, or at minimum provide a sensitivity analysis varying the power threshold and elevation crop to show the conclusions are stable.
  2. [Sec. 4.1, Supp. 7.3] The paper does not specify whether the W=10 frame window used to construct the occupancy map includes the test frames that are held out for novel-view evaluation. If the occupancy map for a training frame is built from a window that contains test frames, then the L_occ supervision (Sec. 3.6) leaks information from held-out views into training, which would inflate both the occupancy estimation and image synthesis metrics. The authors should clarify the exact window construction and, if leakage occurs, rerun the experiments with occupancy maps built only from training frames.
  3. [Supp. Sec. 7.2, Algorithm 1] The denoising algorithm zeros out every range bin outside the first decay region (n_s, n_e) around the global maximum of each smoothed azimuth beam. This heuristic can suppress genuine weak returns located behind a strong reflector, which are precisely the low-power targets that radar is expected to detect in adverse weather. No experiment quantifies the rate of false removal, for example by comparing the denoised signal against LiDAR ground truth before occupancy mapping. Since the resulting occupancy map is the supervision for L_occ, this bias can propagate into the reconstructed geometry. The paper should quantify how many LiDAR-validated points are removed and test a variant that preserves secondary peaks.
  4. [Table 1, Supp. Sec. 8.3] The headline results in Table 1 are single-value aggregates with no error bars or per-sequence variance. The scene-separated results in Supp. Table 5 are helpful but still lack variance. In addition, the noise-detection thresholds (C_th=0.21, A_th=0.3, C'_th=0.2), the multipath view thresholds (r_th=0.5 m, theta_th=10 degrees), and the fitted multipath parameters A_m and gamma_m are all determined on the Boreas data. No sensitivity analysis is provided for these thresholds, so it is unclear whether the reported improvements generalize to other radar sensors or even to other sequences of the same type. The paper should include error bars and a threshold sweep for at least C_th and A_th to demonstrate that the results are not an artifact of tuning.
minor comments (6)
  1. [Sec. 3.5.4] The notation for the inverse rendering operators Pi_rho_alpha and Pi_rho_eta is used without definition; please define these operators explicitly (presumably replacing sigma in Eq. 11 with rho*alpha and rho*eta, respectively).
  2. [Sec. 4.3] Please clarify how the rendered polar-space occupancy I_occ is converted to the BEV occupancy map used for evaluation, and how Radar Fields' elevation-integrated occupancy is converted, since both are thresholded at 0.5 to produce the final maps.
  3. [Supp. Sec. 7.3] The occupancy mapping uses a W=10 frame window and power threshold p_th=0.15; please specify whether the window is centered on the current frame, how frames at sequence boundaries are handled, and whether the window can span the train/test split.
  4. [Eq. (10)] The noise probability eta_i is not explicitly bounded; please state its admissible range (e.g., [0,1]) and whether the min() in Eq. (10) is applied element-wise to enforce alpha_i + eta_i <= 1.
  5. [References] References [6] and [16] both use the name 'Radar Fields' but describe different radar modalities (automotive scanning radar versus synthetic aperture radar); consider renaming or disambiguating them in the text to avoid confusion.
  6. [Sec. 4.3] Typo: 'improving accuracy more than 1.5x compared' should be 'improving accuracy by more than 1.5x compared'.

Circularity Check

2 steps flagged · score 4.0 of 10

Geometry and multipath gains are partially fitted to the same radar data they are measured against; the held-out image-synthesis comparison retains independent content.

  1. fitted input called prediction [Sec. 3.4, Sec. 3.6, Table 3, Supp. 9.1.]
    "L_occ corresponds to the L1 error between the rendered occupancy state I_alpha output by RadarSplat and the initial occupancy map I_occ estimated in the preprocessing step to aid in training. ... Table 3: Init Occ. Map ... Proposed 1.81 0.04 0.90; RadarSplat ... Full Method 1.81 0.04 0.91. ... we adopt a small radar power threshold of 0.1 to remove all the LiDAR points having corresponding radar measurements below the threshold."

    The rendered occupancy I_alpha is trained with the largest loss weight (lambda_3 = 5) to match I_occ, a map produced by the paper's own denoiser from raw radar power data. The LiDAR ground truth used for the Table 1 RMSE/Accuracy metrics is itself pruned by a radar power threshold applied to the same raw radar measurements, so the target and the supervision are both functions of the same radar signal. Table 3 shows that the initial occupancy map alone attains RMSE 1.81, identical to the full RadarSplat result (1.81), so the headline -40% RMSE for '3D reconstruction' is effectively the metric of the preprocessing map, not an independent geometric prediction; any true return suppressed by the denoiser is also removed from the ground truth by the radar-power filter.

  2. fitted input called prediction [Sec. 3.3 (Eqs. 6-7), Sec. 4.2, Fig. 2.]
    "A_m and gamma_m can be estimated by fitting raw data, x[n], with least squares. ... When rendering from novel viewpoints, T_novel, we compare the new view angle and distance differences, Delta r_view and Delta theta_view, to determine if multipath effects are present. If Delta r_view < r_th and Delta theta_view < theta_th, we reconstruct multipath effects x'_m[n] from the novel view using the updated d_m and Eq. 4-7."

