Pith. sign in

REVIEW 5 major objections 7 minor 37 references

3D Face Reconstruction From Radar Images

T0 review · 5 major / 7 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Radar images alone can reconstruct 3D faces: a model-based autoencoder reaches 2.56 mm mean point error on synthetic faces and separates identities on real radar data.

desk verdict First radar-image 3D face reconstruction with a learned renderer, credible synthetic results, but the real-data and finetuning claims need sharper evidence. read the letter →

arxiv 2412.02403 v2 pith:ZCHFTMEY submitted 2024-12-03 cs.CV cs.LG

Valentin Braeutigam , Vanessa Wirth , Ingrid Ullmann , Christian Schü{ss}ler , Martin Vossiek , Matthias Berking , Bernhard Egger This is my paper · ORCID
classification cs.CVcs.LG
keywords 3DfacereconstructionmmWaveradarimagingmorphablemodeldifferentiablerenderinganalysis-by-synthesisrecognitionfromsynthetictrainingdatasleepmonitoring
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

Radar images can reveal the 3D geometry of a face even though no optical camera is involved, and this paper tries to establish that fact with a parametric reconstruction pipeline. The authors generate 10,000 synthetic radar images from a statistical 3D face model using a physics-based ray-tracing renderer, then train a CNN encoder to predict the model's shape, expression, and pose parameters. To make the loop differentiable and fast, they replace the non-differentiable physics renderer with a learned neural renderer and train the whole encoder-plus-renderer as a model-based autoencoder, minimizing both parameter error and image reconstruction error. On synthetic faces this reaches a mean Euclidean point distance of 2.56 mm between reconstructed and ground-truth meshes, and on real radar images of four individuals, instances of the same identity with varied pose and expression are more similar to each other than to other faces. If correct, this means radar, which penetrates fabric and works in darkness, can support 3D face analysis and monitoring where optical sensors cannot.

What carries the argument

The central object is a statistical 3D morphable face model (the BFM 2019 face12 mask), whose Gaussian-sampled shape vector and expression vector define each face instance within a compact latent space. Around that model the method wraps an analysis-by-synthesis loop: a physics-based radar ray-tracing renderer creates synthetic amplitude and depth training images; a convolutional encoder predicts the shape, expression, and pose parameters; a reversed convolutional network acts as a learned differentiable radar renderer that maps parameters back to an image; and the combined autoencoder is trained with a loss of the form $L_{\text{train}} = L_{\text{image}} + \lambda L_{\text{params}}$. The learned renderer is the load-bearing replacement for the non-differentiable physics simulator, and it is what makes unsupervised test-time optimization on the image loss possible.

What would settle it

Collect real radar images with photogrammetry ground truth from a diverse set of subjects (different sexes, skin tones, ages, facial hair, and headwear) and compute the mean Euclidean distance between reconstructed and ground-truth meshes; if identity similarity in the cosine matrices drops to chance for any group, or real-mesh error greatly exceeds the synthetic 2.56 mm baseline, the simulator-transfer assumption fails. A cheaper check is to compare simulated and real radar images of the same photogrammetry mesh and measure whether the pattern and scale mismatch grows with skin properties.

Watch

Extended reading notes

Core claim

The paper's central claim, stated in its conclusion, is that a model-based encoder plus a learned differentiable renderer can reconstruct identity-relevant face geometry from radar images alone. On synthetic data the autoencoder achieves a mean Euclidean 3D point distance of 2.56 mm on the face meshes, and on real data it performs face recognition, because instances of the same shape with different pose and expression have higher similarity than other faces. The autoencoder variant consistently beats the fully supervised encoder on shape and expression parameter error, which the authors attribute to the image-reconstruction loss regularizing the parameter space. The learned renderer is both differentiable and more than 2000 times faster than the physics-based simulator, so the parameters can be further optimized at test time by fixing both networks and minimizing only the image loss.

