The reviewed record of science sign in
Pith

arxiv: 2605.29097 · v1 · pith:YWI4BE3D · submitted 2026-05-27 · cs.CV

GeRaF: Neural Geometry Reconstruction from Radio Frequency Signals

Jiachen Lu , Hailan Shanbhag , Haitham Al Hassanieh This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-06-29 12:54 UTCgrok-4.3pith:YWI4BE3Drecord.jsonopen to challenge →

classification cs.CV
keywords neural implicit learningradio frequency sensing3D geometry reconstructionvolumetric renderinglensless imagingsigned distance functionsphysics-based rendering
0
0 comments X

The pith

GeRaF shows neural implicit learning can recover millimeter-scale 3D geometry from noisy radio frequency signals.

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

The paper sets out to prove that RF signals can support near-range 3D geometry reconstruction by adapting neural implicit representations, even though the signals propagate through entire volumes and produce specular reflections. It tackles the resulting noise and cubic sampling cost with three targeted changes: filter-based rendering, a physics-based volumetric pipeline, and lensless sampling paired with lensless alpha blending. A sympathetic reader cares because RF penetrates occlusions that block cameras and LiDAR, so successful reconstruction would enable geometry recovery in settings where optical methods are blind. The method learns signed distance functions together with reflectiveness and signal power inside MLPs and trainable parameters.

Core claim

GeRaF is the first method to apply neural implicit learning to near-range 3D geometry reconstruction from RF signals. It introduces filter-based rendering to suppress irrelevant signals, implements a physics-based RF volumetric rendering pipeline, and proposes a novel lensless sampling and lensless alpha blending strategy that makes full-space sampling feasible during training. By learning signed distance functions, reflectiveness, and signal power through MLPs and trainable parameters, the approach targets millimeter-level geometry recovery in real-world settings.

What carries the argument

The lensless sampling and lensless alpha blending strategy inside a physics-based RF volumetric rendering pipeline, which suppresses noise from full-space propagation while learning signed distance functions.

If this is right

  • Full-space RF sampling becomes computationally feasible during training without prohibitive cubic complexity.
  • Millimeter-level geometry can be recovered despite the low resolution and noise inherent to lensless RF imaging.
  • Reconstruction remains possible in environments containing occlusions because RF signals penetrate surfaces.
  • Specular reflection behavior is handled directly by the physics-based modeling rather than optical ray assumptions.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Successful millimeter recovery would enable 3D mapping tasks in fully occluded indoor or outdoor scenes where cameras and LiDAR cannot operate.
  • The same lensless volumetric approach could be tested on other penetrating wave modalities such as ultrasound or low-frequency sonar.
  • If the MLP evaluation can be accelerated, the method opens a route to real-time geometry updates from continuously collected RF data.

Load-bearing premise

The filter-based rendering, physics-based volumetric pipeline, and lensless sampling and alpha blending can suppress noise and artifacts sufficiently to permit accurate millimeter-level geometry recovery from RF signals.

What would settle it

An experiment that removes the lensless alpha blending strategy and measures whether the resulting reconstructions deviate from ground truth by more than one millimeter in controlled near-range RF scenes would settle the central claim.

Figures

Figures reproduced from arXiv: 2605.29097 by Hailan Shanbhag, Haitham Al Hassanieh, Jiachen Lu.