    The multipath source map M is built in preprocessing from the input radar images (Fig. 2), and the held-out test frames are part of those input images: the paper creates the train-test split for the learned Gaussians but never states that test frames are excluded from the multipath source fitting. Since x'_m[n] = A_m e^{-gamma_m n} x_m[n] uses A_m and gamma_m fitted by least squares to the raw beam signal x[n], a multipath source first observed at a test viewpoint is rendered by a function fitted to that same test frame's signal. The reported +3.4 PSNR therefore includes a multipath component that can be a copy of the target signal rather than a genuine novel-view prediction; the missing train/test separation for the fitted source map is what makes the claimed synthesis partially circular.

full rationale

Two partial circularities are present, both in the form of fitted radar-derived quantities being fed back into the quantities they are used to predict or evaluate. The occupancy-supervision/evaluation loop makes the geometry numbers in Table 1 close to self-consistency: the rendered occupancy is trained to match the paper's own denoised occupancy map, and the LiDAR ground truth is filtered using the same raw radar power field; Table 3 confirms the initial map alone reproduces the full method's RMSE. The multipath synthesis is also vulnerable to test-frame leakage because the source map is built from the input radar images without an explicit train/test separation, so least-squares-fitted A_m and gamma_m can be fitted to the very test beam that is later rendered and scored. The central claim is not entirely circular: the image-synthesis loss is evaluated on held-out views and the relative comparison to Radar Fields uses the same (if biased) ground-truth filtering, so the headline +3.4 PSNR and the comparison to Radar Fields retain independent content; the severe multipath/saturation modeling also goes beyond the baseline. The paper's stated limitations (static scenes, dynamic objects, occlusion) do not change this assessment. Score 4 reflects partial, not total, circularity.

Assumptions & free parameters 13 free parameters · 6 assumptions · 3 invented entities

The central performance claims depend on a large set of hand-tuned thresholds, per-source fitted parameters, and a modeling assumption that radar noise separates into independent components. These are not derived from first principles and may limit generalization.

free parameters (13)
  • C_th (saturation detection threshold) = 0.21
    Constant ratio threshold in FFT domain, experimentally determined in Sec. 3.2 and set in Sec. 8.3; controls which beams are treated as receiver saturation.
  • A_th (multipath peak magnitude threshold) = 0.3
    Peak magnitude threshold for multipath detection, set in Sec. 8.3.
  • C'_th (relaxed multipath constant ratio) = 0.2
    Relaxed constant ratio threshold used with A_th for multipath detection, set in Sec. 8.3.
  • r_th (multipath novel-view distance threshold) = 0.5 m
    If the distance change to a multipath source is below r_th, multipath is rendered from the novel view; defined in Sec. 3.3.
  • theta_th (multipath novel-view angle threshold) = 10 deg
    Angle change threshold for multipath rendering from novel views, defined in Sec. 3.3.
  • sigma_s (denoising Gaussian smoothing variance) = 5 bins
    Variance of the Gaussian filter used in the denoising algorithm (Sec. 7.2).
  • p_th (occupancy map power threshold) = 0.15
    Power threshold to binarize the occupancy grid map (Sec. 7.3).
  • W (occupancy map window) = 10 frames
    Number of frames used to reconstruct each local occupancy map (Sec. 7.3).
  • sigma_w (spectral leakage Gaussian variance) = 0.17 m
    Gaussian approximation width for Hamming-windowed range FFT, derived from Ts and mu but treated as a fixed model parameter (Sec. 6.2).
  • Q (azimuth projection upsampling factor) = 10
    Intermediate azimuth resolution multiplier for the elevation projection (Sec. 3.5.2, Sec. 8.2).
  • A_m, gamma_m (per-source multipath parameters) = fitted per source
    Power reflection and attenuation rate for each multipath source, fitted to raw data with least squares (Sec. 3.3).
  • Loss weights lambda_1..lambda_5 = 0.8, 0.2, 5, 100, 100
    Weights in the total loss Eq. 12, chosen manually in Sec. 8.3.
  • Initial Gaussian count and size = 2e4 Gaussians, 0.5 m
    Number and size of Gaussians at initialization, with alpha and eta initialized at 0.1 (Sec. 8.2, 8.3).
assumptions (6)
  • domain assumption The radar equation (Eq. 1) models received power as Pt G^2 lambda^2 sigma / ((4 pi)^3 R^4 L).
    Used as the basis for the rendering equation, ignoring lambda and L as constants.
  • ad hoc to paper Multipath and saturation manifest as periodic peaks and DC offsets in the FFT of a range-power beam.
    Heuristic foundation of the noise detection method (Sec. 3.2); no formal justification beyond observation of Figure 3 and 19.
  • ad hoc to paper Hamming-windowed range FFT spectral leakage can be approximated by a Gaussian distribution.
    Justified visually in Sec. 6.2 and Figure 15; the variance sigma_w is then fixed.
  • ad hoc to paper The power return ratio can be decomposed as sigma_i = rho_i * min(alpha_i + eta_i, 1) with occupancy and noise probabilities.
    Modeling choice in Eq. 10; enables inverse rendering but is not derived from radar physics.
  • domain assumption The Boreas dataset provides accurate ground-truth poses and LiDAR maps for evaluation.
    Used for training and evaluation; the paper excludes two snowy scenes where LiDAR is inaccurate, acknowledging limits.
  • domain assumption The scene is static during the radar collection window.
    Explicitly stated as a limitation; dynamic objects are not modeled.
invented entities (3)
  • Noise probability eta_i per Gaussian
    purpose: Disentangle real target power from noise in rendering and enable radar inverse rendering (Eq. 10, Sec. 3.5.4).
    A latent variable without direct physical measurement; its value is learned to fit the rendered image to noisy radar data.
  • Multipath source map M
    purpose: Store location, view angle, range, power reflection, and attenuation of multipath sources so ghost reflections can be rendered from novel views (Eq. 8, Sec. 3.3).
    Built from detecting peaks in training beams and fitting A_m and gamma_m; the map itself is not independently validated beyond image metrics.
  • Radar inverse rendering decomposition (I_target, I_noise, I_M)
    purpose: Decompose a radar image into real target, noise, and multipath components for analysis and simulation (Fig. 9).
    Output of the model with no external ground truth for the separate components; only the full image is supervised.