Load-bearing premise

The load-bearing premise is that the physics-based radar ray-tracing simulator reproduces the real 72-82 GHz near-field backscatter of skin and fabric closely enough that networks trained entirely on synthetic radar images transfer to real radar captures; the paper itself acknowledges a synthetic-to-real domain gap in patterns and scale and validates only on four male European faces.

Editorial extensions

If this is right

  • Adding the learned renderer and image-reconstruction loss reduces shape and expression parameter error compared with fully supervised training, with the best synthetic mesh error at 2.56 mm using depth input.
  • Depth radar images, alone or combined with amplitude images, improve shape estimation over amplitude-only input.
  • Because the learned renderer is differentiable and runs in about 58 ms per image, the latent parameters can be refined at test time by optimizing only the image loss, enabling unsupervised per-instance adaptation.
  • Real radar images carry enough identity information to separate same-identity captures from different identities across varying pose and expression.
  • Reliable expression recognition from real radar data is not yet achieved; the paper finds it only works with identity-specific trained models.

Reading between the lines

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

  • If the simulator-transfer assumption is the bottleneck, then improving the skin reflectance model and closing the pattern/scale gap between simulated and real radar images should directly translate into better real-data reconstructions, since every network is trained on simulated images.
  • The unsupervised test-time finetuning suggests a practical domain-adaptation route: a model pre-trained on synthetic images could be adapted with unlabeled real radar captures of a new environment or subject, without photogrammetry ground truth.
  • The same encoder-plus-learned-renderer architecture should transfer to other static radar imaging targets, such as hands or body parts, whenever a parametric shape model is available; the paper itself notes the method generalizes to static settings.
  • The identity-separation result on real data is restricted to four male European individuals, so it remains an open question whether the approach holds across sex, skin tone, age, facial hair, and headwear.
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

5 major / 7 minor

Summary. The paper proposes a model-based method for 3D face reconstruction from radar images. It generates 10,000 synthetic radar images with a physics-based ray-tracing renderer from BFM 2019 face meshes, trains a ResNet-50/AlexNet encoder to regress 3DMM shape, expression, and pose parameters, and trains a reversed ResNet-50 as a learned differentiable renderer. The encoder and renderer are combined into a model-based autoencoder trained with both an image loss and a parameter loss, and at test time the latent parameters are further optimized by the image loss. The method is evaluated on synthetic validation images and on real radar images of four male European subjects with photogrammetry ground truth. The paper claims a mean Euclidean point distance of 2.56 mm on synthetic data, improved reconstruction from the autoencoder over the supervised encoder, the ability to perform face recognition on real radar data, and a core benefit of unsupervised training.

Significance. If validated, the paper would introduce a genuinely new sensing modality for 3D face reconstruction and a practical way to obtain a differentiable radar renderer, with potential applications in sleep monitoring and privacy-sensitive settings. The strengths are the synthetic dataset built on a physics-based renderer, the sensible baselines in Table 1, the claimed public release of the dataset, and the explicit comparison of amplitude, depth, and combined inputs. However, the central real-data claims rest on a very small and homogeneous test set, the comparison between encoder and autoencoder conflates two independent changes, and the unsupervised-training claim is not demonstrated. The significance is therefore conditional on additional ablations and on substantially tempered real-data conclusions.