Figure 1
Figure 1. Figure 1: GeRaF is the first method to reconstruct millimeter-level geometry of non-line-of-sight ob￾jects using radio frequency (RF) signals by us￾ing neural implicit representations. Right, shows the radar heatmap from one scan, a camera scan, the surface directly reconstructed from the radar heatmap images, and surface output from GeRaF. However, their systems are designed for large￾scale environments and are far… view at source ↗
Figure 2
Figure 2. Figure 2: (a) Frequency Modulated Continuous Waveform, (b) the reflected signal is a delayed [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: a) Comparison between lens-based imaging and lensless imaging models. b) Matched filter [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Overview of the GeRaF framework. (1) Lensless sampling replaces ray-based methods. (2) A neural implicit model predicts geometry, reflectivity, and power. (3) RF volumetric rendering simulates physical signal propagation. (4) Matched filtering produces MF power images (heatmaps). (5) An L2 loss compares the rendered and ground truth power for end-to-end training. 4 Overview Given raw RF signals and the pos… view at source ↗
Figure 5
Figure 5. Figure 5: Left: Naïve sampling traces rays across the entire space for each antenna, resulting in the highest computational complexity. Middle: Lensless sampling strategy samples points along the radar’s primary ray direction and reuses network predictions across antennas. Alpha blending and transmittance are calculated using lens-less alpha blending. Right: Signal Tracing Bank: subset of antennas is processed in ea… view at source ↗
Figure 6
Figure 6. Figure 6: Left: Experimental results of objects with nothing between the radar and the object. Right: Results when the object is occluded from the radar. The camera scan done in line of sight. computational burden by leveraging the fact that the matched filter contains no trainable parameters. This allows us to reuse signal tracing results of most antennas from previous iterations. During training, we implement a me… view at source ↗
Figure 7
Figure 7. Figure 7: Solid lines indicate latency (time per iteration), while dashed lines show GPU memory usage. Green represents the configuration on patch antenna size in the Signal Tracing Bank. Orange represents the ray number, and “w.o. lens-less” refers to disabling lens-less sampling, whose mem￾ory usage is too large to measure experimentally and is instead deduced from theory. Purple repre￾sents the number of sample p… view at source ↗
Figure 8
Figure 8. Figure 8: Rendered heatmaps compared to real heatmaps. Though the output of GeRaF is perhaps, not as detailed as the camera scans, this is primarily because our scans are limited to a single pitch axis, limiting the amount of reflec￾tions we can receive back from the objects, this limitation is discussed in more detail in Appendix E [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Novel view synthesis on four selected novel planes. We visualize the 3D matched filter [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Ablation study on the number of radar scanning planes. F1 denotes the F1-Score and CD denotes the Chamfer Distance. ↑ indicates higher values are better, while ↓ indicates lower values are better. We study the effect of the number of temporal samples Nt on geometry reconstruction quality. According to the matched filter equation: P(x) = [PITH_FULL_IMAGE:figures/full_fig_p011_10.png] view at source ↗
Figure 12
Figure 12. Figure 12: Top: Ablation study on the number of ray samples Nray. Bottom: Ablation study on the number of depth samples Nz. Lat denotes latency, and MEM denotes GPU memory usage. Low frequency levels (e.g., 2 0 , 2 1 ) capture coarse, smooth variations, while high frequency levels (e.g., 2 9 , 2 10) cap￾ture fine, high-frequency details such as sharp edges or thin structures. The number of frequency levels Nlevel co… view at source ↗
Figure 11
Figure 11. Figure 11: Top: Ablation study on the number of temporal samples Nt. Bottom: Ablation study on the positional encoding frequency level Nlevel. Lat denotes latency, and MEM denotes GPU memory usage. For the number of sampling rays, results show that increas￾ing the number of rays leads to higher memory usage and longer latency, but surprisingly worse reconstruction performance. A possible explanation is that in radio… view at source ↗
Figure 13
Figure 13. Figure 13: Ablation study on dynamic loss masking. "w.o." denotes without masking, and "w." denotes with mask￾ing. In Sec. 7, we discuss the motivation for using dynamic loss masking. The primary reason is the inherent ambi￾guity in interpreting low-power regions in matched filter (MF) imaging. Specifically, a weak signal response in the rendered radar image may arise from two distinct causes: (1) low volume density… view at source ↗
Figure 14
Figure 14. Figure 14: Ablation study on radar physics. "Without radar physics" indicates that the Lambertian [PITH_FULL_IMAGE:figures/full_fig_p013_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Difference between far-field and near￾field rays. On the left, it’s shown that the incoming rays can be approximated as nearly parallel, while on the right, when the object is close to the anten￾nas, we can no longer make this assumption. following the scaled Lambertian model men￾tioned above. A.4 Near Field vs. Far Field Wireless imaging systems operate differently depending on the distance between the s… view at source ↗
Figure 16
Figure 16. Figure 16: Difference between processing radar data captured in near-field with matched filter vs beamforming. On the other hand, in this paper, we are operating in the near-field. This mean the object is much closer to the antenna aperture, within a meter in our case. Im￾portantly, this means that the incoming waves that are reflected from the object in the scene can no longer be approximated as parallel. The diffe… view at source ↗
Figure 17
Figure 17. Figure 17: Subset of matched filter images used as ground truth in [PITH_FULL_IMAGE:figures/full_fig_p022_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Novel view synthesis setting. The blue planes represent novel views that are not included during training. We also demonstrate GeRaF’s capability in novel view synthesis (NVS) [39]. To evaluate this, we split the avail￾able views into training and test sets [PITH_FULL_IMAGE:figures/full_fig_p023_18.png] view at source ↗
read the original abstract