how reviews work

0 comments
Cite this review

Pith. "Pith review of RadarSplat: Radar Gaussian Splatting for High-Fidelity Data Synthesis and 3D Reconstruction of Autonomous Driving Scenes." pith.science (2026). https://pith.science/paper/FWPTANDW

@misc{pith2026250601379,
  author       = {Pith},
  title        = {Pith review of: RadarSplat: Radar Gaussian Splatting for High-Fidelity Data Synthesis and 3D Reconstruction of Autonomous Driving Scenes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FWPTANDW}},
  note         = {Machine review of arXiv:2506.01379}
}
read the original abstract

High-Fidelity 3D scene reconstruction plays a crucial role in autonomous driving by enabling novel data generation from existing datasets. This allows simulating safety-critical scenarios and augmenting training datasets without incurring further data collection costs. While recent advances in radiance fields have demonstrated promising results in 3D reconstruction and sensor data synthesis using cameras and LiDAR, their potential for radar remains largely unexplored. Radar is crucial for autonomous driving due to its robustness in adverse weather conditions like rain, fog, and snow, where optical sensors often struggle. Although the state-of-the-art radar-based neural representation shows promise for 3D driving scene reconstruction, it performs poorly in scenarios with significant radar noise, including receiver saturation and multipath reflection. Moreover, it is limited to synthesizing preprocessed, noise-excluded radar images, failing to address realistic radar data synthesis. To address these limitations, this paper proposes RadarSplat, which integrates Gaussian Splatting with novel radar noise modeling to enable realistic radar data synthesis and enhanced 3D reconstruction. Compared to the state-of-the-art, RadarSplat achieves superior radar image synthesis (+3.4 PSNR / 2.6x SSIM) and improved geometric reconstruction (-40% RMSE / 1.5x Accuracy), demonstrating its effectiveness in generating high-fidelity radar data and scene reconstruction. A project page is available at https://umautobots.github.io/radarsplat.

Figures

Figures reproduced from arXiv: 2506.01379 by the authors.

Figure 1
Figure 1. RadarSplat enables radar 2D-to-3D scene reconstruction, image synthesis, and occupancy estimation. RadarSplat outperforms [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. System Overview. RadarSplat takes radar images and poses as input. The preprocessing step includes noise detection and initial [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Three types of radar noise of scanning radar highlighted [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (17 more)
Figure 4
Figure 4. Figure 4: Range-power signal and its FFT of a radar azimuth beam [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Multipath modeling and denoising. The multipath effect is modeled by peak frequency and source power reflection and attenu [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Our proposed radar image denoising method preserves [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: Modeling spectral leakage in the radar-power signal. (a) [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 10
Figure 10. Figure 10: Occupancy estimation. Our method provides clear and [PITH_FULL_IMAGE:figures/full_fig_p007_10.png]
Figure 9
Figure 9. Figure 9: Radar inverse rendering for radar image decomposition. [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]
Figure 11
Figure 11. Figure 11: Comparison of 3D reconstructions from our proposed method RadarSplat and Radar Fields [ [PITH_FULL_IMAGE:figures/full_fig_p008_11.png]
Figure 12
Figure 12. Figure 12: Ablation studies on image synthesis. RadarSplat fails [PITH_FULL_IMAGE:figures/full_fig_p008_12.png]
Figure 13
Figure 13. Figure 13: Robust radar occupancy estimation in extreme condi [PITH_FULL_IMAGE:figures/full_fig_p008_13.png]
Figure 14
Figure 14. Figure 14: Radar imaging with range fast Fourier transform (FFT). [PITH_FULL_IMAGE:figures/full_fig_p012_14.png]
Figure 16
Figure 16. Figure 16: Illustration of blurred range-power signal due to spec [PITH_FULL_IMAGE:figures/full_fig_p013_16.png]
Figure 15
Figure 15. Figure 15: Modeling spectral leakage in the radar image. (a) Ideal [PITH_FULL_IMAGE:figures/full_fig_p013_15.png]
Figure 18
Figure 18. Figure 18: Antenna gain plot. 7. Noise Detection, Denoising, and Occupancy Mapping 7.1. Multipath and Saturation Noise Detection [PITH_FULL_IMAGE:figures/full_fig_p013_18.png]
Figure 19
Figure 19. Figure 19: Illustration of the range-power signal of multipath, receiver saturation, and noise-free azimuth beam and their FFT. Also the [PITH_FULL_IMAGE:figures/full_fig_p014_19.png]
Figure 20
Figure 20. Figure 20: RadarSplat rendering. We first project Gaussians to 2D using the elevation antenna gain, then apply the azimuth antenna gain [PITH_FULL_IMAGE:figures/full_fig_p016_20.png]
Figure 21
Figure 21. Figure 21: Additional results on Boreas Dataset. t = T Occluded Occluded Non-Occluded t = T ⋅ 10 t = T ⋅ 5 Missing Reconstruction Ground-Truth LiDAR Final Occupancy Estimation Radar Input Frames GT Reconstruction [PITH_FULL_IMAGE:figures/full_fig_p017_21.png]
Figure 22
Figure 22. Figure 22: Wrong occupancy estimation caused by the occlusion problem. [PITH_FULL_IMAGE:figures/full_fig_p017_22.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Millimeter Wave Radar: From Synthetic Aperture to Probabilistic Mapping

    cs.RO 2026-07 accept novelty 6.0 of 10

    A SAR-plus-Rayleigh pipeline converts raw cascaded mmWave radar signals into occupancy maps that outperform CFAR and range-azimuth baselines on indoor geometry fidelity and A* planning success.