major comments (5)
  1. [Section 3.4/4, Table 1] The comparison between the 'Encoder' and 'Autoencoder' rows in Table 1 changes two factors at once: the autoencoder is trained with the additional image reconstruction term L_image in Eq. (1), and during evaluation its latent parameters are further optimized with the image loss (Section 3.4 states: 'During evaluation both the encoder and the decoder are fixed and the latent space variables are further optimized by the image loss'). No experiment separates these effects, so the reported improvements (e.g., 2.77 mm to 2.56 mm for depth input) cannot be attributed to test-time finetuning or to the joint-training regularization. Please add ablations that report encoder predictions before and after test-time finetuning and autoencoder training without test-time finetuning.
  2. [Section 4, Figures 6a-c and Conclusion] The claim that 'we can perform face recognition' on real data is not supported by the reported evidence. The 4x4 cosine-similarity matrices are presented without a recognition metric, error bars, statistical tests, or a chance baseline, and the diagonal is not consistently dominant: in Figure 6b the off-diagonal similarity between identities 0 and 2 (0.574) exceeds the diagonal value for identity 0 (0.372). Since the real set contains only four male European subjects, the evidence cannot support a general recognition claim. Please report identity-retrieval accuracy with confidence intervals, compare to chance, and provide per-subject mesh errors against the available photogrammetry ground truth.
  3. [Section 3.1/4] The paper collects photogrammetry ground truth for four real faces and generates real-to-synth images from those meshes, but it never reports point distances or parameter errors on these real-to-synth images. This is a missed opportunity to quantify simulator transfer without the ambiguity of real-image noise. Please add these numbers (and compare them with the real-image results) to support the transfer claim.
  4. [Conclusion] The statement that 'the core benefit of our method is that it can be trained in an unsupervised fashion as model-based autoencoder' is not supported. In Section 3.4 the autoencoder is trained with L_params in Eq. (1), which requires ground-truth parameters from the synthetic dataset; only the evaluation-time optimization is unsupervised. No experiment demonstrates training on unlabeled radar images. Please either remove or substantially qualify this claim, or provide an experiment with encoder training that does not use parameter labels.
  5. [Section 3.2/3.4 and Limitations] The learned decoder is trained on images from the same physics-based simulator, and the paper's own Limitations section acknowledges a synthetic-to-real domain gap in patterns and scale and states that the renderer only approximates skin reflectance. Since the test-time optimization minimizes image loss through this learned decoder, the optimization may drive latent codes toward renderer-specific artifacts rather than correct geometry; no ablation or real-data error metric rules this out. Please quantify the effect of the domain gap, for example by reporting real-image point distances to photogrammetry meshes and comparing optimization through the learned decoder against direct encoder outputs.
minor comments (7)
  1. [Section 3.2 and Contributions] The mesh mask 'face12' should be defined, and the claimed publicly available dataset should include a URL or repository information; the contribution list promises release but no link is given.
  2. [Figures 2 and 5] The captions use the terms 'synthetic-real', 'normal', and 'additionally_depth' inconsistently; please unify the terminology and define all abbreviations in the text.
  3. [Section 3.3] The sentence 'the L2 loss between those parameters and the resulting parameters' appears to contain a typo; 'resulting' should probably be 'ground truth'.
  4. [Section 3.4 and Eq. (1)] Please specify the dimensionality and image representation used in L_image when depth or amplitude-depth inputs are used, and state how the image loss is weighted across channels.
  5. [Table 1] No standard deviations or confidence intervals are reported for the validation results; given that the validation set contains 1,500 images, this limits reproducibility and comparability.
  6. [Section 3.4] Details of the test-time optimization (number of iterations, learning rate, early stopping, initialization) are missing; without them the unsupervised finetuning cannot be reproduced.
  7. [References] References [8], [9], and [17] contain incomplete or malformed bibliographic entries (for example, 'arXiv: Computer Vision and Pattern Recognition' and a S2ID field); please fix them.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: synthetic evaluation is anchored to external ground-truth parameters, real evaluation uses photogrammetry, and the claimed finetuning benefit is confounded but not by-construction.