GeRaF is the first method to use neural implicit learning for near-range 3D geometry reconstruction from radio frequency (RF) signals. Unlike RGB or LiDAR-based methods, RF sensing can see through occlusion but suffers from low resolution and noise due to its lensless imaging nature. While lenses in RGB imaging constrain sampling to 1D rays, RF signals propagate through the entire space, introducing significant noise and leading to cubic complexity in volumetric rendering. Moreover, RF signals interact with surfaces via specular reflections, requiring fundamentally different modeling. To address these challenges, GeRaF (1) introduces filter-based rendering to suppress irrelevant signals, (2) implements a physics-based RF volumetric rendering pipeline, and (3) proposes a novel lensless sampling and lensless alpha blending strategy that makes full-space sampling feasible during training. By learning signed distance functions, reflectiveness, and signal power through MLPs and trainable parameters, GeRaF takes the first step towards reconstructing millimeter-level geometry from RF signals in real-world settings.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The manuscript introduces GeRaF as the first neural implicit method for near-range 3D geometry reconstruction from RF signals. It identifies challenges of lensless propagation (noise, cubic complexity, specular reflections) and proposes three components—filter-based rendering to suppress irrelevant signals, a physics-based RF volumetric rendering pipeline, and a lensless sampling/alpha-blending strategy—to enable full-space sampling and SDF learning. The method parameterizes signed distance functions, reflectiveness, and signal power via MLPs and trainable parameters.

Significance. If the three proposed components jointly achieve the required noise and artifact suppression, the result would be significant: it would demonstrate the first viable neural-implicit pipeline for millimeter-scale geometry recovery from RF, extending implicit representations beyond optical modalities into occluded, lensless sensing scenarios with potential applications in robotics and through-wall imaging.

major comments (1)
  1. [Abstract] Abstract (and central claim): the assertion that filter-based rendering, physics-based volumetric rendering, and lensless alpha blending together suppress noise/artifacts sufficiently for millimeter-level SDF recovery is load-bearing, yet the provided text contains no quantitative evidence (error metrics, ablation results, or ground-truth comparisons) demonstrating that residual RF interactions do not prevent precise geometry recovery.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed review and the identification of a key point regarding the abstract. We address the concern below.

read point-by-point responses

  1. Referee: [Abstract] Abstract (and central claim): the assertion that filter-based rendering, physics-based volumetric rendering, and lensless alpha blending together suppress noise/artifacts sufficiently for millimeter-level SDF recovery is load-bearing, yet the provided text contains no quantitative evidence (error metrics, ablation results, or ground-truth comparisons) demonstrating that residual RF interactions do not prevent precise geometry recovery.

    Authors: We agree that the abstract, being a high-level summary, does not embed specific quantitative metrics or ablation tables. The full manuscript contains the supporting evidence: Section 4 reports mean absolute distance errors on the order of 1-3 mm against ground-truth meshes, with ablation studies (Table 2) quantifying the contribution of each component to noise suppression and artifact reduction, and direct comparisons showing that residual specular and multipath effects are mitigated sufficiently for SDF convergence at millimeter scale. To make this explicit in the abstract, we will add a concise clause referencing the achieved reconstruction accuracy. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on novel proposed components without reduction to inputs

full rationale

The provided abstract and description introduce three new technical components (filter-based rendering, physics-based RF volumetric rendering, and lensless sampling/alpha blending) to address RF-specific challenges, then learn SDFs and related quantities via MLPs. No equations, fitted parameters renamed as predictions, or self-citations are shown that would make any claimed result equivalent to its inputs by construction. The central claim of millimeter-level reconstruction is presented as enabled by these independent methodological contributions rather than tautological definitions or load-bearing prior self-work, rendering the chain self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on domain assumptions about RF signal propagation and the effectiveness of the three proposed adaptations; no free parameters or invented entities are explicitly quantified in the abstract.

axioms (2)
  • domain assumption RF signals propagate through the entire space rather than along constrained rays, introducing cubic complexity and noise.
    Stated directly in the abstract as the core challenge of lensless RF imaging.
  • domain assumption Specular reflections and low-resolution noisy measurements can be mitigated by filter-based rendering and lensless alpha blending.
    The abstract presents these as the solutions that make the method feasible.

pith-pipeline@v0.9.1-grok · 5710 in / 1255 out tokens · 27310 ms · 2026-06-29T12:54:38.947278+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

72 extracted references · 1 canonical work pages

  1. [1]

    Capturing the human figure through a wall.ACM Transactions on Graphics (TOG), 2015

    Fadel Adib, Chen-Yu Hsu, Hongzi Mao, Dina Katabi, and Frédo Durand. Capturing the human figure through a wall.ACM Transactions on Graphics (TOG), 2015

    2015

  2. [2]

    See through walls with wifi! InSIGCOMM, 2013

    Fadel Adib and Dina Katabi. See through walls with wifi! InSIGCOMM, 2013

    2013

  3. [3]

    Fast fourier methods for synthetic aperture radar imaging.IEEE Transactions on Aerospace and Electronic Systems, 2012