Reference graph

Works this paper leans on

67 extracted references · 48 canonical work pages · cited by 1 Pith paper

  1. [1]

    Daniel Adolfsson, Martin Magnusson, Anas Alhashimi, Achim J Lilienthal, and Henrik Andreasson. Lidar-level lo- calization with radar? the cfear approach to accurate, fast, and robust large-scale radar odometry in diverse environ- ments.IEEE Transactions on robotics, 39(2):1476–1495,

    work page

  2. [2]

    Raysar- 3d sar simulator: Now open source

    Stefan Auer, Richard Bamler, and Peter Reinartz. Raysar- 3d sar simulator: Now open source. In2016 IEEE In- ternational Geoscience and Remote Sensing Symposium (IGARSS), pages 6730–6733. IEEE, 2016. 3

    work page 2016

  3. [3]

    The oxford radar robotcar dataset: A radar extension to the oxford robotcar dataset

    Dan Barnes, Matthew Gadd, Paul Murcutt, Paul Newman, and Ingmar Posner. The oxford radar robotcar dataset: A radar extension to the oxford robotcar dataset. In2020 IEEE international conference on robotics and automation (ICRA), pages 6433–6438. IEEE, 2020. 2

    work page 2020

  4. [4]

    Environmental Research Laboratories, 1973

    Donald E Barrick.FM/CW radar signals and digital pro- cessing. Environmental Research Laboratories, 1973. 12

    work page 1973

  5. [5]

    Artech House, 2013

    David Knox Barton.Radar equations for modern radar. Artech House, 2013. 3

    work page 2013

  6. [6]

    Radar fields: Frequency-space neural scene representa- tions for fmcw radar

    David Borts, Erich Liang, Tim Broedermann, Andrea Ra- mazzina, Stefanie Walz, Edoardo Palladin, Jipeng Sun, David Brueggemann, Christos Sakaridis, Luc Van Gool, et al. Radar fields: Frequency-space neural scene representa- tions for fmcw radar. InACM SIGGRAPH 2024 Conference Papers, pages 1–10, 2024. 1, 2, 4, 5, 6, 7, 8, 14, 16, 17

    work page 2024

  7. [7]

    Are we ready for radar to replace lidar in all-weather mapping and localization?IEEE Robotics and Automation Letters, 7(4):10328–10335, 2022

    Keenan Burnett, Yuchen Wu, David J Yoon, Angela P Schoellig, and Timothy D Barfoot. Are we ready for radar to replace lidar in all-weather mapping and localization?IEEE Robotics and Automation Letters, 7(4):10328–10335, 2022. 2

    work page 2022

  8. [8]

    Boreas: A multi-season autonomous driving dataset.The International Journal of Robotics Research, 42(1-2):33–42, 2023

    Keenan Burnett, David J Yoon, Yuchen Wu, Andrew Z Li, Haowei Zhang, Shichen Lu, Jingxing Qian, Wei-Kang Tseng, Andrew Lambert, Keith YK Leung, et al. Boreas: A multi-season autonomous driving dataset.The International Journal of Robotics Research, 42(1-2):33–42, 2023. 2, 6, 13

    work page 2023

Show all 67 references
  1. [9]

    Cloner: Camera-lidar fu- sion for occupancy grid-aided neural representations.IEEE Robotics and Automation Letters, 8(5):2812–2819, 2023

    Alexandra Carlson, Manikandasriram S Ramanagopal, Nathan Tseng, Matthew Johnson-Roberson, Ram Vasude- van, and Katherine A Skinner. Cloner: Camera-lidar fu- sion for occupancy grid-aided neural representations.IEEE Robotics and Automation Letters, 8(5):2812–2819, 2023. 2

    2023

  2. [10]

    Lidar-gs: Real-time lidar re-simulation us- ing gaussian splatting.arXiv preprint arXiv:2410.05111,

    Qifeng Chen, Sheng Yang, Sicong Du, Tao Tang, Peng Chen, and Yuchi Huo. Lidar-gs: Real-time lidar re-simulation us- ing gaussian splatting.arXiv preprint arXiv:2410.05111,

  3. [11]

    Periodic vibration gaussian: Dynamic urban scene reconstruction and real-time rendering.arXiv preprint arXiv:2311.18561, 2023

    Yurui Chen, Chun Gu, Junzhe Jiang, Xiatian Zhu, and Li Zhang. Periodic vibration gaussian: Dynamic urban scene reconstruction and real-time rendering.arXiv preprint arXiv:2311.18561, 2023. 2

    2023 arXiv

  4. [12]

    A ray tracing formulation and its application to some problems in over-the-horizon radar.Radio Science, 33 (4):1187–1197, 1998

    CJ Coleman. A ray tracing formulation and its application to some problems in over-the-horizon radar.Radio Science, 33 (4):1187–1197, 1998. 3

    1998

  5. [13]