full rationale

Synthetic evaluation is anchored to external ground truth: Table 1 compares predicted BFM parameters against the known Gaussian-sampled shape, expression, and pose parameters used to generate each synthetic image (Sections 3.2 and 4), and real-data evaluation relies on photogrammetric meshes captured with five RGB cameras (Section 3.1, Figure 1), not on the paper's own outputs. The learned renderer is trained on the same physics-based simulator that generated the training set, but the reported metrics (L2 parameter error and mesh point distance against ground truth) do not reduce to the renderer's outputs; optimizing the image loss through a learned decoder could in principle move latent parameters away from the ground truth, so the improvement in Table 1 is an empirical result, not an identity by construction. The radar simulator itself is inherited from self-cited prior work (Schuessler et al. [27] and Braeunig et al. [7]), and its fidelity is load-bearing for synthetic-to-real transfer; however, those citations are to peer-reviewed physics-based simulators, the paper includes an independent real-versus-synthetic image comparison (Figure 2), and the Limitations section explicitly acknowledges the domain gap ('different patterns between the radar simulator and the real radar images and different scales') and the approximate skin reflectance. These are validation risks, not definitional loops. The Conclusion's own admission that 'for real data there is no discernible correlation between the ground truth and the resulting meshes concerning shape and expression in most of the results' further weakens the real-data claim without making it circular. The main experimental weakness is that the autoencoder row differs from the encoder row by both joint image-loss training and test-time finetuning, so the finetuning benefit is not isolated; that is a confound, not a circular reduction. No equation or parameter in the paper is defined in terms of the quantity it is claimed to predict.

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

The central claim rests on the BFM face model, the physics-based radar simulator, and the learned renderer's fidelity. The free parameters are mostly hyperparameters and truncation choices, not calibrated physical constants. The main unverified premise is synthetic-to-real transfer: the real test set is four subjects and the paper admits a domain gap. No new physical entities are introduced.

free parameters (4)
  • loss weight λ = 1
    Eq. 1 combines image and parameter losses; set to 1 with the claim that other values did not change results substantially, but no ablation is shown.
  • tanh scaling factor for 3DMM coefficients = 3
    Section 3.3: outputs scaled by 3 to cover 99.8% of sampled values; hand-chosen, not optimized.
  • shape/expression coefficient truncation = α∈R^10, γ∈R^7
    Section 3.3: predicting first 10 shape and 7 expression coefficients covers ~85% shape, ~76% expression variance; this truncation discards the remaining variance and limits reconstruction fidelity.
  • decoder dynamic range = -15 dB
    Section 3.1 and 3.4: amplitude images for the decoder use a fixed dynamic range of -15 dB while encoder training samples -15 to -30 dB; the choice affects the image content the learned renderer sees.
assumptions (5)
  • domain assumption BFM 2019 is a sufficient face prior: first 10 shape and 7 expression coefficients capture enough identity/expression variance (about 85% and 76%) for reconstruction.
    Section 3.3 truncates α∈R^10, γ∈R^7; any information in higher-order coefficients is discarded, which limits the best possible reconstruction.
  • domain assumption The radar ray-tracing simulator [27,7] accurately models the real 72-82 GHz MIMO radar system for faces at 25 cm.
    Section 3.2: the synthetic dataset that trains the encoder and the learned renderer is generated by this simulator; if its scattering model for skin is wrong, both synthetic results and transfer to real data are invalid. Paper notes the physics-based renderer only approximates the reflectance properties of skin.
  • ad hoc to paper The learned decoder (reversed ResNet-50) is an accurate differentiable proxy for the radar renderer for use in test-time optimization.
    Section 3.4: the decoder is trained on the synthetic dataset and then fixed; test-time finetuning optimizes the image loss through this decoder. No quantitative evaluation of the decoder's fidelity on real radar images is provided.
  • domain assumption Real-captured faces are adequately represented by the Gaussian sampling distribution used for synthetic training (N(0,1) coefficients, poses within ±5° yaw, ±10° pitch/roll, ±5 cm translation).
    Section 3.2: if real faces or poses in the sleep-lab setting fall outside this distribution, encoder and finetuning results degrade; the real test set of 4 subjects cannot validate this distribution.
  • domain assumption Photogrammetry from five optical cameras provides accurate 3D ground truth for the real-subject meshes.
    Section 3.1: real evaluation compares against photogrammetrically reconstructed meshes; this ground truth is assumed accurate and is not independently verified in the paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of 3D Face Reconstruction From Radar Images." pith.science (2026). https://pith.science/paper/ZCHFTMEY