    Fredrik Andersson, Randolph Moses, and Frank Natterer. Fast fourier methods for synthetic aperture radar imaging.IEEE Transactions on Aerospace and Electronic Systems, 2012

    2012

  4. [4]

    Multi-view 3d surface reconstruction from sar images by inverse rendering, 2025

    Emile Barbier-Renard, Florence Tupin, Nicolas Trouvé, and Loïc Denis. Multi-view 3d surface reconstruction from sar images by inverse rendering, 2025

    2025

  5. [5]

    Conventional beamforming techniques

    Jacob Benesty, Jingdong Chen, and Yiteng Huang. Conventional beamforming techniques. Microphone array signal processing, 2008

    2008

  6. [6]

    Radar fields: Frequency-space neural scene representations for fmcw radar

    David Borts, Erich Liang, Tim Broedermann, Andrea Ramazzina, Stefanie Walz, Edoardo Palladin, Jipeng Sun, David Brueggemann, Christos Sakaridis, Luc Van Gool, et al. Radar fields: Frequency-space neural scene representations for fmcw radar. InSIGGRAPH, 2024

    2024

  7. [7]

    Nerd: Neural reflectance decomposition from image collections

    Mark Boss, Raphael Braun, Varun Jampani, Jonathan T Barron, Ce Liu, and Hendrik Lensch. Nerd: Neural reflectance decomposition from image collections. InICCV, 2021

    2021

  8. [8]

    Neusg: Neural implicit surface reconstruction with 3d gaussian splatting guidance.arXiv preprint, 2023

    Hanlin Chen, Chen Li, and Gim Hee Lee. Neusg: Neural implicit surface reconstruction with 3d gaussian splatting guidance.arXiv preprint, 2023

    2023

  9. [9]

    Radio frequency ray tracing with neural object representation.arXiv preprint, 2024

    Xingyu Chen, Zihao Feng, Kun Qian, and Xinyu Zhang. Radio frequency ray tracing with neural object representation.arXiv preprint, 2024

    2024

  10. [10]

    MMDetection3D: OpenMMLab next-generation platform for general 3D object detection.https://github.com/open-mmlab/mmdetection3d, 2020

    MMDetection3D Contributors. MMDetection3D: OpenMMLab next-generation platform for general 3D object detection.https://github.com/open-mmlab/mmdetection3d, 2020

    2020

  11. [11]

    Torrance

    Robert L Cook and Kenneth E. Torrance. A reflectance model for computer graphics.ACM Transactions on Graphics (TOG), 1982

    1982

  12. [12]

    Mito: A millimeter-wave dataset and simulator for non-line-of-sight perception.arXiv preprint, 2025

    Laura Dodds, Tara Boroushaki, Cusuh Ham, and Fadel Adib. Mito: A millimeter-wave dataset and simulator for non-line-of-sight perception.arXiv preprint, 2025

    2025

  13. [13]

    Around the corner mmwave imaging in practical environments

    Laura Dodds, Hailan Shanbhag, Junfeng Guan, Saurabh Gupta, and Haitham Hassanieh. Around the corner mmwave imaging in practical environments. InProceedings of the 30th Annual International Conference on Mobile Computing and Networking, pages 953–967, 2024

    2024

  14. [14]

    Digital twin catalog: A large-scale photorealistic 3d object digital twin dataset

    Zhao Dong, Ka Chen, Zhaoyang Lv, Hong-Xing Yu, Yunzhi Zhang, Cheng Zhang, Yufeng Zhu, Stephen Tian, Zhengqin Li, Geordie Moffatt, et al. Digital twin catalog: A large-scale photorealistic 3d object digital twin dataset. InCVPR, 2025

    2025

  15. [15]

    On the shape of a set of points in the plane.IEEE Transactions on information theory, 1983

    Herbert Edelsbrunner, David Kirkpatrick, and Raimund Seidel. On the shape of a set of points in the plane.IEEE Transactions on information theory, 1983

    1983

  16. [16]

    Multi-view stereo: A tutorial.Foundations and trends® in Computer Graphics and Vision, 2015

    Yasutaka Furukawa, Carlos Hernández, et al. Multi-view stereo: A tutorial.Foundations and trends® in Computer Graphics and Vision, 2015

    2015

  17. [17]

    Accurate, dense, and robust multiview stereopsis.IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2009

    Yasutaka Furukawa and Jean Ponce. Accurate, dense, and robust multiview stereopsis.IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2009

    2009

  18. [18]

    Relightable 3d gaussians: Realistic point cloud relighting with brdf decomposition and ray tracing

    Jian Gao, Chun Gu, Youtian Lin, Zhihao Li, Hao Zhu, Xun Cao, Li Zhang, and Yao Yao. Relightable 3d gaussians: Realistic point cloud relighting with brdf decomposition and ray tracing. InECCV, 2024. 14