    Letsgo: Large-scale garage modeling and ren- dering via lidar-assisted gaussian primitives.arXiv preprint arXiv:2404.09748, 2024

    Jiadi Cui, Junming Cao, Yuhui Zhong, Liao Wang, Fuqiang Zhao, Penghao Wang, Yifan Chen, Zhipeng He, Lan Xu, Yu- jiao Shi, et al. Letsgo: Large-scale garage modeling and ren- dering via lidar-assisted gaussian primitives.arXiv preprint arXiv:2404.09748, 2024. 2

    2024 arXiv

  6. [14]

    Isar-nerf: Neural radiance fields for 3-d imaging of space target from multiview isar images.IEEE Sensors Journal, 24(7):11705–11722, 2024

    Junyuan Deng, Pengfei Xie, Lei Zhang, and Yulai Cong. Isar-nerf: Neural radiance fields for 3-d imaging of space target from multiview isar images.IEEE Sensors Journal, 24(7):11705–11722, 2024. 2

    2024

  7. [15]

    Yaw aided radar inertial odometry using manhattan world assumptions

    Christopher Doer and Gert F Trommer. Yaw aided radar inertial odometry using manhattan world assumptions. In 2021 28th Saint Petersburg International Conference on Integrated Navigation Systems (ICINS), pages 1–9. IEEE,

    2021

  8. [16]

    Radar fields: An extension of radiance fields to sar

    Thibaud Ehret, Roger Mar ´ı, Dawa Derksen, Nicolas Gas- nier, and Gabriele Facciolo. Radar fields: An extension of radiance fields to sar. InProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 564–574, 2024. 2

    2024

  9. [17]

    Simple-bev: What really mat- ters for multi-sensor bev perception? In2023 IEEE Inter- national Conference on Robotics and Automation (ICRA), pages 2759–2765

    Adam W Harley, Zhaoyuan Fang, Jie Li, Rares Ambrus, and Katerina Fragkiadaki. Simple-bev: What really mat- ters for multi-sensor bev perception? In2023 IEEE Inter- national Conference on Robotics and Automation (ICRA), pages 2759–2765. IEEE, 2023. 2

    2023

  10. [18]

    A new wave in robotics: Sur- vey on recent mmwave radar applications in robotics.IEEE Transactions on Robotics, 2024

    Kyle Harlow, Hyesu Jang, Timothy D Barfoot, Ayoung Kim, and Christoffer Heckman. A new wave in robotics: Sur- vey on recent mmwave radar applications in robotics.IEEE Transactions on Robotics, 2024. 2

    2024

  11. [19]

    A ray launching approach for modeling an fmcw radar system

    Nils Hirsenkorn, Paul Subkowski, Timo Hanke, Alexander Schaermann, Andreas Rauch, Ralph Rasshofer, and Erwin Biebl. A ray launching approach for modeling an fmcw radar system. In2017 18th International Radar Symposium (IRS), pages 1–10. IEEE, 2017. 3

  12. [20]

    Liv-gaussmap: Lidar-inertial-visual fusion for real-time 3d radiance field map rendering.IEEE Robotics and Automation Letters, 2024

    Sheng Hong, Junjie He, Xinhu Zheng, Chunran Zheng, and Shaojie Shen. Liv-gaussmap: Lidar-inertial-visual fusion for real-time 3d radiance field map rendering.IEEE Robotics and Automation Letters, 2024. 2

    2024

  13. [21]

    Radar slam: A robust slam system for all weather conditions.arXiv preprint arXiv:2104.05347, 2021

    Ziyang Hong, Yvan Petillot, Andrew Wallace, and Sen Wang. Radar slam: A robust slam system for all weather conditions.arXiv preprint arXiv:2104.05347, 2021. 2

    2021 arXiv

  14. [22]

    Neural lidar fields for novel view synthesis

    Shengyu Huang, Zan Gojcic, Zian Wang, Francis Williams, Yoni Kasten, Sanja Fidler, Konrad Schindler, and Or Litany. Neural lidar fields for novel view synthesis. InProceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 18236–18246, 2023. 2

    2023

  15. [23]

    Dart: Im- plicit doppler tomography for radar novel view synthesis

    Tianshu Huang, John Miller, Akarsh Prabhakara, Tao Jin, Tarana Laroia, Zico Kolter, and Anthony Rowe. Dart: Im- plicit doppler tomography for radar novel view synthesis. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 24118–24129, 2024. 2

    2024

  16. [24]

    Cramnet: Camera-radar fusion with ray-constrained cross-attention for robust 3d object detec- tion

    Jyh-Jing Hwang, Henrik Kretzschmar, Joshua Manela, Sean Rafferty, Nicholas Armstrong-Crews, Tiffany Chen, and Dragomir Anguelov. Cramnet: Camera-radar fusion with ray-constrained cross-attention for robust 3d object detec- tion. InEuropean conference on computer vision, pages ...