@misc{pith2026241202403,
  author       = {Pith},
  title        = {Pith review of: 3D Face Reconstruction From Radar Images},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZCHFTMEY}},
  note         = {Machine review of arXiv:2412.02403}
}
read the original abstract

The 3D reconstruction of faces gains wide attention in computer vision and is used in many fields of application, for example, animation, virtual reality, and even forensics. This work is motivated by monitoring patients in sleep laboratories. Due to their unique characteristics, sensors from the radar domain have advantages compared to optical sensors, namely penetration of electrically non-conductive materials and independence of light. These advantages of radar signals unlock new applications and require adaptation of 3D reconstruction frameworks. We propose a novel model-based method for 3D reconstruction from radar images. We generate a dataset of synthetic radar images with a physics-based but non-differentiable radar renderer. This dataset is used to train a CNN-based encoder to estimate the parameters of a 3D morphable face model. Whilst the encoder alone already leads to strong reconstructions of synthetic data, we extend our reconstruction in an Analysis-by-Synthesis fashion to a model-based autoencoder. This is enabled by learning the rendering process in the decoder, which acts as an object-specific differentiable radar renderer. Subsequently, the combination of both network parts is trained to minimize both, the loss of the parameters and the loss of the resulting reconstructed radar image. This leads to the additional benefit, that at test time the parameters can be further optimized by finetuning the autoencoder unsupervised on the image loss. We evaluated our framework on generated synthetic face images as well as on real radar images with 3D ground truth of four individuals.

Figures

Figures reproduced from arXiv: 2412.02403 by the authors.

Figure 1
Figure 1. The real radar setup and RGB cameras for photogram [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Examples for a real radar images (left) and synthetic-real [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Examples of a synthetic amplitude image (left) with a [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (14 more)
Figure 4
Figure 4. Figure 4: Overview of our method. The input image is fed to three encoder networks which predict the shape, expression, and pose of the [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 7
Figure 7. Figure 7: Reconstructions of the autoencoder models with the different settings. The first row contains the results for the synthetic radar [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 8
Figure 8. Figure 8: The resulting meshes generated with parameters of the different models. The left half depicts the results from the models [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 6
Figure 6. Figure 6: Cosine similarity comparison between the shape and [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 5
Figure 5. Figure 5: Cosine similarity comparison between the face model [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 10
Figure 10. Figure 10: Comparison of the cosine similarity of the shape pa [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]
Figure 11
Figure 11. Figure 11: Comparison of the cosine similarity of the shape pa [PITH_FULL_IMAGE:figures/full_fig_p011_11.png]
Figure 12
Figure 12. Figure 12: Comparison of the cosine similarity of the shape pa [PITH_FULL_IMAGE:figures/full_fig_p012_12.png]
Figure 15
Figure 15. Figure 15: Comparison of the cosine similarity of the shape pa [PITH_FULL_IMAGE:figures/full_fig_p012_15.png]
Figure 16
Figure 16. Figure 16: Cosine similarity comparison between the shape and [PITH_FULL_IMAGE:figures/full_fig_p013_16.png]
Figure 18
Figure 18. Figure 18: Resulting amplitude images of the autoencoder model trained on amplitude, depth, and amplitude-depth images for different synthetic face instances. 4 [PITH_FULL_IMAGE:figures/full_fig_p014_18.png]
Figure 19
Figure 19. Figure 19: Resulting amplitude images of the autoencoder model trained on amplitude, depth, and amplitude-depth images for different real face instances. 5 [PITH_FULL_IMAGE:figures/full_fig_p015_19.png]
Figure 20
Figure 20. Figure 20: Mesh reconstructions of the results from the [PITH_FULL_IMAGE:figures/full_fig_p016_20.png]
Figure 21
Figure 21. Figure 21: Mesh reconstructions of the results from the [PITH_FULL_IMAGE:figures/full_fig_p017_21.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 33 canonical work pages