    2024

  19. [19]

    Irgs: Inter-reflective gaussian splatting with 2d gaussian ray tracing

    Chun Gu, Xiaofei Wei, Zixuan Zeng, Yuxuan Yao, and Li Zhang. Irgs: Inter-reflective gaussian splatting with 2d gaussian ray tracing. InCVPR, 2025

    2025

  20. [20]

    Through fog high-resolution imaging using millimeter wave radar

    Junfeng Guan, Sohrab Madani, Suraj Jog, Saurabh Gupta, and Haitham Hassanieh. Through fog high-resolution imaging using millimeter wave radar. InCVPR, 2020

    2020

  21. [21]

    Articulatedgs: Self- supervised digital twin modeling of articulated objects using 3d gaussian splatting.arXiv preprint, 2025

    Junfu Guo, Yu Xin, Gaoyi Liu, Kai Xu, Ligang Liu, and Ruizhen Hu. Articulatedgs: Self- supervised digital twin modeling of articulated objects using 3d gaussian splatting.arXiv preprint, 2025

    2025

  22. [22]

    Carto: Category and joint agnostic reconstruction of articulated objects

    Nick Heppert, Muhammad Zubair Irshad, Sergey Zakharov, Katherine Liu, Rares Andrei Ambrus, Jeannette Bohg, Abhinav Valada, and Thomas Kollar. Carto: Category and joint agnostic reconstruction of articulated objects. InCVPR, 2023

    2023

  23. [23]

    2d gaussian splatting for geometrically accurate radiance fields

    Binbin Huang, Zehao Yu, Anpei Chen, Andreas Geiger, and Shenghua Gao. 2d gaussian splatting for geometrically accurate radiance fields. InSIGGRAPH, 2024

    2024

  24. [24]

    Dart: Implicit doppler tomography for radar novel view synthesis

    Tianshu Huang, John Miller, Akarsh Prabhakara, Tao Jin, Tarana Laroia, Zico Kolter, and Anthony Rowe. Dart: Implicit doppler tomography for radar novel view synthesis. InCVPR, 2024

    2024

  25. [25]

    Gaussianshader: 3d gaussian splatting with shading functions for reflective surfaces

    Yingwenqi Jiang, Jiadong Tu, Yuan Liu, Xifeng Gao, Xiaoxiao Long, Wenping Wang, and Yuexin Ma. Gaussianshader: 3d gaussian splatting with shading functions for reflective surfaces. InCVPR, 2024

    2024

  26. [26]

    Tensoir: Tensorial inverse rendering

    Haian Jin, Isabella Liu, Peijia Xu, Xiaoshuai Zhang, Songfang Han, Sai Bi, Xiaowei Zhou, Zexiang Xu, and Hao Su. Tensoir: Tensorial inverse rendering. InCVPR, 2023

    2023

  27. [27]

    Poisson surface reconstruction

    Michael Kazhdan, Matthew Bolitho, and Hugues Hoppe. Poisson surface reconstruction. In Proceedings of the fourth Eurographics symposium on Geometry processing, 2006

    2006

  28. [28]

    3d gaussian splatting for real-time radiance field rendering.ACM Transactions on Graphics (TOG), 2023

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering.ACM Transactions on Graphics (TOG), 2023

    2023

  29. [29]

    Robot see robot do: Imitating articulated object manipulation with monocu- lar 4d reconstruction

    Justin Kerr, Chung Min Kim, Mingxuan Wu, Brent Yi, Qianqian Wang, Ken Goldberg, and Angjoo Kanazawa. Robot see robot do: Imitating articulated object manipulation with monocu- lar 4d reconstruction. InConference on Robot Learning (CoRL), 2024

    2024

  30. [30]

    Xmodal-id: Using wifi for through-wall person identification from candidate video footage

    Belal Korany, Chitra R Karanam, Hong Cai, and Yasamin Mostofi. Xmodal-id: Using wifi for through-wall person identification from candidate video footage. InMobiCom, 2019

    2019

  31. [31]

    Enabling visual recognition at radio frequency

    Haowen Lai, Gaoxiang Luo, Yifei Liu, and Mingmin Zhao. Enabling visual recognition at radio frequency. InMobiCom, 2024

    2024

  32. [32]

    Single view garment reconstruction using diffusion mapping via pattern coordinates

    Ren Li, Cong Cao, Corentin Dumery, Yingxuan You, Hao Li, and Pascal Fua. Single view garment reconstruction using diffusion mapping via pattern coordinates. InSIGGRAPH, 2025

    2025

  33. [33]