    2022

  17. [25]

    Loner: Lidar only neu- ral representations for real-time slam.IEEE Robotics and Automation Letters, 2023

    Seth Isaacson, Pou-Chun Kung, Mani Ramanagopal, Ram Vasudevan, and Katherine A Skinner. Loner: Lidar only neu- ral representations for real-time slam.IEEE Robotics and Automation Letters, 2023. 2

    2023

  18. [26]

    Li-gs: Gaussian splatting with lidar 9 incorporated for accurate large-scale reconstruction.arXiv preprint arXiv:2409.12899, 2024

    Changjian Jiang, Ruilan Gao, Kele Shao, Yue Wang, Rong Xiong, and Yu Zhang. Li-gs: Gaussian splatting with lidar 9 incorporated for accurate large-scale reconstruction.arXiv preprint arXiv:2409.12899, 2024. 2

    2024 arXiv

  19. [27]

    3d gaussian splatting for real-time radiance field rendering.ACM Trans

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimk ¨uhler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering.ACM Trans. Graph., 42(4):139–1,

  20. [28]

    Autosplat: Constrained gaussian splatting for autonomous driving scene reconstruction.arXiv preprint arXiv:2407.02598, 2024

    Mustafa Khan, Hamidreza Fazlali, Dhruv Sharma, Tongtong Cao, Dongfeng Bai, Yuan Ren, and Bingbing Liu. Autosplat: Constrained gaussian splatting for autonomous driving scene reconstruction.arXiv preprint arXiv:2407.02598, 2024. 1

    2024 arXiv

  21. [29]

    Mulran: Multimodal range dataset for urban place recognition

    Giseop Kim, Yeong Sang Park, Younghun Cho, Jinyong Jeong, and Ayoung Kim. Mulran: Multimodal range dataset for urban place recognition. 2

  22. [30]

    Hercules: Heterogeneous radar dataset in complex ur- ban environment for multi-session radar slam.arXiv preprint arXiv:2502.01946, 2025

    Hanjun Kim, Minwoo Jung, Chiyun Noh, Sangwoo Jung, Hyunho Song, Wooseong Yang, Hyesu Jang, and Ayoung Kim. Hercules: Heterogeneous radar dataset in complex ur- ban environment for multi-session radar slam.arXiv preprint arXiv:2502.01946, 2025. 2

    2025 arXiv

  23. [31]

    A normal distribution transform-based radar odometry de- signed for scanning and automotive radars

    Pou-Chun Kung, Chieh-Chih Wang, and Wen-Chieh Lin. A normal distribution transform-based radar odometry de- signed for scanning and automotive radars. In2021 IEEE In- ternational Conference on Robotics and Automation (ICRA), pages 14417–14423. IEEE, 2021. 2

    2021

  24. [32]

    Radar occupancy prediction with lidar supervision while preserving long-range sensing and penetrating capabilities

    Pou-Chun Kung, Chieh-Chih Wang, and Wen-Chieh Lin. Radar occupancy prediction with lidar supervision while preserving long-range sensing and penetrating capabilities. IEEE Robotics and Automation Letters, 7(2):2637–2643,

  25. [33]

    Sad-gs: Shape-aligned depth- supervised gaussian splatting

    Pou-Chun Kung, Seth Isaacson, Ram Vasudevan, and Katherine A Skinner. Sad-gs: Shape-aligned depth- supervised gaussian splatting. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2842–2851, 2024. 2

    2024

  26. [34]

    Lihi-gs: Lidar-supervised gaussian splat- ting for highway driving scene reconstruction.arXiv preprint arXiv:2412.15447, 2024

    Pou-Chun Kung, Xianling Zhang, Katherine A Skinner, and Nikita Jaipuria. Lihi-gs: Lidar-supervised gaussian splat- ting for highway driving scene reconstruction.arXiv preprint arXiv:2412.15447, 2024. 1, 2, 5

    2024

  27. [35]

    Sar-nerf: Neural radiance fields for synthetic aperture radar multi-view representation.IEEE Transactions on Geoscience and Remote Sensing, 2024

    Zhengxin Lei, Feng Xu, Jiangtao Wei, Feng Cai, Feng Wang, and Ya-Qiu Jin. Sar-nerf: Neural radiance fields for synthetic aperture radar multi-view representation.IEEE Transactions on Geoscience and Remote Sensing, 2024. 2

    2024

  28. [36]

    Ranerf: Neural 3-d reconstruction of space tar- gets from isar image sequences.IEEE Transactions on Geo- science and Remote Sensing, 61:1–15, 2023

    Afei Liu, Shuanghui Zhang, Chi Zhang, Shuaifeng Zhi, and Xiang Li. Ranerf: Neural 3-d reconstruction of space tar- gets from isar image sequences.IEEE Transactions on Geo- science and Remote Sensing, 61:1–15, 2023. 2

    2023

  29. [37]

    Nerf: Representing scenes as neural radiance fields for view syn- thesis.Communications of the ACM, 65(1):99–106, 2021

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoorthi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view syn- thesis.Communications of the ACM, 65(1):99–106, 2021. 2

    2021

  30. [38]

    Neural scene graphs for dynamic scenes

    Julian Ost, Fahim Mannan, Nils Thuerey, Julian Knodt, and Felix Heide. Neural scene graphs for dynamic scenes. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 2856–2865, 2021. 2

    2021

  31. [39]

    K-radar: 4d radar object detection for autonomous driving in various weather conditions.Advances in Neural Information Processing Systems, 35:3819–3829, 2022

    Dong-Hee Paek, Seung-Hyun Kong, and Kevin Tirta Wijaya. K-radar: 4d radar object detection for autonomous driving in various weather conditions.Advances in Neural Information Processing Systems, 35:3819–3829, 2022. 2

    2022

  32. [40]

    Transloc4d: Transformer-based 4d radar place recognition

    Guohao Peng, Heshan Li, Yangyang Zhao, Jun Zhang, Zhenyu Wu, Pengyu Zheng, and Danwei Wang. Transloc4d: Transformer-based 4d radar place recognition. InProceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17595–17605, 2024. 2

    2024

  33. [41]