  1. [1]

    rohde - schwarz

    Rohde & Schwarz IMAGER, [online] available: https : / / www . rohde - schwarz . com / de / produkte / messtechnik / microwave - imaging / imager _ 256948 . html/ , Nov. 2024. 3

    work page 2024

  2. [2]

    tripsavvy

    Which airports have full body scanners?, [online] avail- able: https : / / www . tripsavvy . com / which - airports - have - full - body - scanners - 3150257, Oct. 2024. 1

    work page 2024

  3. [3]

    rohde - schwarz

    Rohde & Schwarz QAR50 automotive radome tester, [on- line] available: https : / / www . rohde - schwarz . com/qar50/, Oct. 2024. 3

    work page 2024

  4. [4]

    Sherif S. Ahmed. Microwave imaging in security — two decades of innovation. IEEE Journal of Microwaves , 1(1): 191–201, 2021. 1, 3

    work page 2021

  5. [5]

    Advanced Microwave Imaging

    Sherif Sayed Ahmed, Andreas Schiessl, Frank Gumb- mann, Marc Tiebout, Sebastian Methfessel, and Lorenz- Peter Schmidt. Advanced Microwave Imaging. IEEE Mi- crowave Magazine, 13(6):26–43, 2012. 3, 4

    work page 2012

  6. [6]

    Optimal landmark detec- tion using shape models and branch and bound

    Brian Amberg and Thomas Vetter. Optimal landmark detec- tion using shape models and branch and bound. In 2011 In- ternational Conference on Computer Vision, pages 455–462,

    work page 2011

  7. [7]

    A realistic radar ray tracing simulator for hand pose imaging

    Johanna Br ¨aunig, Christian Sch ¨ußler, Vanessa Wirth, Marc Stamminger, Ingrid Ullmann, and Martin V ossiek. A realistic radar ray tracing simulator for hand pose imaging. In 2023 20th European Radar Conference (EuRAD), pages 238–341,

    work page 2023

  8. [8]

    FacePoseNet: Making a Case for Landmark-Free Face Alignment

    Fengju Chang, Anh Tuan Tran, Tal Hassner, Iacopo Masi, Ram Nevatia, and Gerard Medioni. FacePoseNet: Making a Case for Landmark-Free Face Alignment. arXiv: Computer Vision and Pattern Recognition , 2017. arXiv:1708.07517 [cs]. 2, 4

    work page Pith review arXiv 2017

Show all 37 references
  1. [9]

    ExpNet: Landmark- Free, Deep, 3D Facial Expressions

    Feng-Ju Chang, Anh Tuan Tran, Tal Hassner, Iacopo Masi, Ram Nevatia, and Gerard Medioni. ExpNet: Landmark- Free, Deep, 3D Facial Expressions. In2018 13th IEEE Inter- national Conference on Automatic Face & Gesture Recogni- tion (FG 2018), pages 122–129, 2018. 2, 4

    2018

  2. [10]

    ImmFusion: Robust mmWave-RGB Fusion for 3D Human Body Reconstruction in All Weather Conditions

    Anjun Chen, Xiangyu Wang, Kun Shi, Shaohao Zhu, Bin Fang, Yingfeng Chen, Jiming Chen, Yuchi Huo, and Qi Ye. ImmFusion: Robust mmWave-RGB Fusion for 3D Human Body Reconstruction in All Weather Conditions. In 2023 IEEE International Conference on Robotics and Automation (ICRA), ...

    2023

  3. [11]

    Vetter, B

    Chunlu Li, Andreas Morel-Forster, T. Vetter, B. Egger, and Adam Kortylewski. Robust Model-based Face Recon- struction through Weakly-Supervised Outlier Segmentation. arXiv: Computer Vision and Pattern Recognition, 2021. 2

    2021

  4. [12]

    Accurate 3D Face Reconstruction with Weakly-Supervised Learning: From Single Image to Image Set

    Yu Deng, Jiaolong Yang, Sicheng Xu, Dong Chen, Yunde Jia, and Xin Tong. Accurate 3D Face Reconstruction with Weakly-Supervised Learning: From Single Image to Image Set. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR) Workshops ,

  5. [13]