    Gs-ir: 3d gaussian splatting for inverse rendering

    Zhihao Liang, Qi Zhang, Ying Feng, Ying Shan, and Kui Jia. Gs-ir: 3d gaussian splatting for inverse rendering. InCVPR, 2024

    2024

  34. [34]

    Echoes beyond points: Unleashing the power of raw radar data in multi-modality fusion

    Yang Liu, Feng Wang, Naiyan Wang, and Zhaoxiang Zhang. Echoes beyond points: Unleashing the power of raw radar data in multi-modality fusion. InNeurIPS, 2023

    2023

  35. [35]

    Nero: Neural geometry and brdf reconstruction of reflective objects from multiview images.ACM Transactions on Graphics (TOG), 2023

    Yuan Liu, Peng Wang, Cheng Lin, Xiaoxiao Long, Jiepeng Wang, Lingjie Liu, Taku Komura, and Wenping Wang. Nero: Neural geometry and brdf reconstruction of reflective objects from multiview images.ACM Transactions on Graphics (TOG), 2023

    2023

  36. [36]

    Newrf: A deep learning framework for wireless radiation field reconstruction and channel prediction

    Haofan Lu, Christopher Vattheuer, Baharan Mirzasoleiman, and Omid Abari. Newrf: A deep learning framework for wireless radiation field reconstruction and channel prediction. InICML, 2024

    2024

  37. [37]

    Rada- tron: Accurate detection using multi-resolution cascaded mimo radar

    Sohrab Madani, Jayden Guan, Waleed Ahmed, Saurabh Gupta, and Haitham Hassanieh. Rada- tron: Accurate detection using multi-resolution cascaded mimo radar. InECCV, 2022. 15

    2022

  38. [38]

    Ganesh: Generalizable nerf for lensless imaging

    Rakesh Raj Madavan, Akshat Kaimal, Vinayak Gupta, Rohit Choudhary, Chandrakala Shanmu- ganathan, Kaushik Mitra, et al. Ganesh: Generalizable nerf for lensless imaging. InWACV, 2025

    2025

  39. [39]

    Nerf: Representing scenes as neural radiance fields for view synthesis

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoor- thi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view synthesis. Communications of the ACM, 2021

    2021

  40. [40]

    Instant neural graphics primitives with a multiresolution hash encoding.ACM Transactions on Graphics (TOG), 2022

    Thomas Müller, Alex Evans, Christoph Schied, and Alexander Keller. Instant neural graphics primitives with a multiresolution hash encoding.ACM Transactions on Graphics (TOG), 2022

    2022

  41. [41]

    Extracting triangular 3d models, materials, and lighting from images

    Jacob Munkberg, Jon Hasselgren, Tianchang Shen, Jun Gao, Wenzheng Chen, Alex Evans, Thomas Müller, and Sanja Fidler. Extracting triangular 3d models, materials, and lighting from images. InCVPR, 2022

    2022

  42. [42]

    Differentiable volumetric rendering: Learning implicit 3d representations without 3d supervision

    Michael Niemeyer, Lars Mescheder, Michael Oechsle, and Andreas Geiger. Differentiable volumetric rendering: Learning implicit 3d representations without 3d supervision. InCVPR, 2020

    2020

  43. [43]

    Diffuse scattering directive model parameterization method for construction materials at mmwave frequencies.International Journal of Antennas and Propagation, 2020

    Ming-Hao Ren, Xi Liao, Jihua Zhou, Yang Wang, Yu Shao, Shasha Liao, and Jie Zhang. Diffuse scattering directive model parameterization method for construction materials at mmwave frequencies.International Journal of Antennas and Propagation, 2020

    2020

  44. [44]

    Principles of modern radar

    Mark A Richards, Jim Scheer, William A Holm, and William L Melvin. Principles of modern radar. 2010

    2010

  45. [45]

    Three-dimensional millimeter- wave imaging for concealed weapon detection.IEEE Transactions on microwave theory and techniques, 2001

    David M Sheen, Douglas L McMakin, and Thomas E Hall. Three-dimensional millimeter- wave imaging for concealed weapon detection.IEEE Transactions on microwave theory and techniques, 2001

    2001

  46. [46]

    Nerv: Neural reflectance and visibility fields for relighting and view synthesis

    Pratul P Srinivasan, Boyang Deng, Xiuming Zhang, Matthew Tancik, Ben Mildenhall, and Jonathan T Barron. Nerv: Neural reflectance and visibility fields for relighting and view synthesis. InCVPR, 2021

    2021

  47. [47]

    3drimr: 3d reconstruc- tion and imaging via mmwave radar based on deep learning

    Yue Sun, Zhuoming Huang, Honggang Zhang, Zhi Cao, and Deqiang Xu. 3drimr: 3d reconstruc- tion and imaging via mmwave radar based on deep learning. InIEEE International Performance, Computing, and Communications Conference (IPCCC), 2021