    Neural implicit surface reconstruction using imaging sonar

    Mohamad Qadri, Michael Kaess, and Ioannis Gkioulekas. Neural implicit surface reconstruction using imaging sonar. In2023 IEEE International Conference on Robotics and Au- tomation (ICRA), pages 1040–1047. IEEE, 2023. 2

    2023

  34. [42]

    Z-splat: Z-axis gaussian splatting for camera-sonar fusion.IEEE Transactions on Pat- tern Analysis and Machine Intelligence, 2024

    Ziyuan Qu, Omkar Vengurlekar, Mohamad Qadri, Kevin Zhang, Michael Kaess, Christopher Metzler, Suren Jaya- suriya, and Adithya Pediredla. Z-splat: Z-axis gaussian splatting for camera-sonar fusion.IEEE Transactions on Pat- tern Analysis and Machine Intelligence, 2024. 2

    2024

  35. [43]

    Neural vol- umetric reconstruction for coherent synthetic aperture sonar

    Albert Reed, Juhyeon Kim, Thomas Blanford, Adithya Pediredla, Daniel Brown, and Suren Jayasuriya. Neural vol- umetric reconstruction for coherent synthetic aperture sonar. ACM Transactions on Graphics (TOG), 42(4):1–20, 2023. 2

    2023

  36. [44]

    Urban radiance fields

    Konstantinos Rematas, Andrew Liu, Pratul P Srini- vasan, Jonathan T Barron, Andrea Tagliasacchi, Thomas Funkhouser, and Vittorio Ferrari. Urban radiance fields. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 12932–12942, 2022. 2

    2022

  37. [45]

    Radar cfar thresholding in clutter and multiple target situations.IEEE transactions on aerospace and electronic systems, (4):608–621, 1983

    Hermann Rohling. Radar cfar thresholding in clutter and multiple target situations.IEEE transactions on aerospace and electronic systems, (4):608–621, 1983. 3

    1983

  38. [46]

    Sonarsplat: Novel view synthesis of imaging sonar via gaussian splatting.arXiv preprint arXiv:2504.00159, 2025

    Advaith V Sethuraman, Max Rucker, Onur Bagoren, Pou- Chun Kung, Nibarkavi NB Amutha, and Katherine A Skin- ner. Sonarsplat: Novel view synthesis of imaging sonar via gaussian splatting.arXiv preprint arXiv:2504.00159, 2025. 2

    2025

  39. [47]

    Radiate: A radar dataset for automotive perception in bad weather

    Marcel Sheeny, Emanuele De Pellegrin, Saptarshi Mukher- jee, Alireza Ahrabian, Sen Wang, and Andrew Wallace. Radiate: A radar dataset for automotive perception in bad weather. In2021 IEEE International Conference on Robotics and Automation (ICRA), pages 1–7. IEEE, 2021. 2

    2021

  40. [48]

    Li- rafusion: Deep adaptive lidar-radar fusion for 3d object de- tection

    Jingyu Song, Lingjun Zhao, and Katherine A Skinner. Li- rafusion: Deep adaptive lidar-radar fusion for 3d object de- tection. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 18250–18257. IEEE, 2024. 2

    2024

  41. [49]

    Lidar-nerf: Novel li- dar view synthesis via neural radiance fields

    Tang Tao, Longfei Gao, Guangrun Wang, Yixing Lao, Peng Chen, Hengshuang Zhao, Dayang Hao, Xiaodan Liang, Mathieu Salzmann, and Kaicheng Yu. Lidar-nerf: Novel li- dar view synthesis via neural radiance fields. InProceedings of the 32nd ACM International Conference on Multimedia...

    2024

  42. [50]

    Neurad: Neural rendering for autonomous driving

    Adam Tonderski, Carl Lindstr ¨om, Georg Hess, William Ljungbergh, Lennart Svensson, and Christoffer Petersson. Neurad: Neural rendering for autonomous driving. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14895–14904, 2024. 1, 2

    2024

  43. [51]

    Bi-lrfusion: Bi-directional lidar-radar fusion for 3d dynamic object detection

    Yingjie Wang, Jiajun Deng, Yao Li, Jinshui Hu, Cong Liu, Yu Zhang, Jianmin Ji, Wanli Ouyang, and Yanyong Zhang. Bi-lrfusion: Bi-directional lidar-radar fusion for 3d dynamic object detection. InProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition...

    2023

  44. [52]

    High- dynamic radar sequence prediction for weather nowcasting using spatiotemporal coherent gaussian representation.arXiv preprint arXiv:2502.14895, 2025

    Ziye Wang, Yiran Qin, Lin Zeng, and Ruimao Zhang. High- dynamic radar sequence prediction for weather nowcasting using spatiotemporal coherent gaussian representation.arXiv preprint arXiv:2502.14895, 2025. 2

    2025 arXiv

  45. [53]

    Dynamic lidar re- simulation using compositional neural fields

    Hanfeng Wu, Xingxing Zuo, Stefan Leutenegger, Or Litany, Konrad Schindler, and Shengyu Huang. Dynamic lidar re- simulation using compositional neural fields. InProceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 19988–19998, 2024. 2

    2024

  46. [54]

    Mars: An instance-aware, mod- ular and realistic simulator for autonomous driving

    Zirui Wu, Tianyu Liu, Liyi Luo, Zhide Zhong, Jianteng Chen, Hongmin Xiao, Chao Hou, Haozhe Lou, Yuantao Chen, Runyi Yang, et al. Mars: An instance-aware, mod- ular and realistic simulator for autonomous driving. InCAAI International Conference on Artificial Intelligence, pages 3–

  47. [55]

    Depthsplat: Connecting gaussian splatting and depth.arXiv preprint arXiv:2410.13862, 2024

    Haofei Xu, Songyou Peng, Fangjinhua Wang, Hermann Blum, Daniel Barath, Andreas Geiger, and Marc Pollefeys. Depthsplat: Connecting gaussian splatting and depth.arXiv preprint arXiv:2410.13862, 2024. 2

    2024 arXiv

  48. [56]

    Street gaussians for modeling dynamic ur- ban scenes.arXiv preprint arXiv:2401.01339, 2024

    Yunzhi Yan, Haotong Lin, Chenxu Zhou, Weijie Wang, Haiyang Sun, Kun Zhan, Xianpeng Lang, Xiaowei Zhou, and Sida Peng. Street gaussians for modeling dynamic ur- ban scenes.arXiv preprint arXiv:2401.01339, 2024. 2, 8, 17

    2024 arXiv

  49. [57]

    Emernerf: Emergent spatial-temporal scene decomposition via self-supervision.arXiv preprint arXiv:2311.02077, 2023

    Jiawei Yang, Boris Ivanovic, Or Litany, Xinshuo Weng, Se- ung Wook Kim, Boyi Li, Tong Che, Danfei Xu, Sanja Fidler, Marco Pavone, et al. Emernerf: Emergent spatial-temporal scene decomposition via self-supervision.arXiv preprint arXiv:2311.02077, 2023. 2

    2023 arXiv

  50. [58]

    Unisim: A neural closed-loop sensor simulator

    Ze Yang, Yun Chen, Jingkang Wang, Sivabalan Mani- vasagam, Wei-Chiu Ma, Anqi Joyce Yang, and Raquel Ur- tasun. Unisim: A neural closed-loop sensor simulator. In Proceedings of the IEEE/CVF Conference on Computer Vi- sion and Pattern Recognition, pages 1389–1399, 2023. 1, 2

    2023

  51. [59]

    Tclc-gs: Tightly coupled lidar-camera gaussian splatting for surrounding autonomous driving scenes.arXiv preprint arXiv:2404.02410, 2024

    Cheng Zhao, Su Sun, Ruoyu Wang, Yuliang Guo, Jun-Jun Wan, Zhou Huang, Xinyu Huang, Yingjie Victor Chen, and Liu Ren. Tclc-gs: Tightly coupled lidar-camera gaussian splatting for surrounding autonomous driving scenes.arXiv preprint arXiv:2404.02410, 2024. 2

    2024 arXiv

  52. [60]

    Crkd: Enhanced camera-radar object detection with cross-modality knowledge distillation

    Lingjun Zhao, Jingyu Song, and Katherine A Skinner. Crkd: Enhanced camera-radar object detection with cross-modality knowledge distillation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15470–15480, 2024. 2

    2024

  53. [61]

    Lidar4d: Dynamic neural fields for novel space-time view lidar synthesis

    Zehan Zheng, Fan Lu, Weiyi Xue, Guang Chen, and Changjun Jiang. Lidar4d: Dynamic neural fields for novel space-time view lidar synthesis. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5145–5154, 2024. 2

    2024

  54. [62]

    Drivinggaussian: Composite gaussian splatting for surrounding dynamic au- tonomous driving scenes

    Xiaoyu Zhou, Zhiwei Lin, Xiaojun Shan, Yongtao Wang, Deqing Sun, and Ming-Hsuan Yang. Drivinggaussian: Composite gaussian splatting for surrounding dynamic au- tonomous driving scenes. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2...

    2024

  55. [63]

    Frequency Modulated Continuous Wave (FMCW) Radar In FMCW radar systems, the transmitted signal is a lin- ear frequency-modulated chirp

    Radar Sensing Primer 6.1. Frequency Modulated Continuous Wave (FMCW) Radar In FMCW radar systems, the transmitted signal is a lin- ear frequency-modulated chirp. The most common chirp is with the sawtooth pattern. The designed chirp slope is related to the bandwidthBand chirp ...

  56. [64]

    Multipath and Saturation Noise Detection Figure 19 shows the range-power signal and FFT of mul- tipath, receiver saturation, and noise-free azimuth beams

    Noise Detection, Denoising, and Occupancy Mapping 7.1. Multipath and Saturation Noise Detection Figure 19 shows the range-power signal and FFT of mul- tipath, receiver saturation, and noise-free azimuth beams. The saturation beam is detected when constant ratioC > Cth. The mul...

  57. [65]

    RadarSplat Implementation Details 8.1. Input Format We set the maximum range of the input radar image to 50 m, with an azimuth resolution of 0.9° and a range resolu- Algorithm 1Denoising algorithm with Decay Regions in Radar Range-Power Data Require:P(n)(1D array of radar powe...

  58. [66]

    Scene Reconstruction Evaluation We construct a LiDAR pointcloud map to obtain ground- truth geometry for evaluation

    Evaluation Details and Extra Evaluation 9.1. Scene Reconstruction Evaluation We construct a LiDAR pointcloud map to obtain ground- truth geometry for evaluation. Similar to building an oc- cupancy map, we use a W-frame window to reconstruct a local map for each frame, preventi...

  59. [67]

    Limitations Occlusion Problem. Although radar provides bird-eye- view (BEV) power images with radar waves penetrating and bouncing off to see through occluded objects, occlusion can still happen in the radar image if the objects have high re- flectivity. Figure 22 illustrates ...

Pith tools

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