    2021

  48. [48]

    3d reconstruction of multiple objects by mmwave radar on uav

    Yue Sun, Zhuoming Huang, Honggang Zhang, and Xiaohui Liang. 3d reconstruction of multiple objects by mmwave radar on uav. In2022 IEEE 19th International Conference on Mobile Ad Hoc and Smart Systems (MASS), 2022

    2022

  49. [49]

    R2p: A deep learning model from mmwave radar to point cloud

    Yue Sun, Honggang Zhang, Zhuoming Huang, and Benyuan Liu. R2p: A deep learning model from mmwave radar to point cloud. InInternational Conference on Artificial Neural Networks, 2022

    2022

  50. [50]

    Spinr: Neural volumetric reconstruction for fmcw radars.arXiv preprint, 2025

    Harshvardhan Takawale and Nirupam Roy. Spinr: Neural volumetric reconstruction for fmcw radars.arXiv preprint, 2025

    2025

  51. [51]

    Neural geometric level of detail: Real-time rendering with implicit 3d shapes

    Towaki Takikawa, Joey Litalien, Kangxue Yin, Karsten Kreis, Charles Loop, Derek Nowrouzezahrai, Alec Jacobson, Morgan McGuire, and Sanja Fidler. Neural geometric level of detail: Real-time rendering with implicit 3d shapes. InCVPR, 2021

    2021

  52. [52]

    Fast 3d reconstruction of space targets from isar image sequences based on neural network

    Weiyi Tan, Yu Wang, Biao Tian, and Shiyou Xu. Fast 3d reconstruction of space targets from isar image sequences based on neural network. InIEEE 8th International Conference on Vision, Image and Signal Processing (ICVISP), 2024

    2024

  53. [53]

    Texas instrument awr1843.https://www.ti.com/product/AWR1843, 2023

    TI Inc. Texas instrument awr1843.https://www.ti.com/product/AWR1843, 2023

    2023

  54. [54]

    Ref-nerf: Structured view-dependent appearance for neural radiance fields

    Dor Verbin, Peter Hedman, Ben Mildenhall, Todd Zickler, Jonathan T Barron, and Pratul P Srinivasan. Ref-nerf: Structured view-dependent appearance for neural radiance fields. In CVPR, 2022. 16

    2022

  55. [55]

    Neus: Learning neural implicit surfaces by volume rendering for multi-view reconstruction

    Peng Wang, Lingjie Liu, Yuan Liu, Christian Theobalt, Taku Komura, and Wenping Wang. Neus: Learning neural implicit surfaces by volume rendering for multi-view reconstruction. In NeurIPS, 2021

    2021

  56. [56]

    Safe for generations to come: Considerations of safety for millimeter waves in wireless communications.IEEE microwave magazine, 2015

    Ting Wu, Theodore S Rappaport, and Christopher M Collins. Safe for generations to come: Considerations of safety for millimeter waves in wireless communications.IEEE microwave magazine, 2015

    2015

  57. [57]

    Near-field 2-d sar imaging by millimeter-wave radar for concealed item detection

    Muhammet Emin Yanik and Murat Torlak. Near-field 2-d sar imaging by millimeter-wave radar for concealed item detection. In2019 IEEE radio and Wireless Symposium (RWS), 2019

    2019

  58. [58]

    Mvsnet: Depth inference for unstructured multi-view stereo

    Yao Yao, Zixin Luo, Shiwei Li, Tian Fang, and Long Quan. Mvsnet: Depth inference for unstructured multi-view stereo. InECCV, 2018

    2018

  59. [59]

    Neilf: Neural incident light field for physically-based material estimation

    Yao Yao, Jingyang Zhang, Jingbo Liu, Yihang Qu, Tian Fang, David McKinnon, Yanghai Tsin, and Long Quan. Neilf: Neural incident light field for physically-based material estimation. In ECCV, 2022

    2022

  60. [60]

    Reflective gaussian splatting

    Yuxuan Yao, Zixuan Zeng, Chun Gu, Xiatian Zhu, and Li Zhang. Reflective gaussian splatting. InICLR, 2025

    2025

  61. [61]

    V olume rendering of neural implicit surfaces

    Lior Yariv, Jiatao Gu, Yoni Kasten, and Yaron Lipman. V olume rendering of neural implicit surfaces. InNeurIPS, 2021

    2021

  62. [62]

    Multiview neural surface reconstruction by disentangling geometry and appearance

    Lior Yariv, Yoni Kasten, Dror Moran, Meirav Galun, Matan Atzmon, Basri Ronen, and Yaron Lipman. Multiview neural surface reconstruction by disentangling geometry and appearance. InNeurIPS, 2020

    2020

  63. [63]

    Fast-mvsnet: Sparse-to-dense multi-view stereo with learned propagation and gauss-newton refinement

    Zehao Yu and Shenghua Gao. Fast-mvsnet: Sparse-to-dense multi-view stereo with learned propagation and gauss-newton refinement. InCVPR, 2020

    2020

  64. [64]

    A modular robotic arm control stack for research: Franka-interface and frankapy.arXiv preprint arXiv:2011.02398, 2020

    Kevin Zhang, Mohit Sharma, Jacky Liang, and Oliver Kroemer. A modular robotic arm control stack for research: Franka-interface and frankapy.arXiv preprint arXiv:2011.02398, 2020

    work page arXiv 2011

  65. [65]

    Through-wall human pose estimation using radio signals

    Mingmin Zhao, Tianhong Li, Mohammad Abu Alsheikh, Yonglong Tian, Hang Zhao, Antonio Torralba, and Dina Katabi. Through-wall human pose estimation using radio signals. InCVPR, 2018

    2018

  66. [66]

    Through-wall human mesh recovery using radio signals

    Mingmin Zhao, Yingcheng Liu, Aniruddh Raghu, Tianhong Li, Hang Zhao, Antonio Torralba, and Dina Katabi. Through-wall human mesh recovery using radio signals. InICCV, 2019

    2019

  67. [67]

    Rf-based 3d skeletons

    Mingmin Zhao, Yonglong Tian, Hang Zhao, Mohammad Abu Alsheikh, Tianhong Li, Rumen Hristov, Zachary Kabelac, Dina Katabi, and Antonio Torralba. Rf-based 3d skeletons. In Proceedings of the Conference of the ACM Special Interest Group on Data Communication, 2018

    2018

  68. [68]

    Nerf2: Neural radio-frequency radiance fields

    Xiaopeng Zhao, Zhenlin An, Qingrui Pan, and Lei Yang. Nerf2: Neural radio-frequency radiance fields. InMobiCom, 2023. 17 Appendix This appendix is organized as follows: • Section A provides more extensive radar background information. • Section B provides the proof for our lenless alpha blending. • Section C describes in detail the experimental parameters...

    2023

  69. [69]

    , N ant −1,t= 0,

    Forward:Input signals(i, t)wherei= 0, . . . , N ant −1,t= 0, . . . , N t −1; sampling pointsx j wherej= 0, . . . , N rayNs −1 for Parallelj= 0, . . . , N rayNs −1do P(x j)←0 fori= 0, . . . , N ant −1do fort= 0, . . . , N t −1do P(x j)←P(x j) +s(i, t)·e j2πkτ it ·e j2πf τi P(x j)← ∥P(x j)∥ Output:P(x j)for allj

  70. [70]

    , N rayNs −1; Input signals(i, t)wherei= 0,

    Backward:Gradient of matched filter power ∂L ∂P(x j) forj= 0, . . . , N rayNs −1; Input signals(i, t)wherei= 0, . . . , N ant −1,t= 0, . . . , N t −1; Sampling pointsx j wherej= 0, . . . , N rayNs −1 for Paralleli= 0, . . . , N ant −1do for Parallelt= 0, . . . , N t −1do ∂L ∂s(i,t) ←0 forj= 0, . . . , N rayNs −1do ∂L ∂s(i,t) ← ∂L ∂s(i,t) + 1 P(x j) · ∂L ∂...

  71. [71]

    , N rayNs −1; AmplitudeA rx(xj)wherej= 0,

    Forward:Sampling pointsx j wherej= 0, . . . , N rayNs −1; AmplitudeA rx(xj)wherej= 0, . . . , N rayNs −1 for Paralleli= 0, . . . , N ant −1do for Parallelt= 0, . . . , N t −1do s(i, t)←0 forj= 0, . . . , N rayNs −1do s(i, t)←s(i, t) +A rx(x)·e −j2πkτ it ·e −j2πf τi Output:s(i, t)for alli, t

  72. [72]

    , N ant −1,t= 0,

    Backward:Gradient of signal ∂L ∂s(i,t) fori= 0, . . . , N ant −1,t= 0, . . . , N t −1 Sampling pointsx j wherej= 0, . . . , N rayNs −1 for Parallelj= 0, . . . , N rayNs −1do ∂L ∂Arx(xj) ←0 fori= 0, . . . , N ant −1do fort= 0, . . . , N t −1do ∂L ∂Arx(xj) ← ∂L ∂Arx(xj) + ∂L ∂s(i,t) ·e j2πkτ it ·e j2πf τi Output: ∂L ∂Arx(xj) for allj 20 B Lensless Alpha Ble...