    3d morphable face models—past, present, and future

    Bernhard Egger, William AP Smith, Ayush Tewari, Stefanie Wuhrer, Michael Zollhoefer, Thabo Beeler, Florian Bernard, Timo Bolkart, Adam Kortylewski, Sami Romdhani, et al. 3d morphable face models—past, present, and future. ACM Transactions on Graphics (ToG), 39(5):1–38, 2020. 2

    2020

  6. [14]

    Morphable Face Models - An Open Frame- work

    Thomas Gerig, Andreas Morel-Forster, Clemens Blumer, Bernhard Egger, Marcel L ¨uthi, Sandro Schoenborn, and Thomas Vetter. Morphable Face Models - An Open Frame- work. 2018 13th IEEE International Conference on Auto- matic Face & Gesture Recognition (FG 2018), pages 75–82,

    2018

  7. [15]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceed- ings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016. 2, 4

    2016

  8. [16]

    Mask r-cnn

    Kaiming He, Georgia Gkioxari, Piotr Doll ´ar, and Ross Gir- shick. Mask r-cnn. In Proceedings of the IEEE international conference on computer vision, pages 2961–2969, 2017. 2

    2017

  9. [17]

    Br ¨aunig, V

    J. Br ¨aunig, V . Wirth, Christoph Kammel, Chris- tian Sch ¨ußler, I. Ullmann, M. Stamminger, and M. V ossiek. An Ultra-Efficient Approach for High- Resolution MIMO Radar Imaging of Human Hand Poses. IEEE Transactions on Radar Systems , 2023. S2ID: aecd085eab99912652a22a413d10...

    2023

  10. [18]

    MR angiography with two- dimensional acquisition and three-dimensional display

    P J Keller, B P Drayer, E K Fram, K D Williams, C L Dumoulin, and S P Souza. MR angiography with two- dimensional acquisition and three-dimensional display. work in progress. Radiology, 173(2):527–532, 1989. 3

    1989

  11. [19]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv: Computer Science - Machine Learning, 2017. 5

    2017

  12. [20]

    Imagenet classification with deep convolutional neural net- works

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural net- works. In Advances in Neural Information Processing Sys- tems. Curran Associates, Inc., 2012. 2, 4

    2012

  13. [21]

    Black, Hao Li, and Javier Romero

    Tianye Li, Timo Bolkart, Michael J. Black, Hao Li, and Javier Romero. Learning a model of facial shape and ex- pression from 4D scans. ACM Transactions on Graphics, 36 (6):194, 2017. 2

    2017

  14. [22]

    A convnet for the 2020s

    Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feicht- enhofer, Trevor Darrell, and Saining Xie. A convnet for the 2020s. In Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition , pages 11976–11986,

  15. [23]

    Matthew Loper, Naureen Mahmood, Javier Romero, Gerard Pons-Moll, and Michael J. Black. SMPL: a skinned multi- person linear model. ACM Transactions on Graphics, 34(6): 1–16, 2015. 2

    2015

  16. [24]

    Paysan, R

    P. Paysan, R. Knothe, B. Amberg, S. Romdhani, and T. Vet- ter. A 3d face model for pose and illumination invariant face recognition. Proceedings of the 6th IEEE International Con- ference on Advanced Video and Signal based Surveillance (AVSS) for Security, Safety and Monitorin...

    2009

  17. [25]

    Soubhik Sanyal, Timo Bolkart, Haiwen Feng, and Michael J. Black. Learning to Regress 3D Face Shape and Expression From an Image Without 3D Supervision. In2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 7755–7764, Long Beach, CA, USA, 2019. IEEE. 2

    2019

  18. [26]

    Radar-based recognition of static hand gestures in american sign language

    Christian Schuessler, Wenxuan Zhang, Johanna Br ¨aunig, Marcel Hoffmann, Michael Stelzig, and Martin V ossiek. Radar-based recognition of static hand gestures in american sign language. In 2024 IEEE Radar Conference (Radar- Conf24), pages 1–6, 2024. 3

    2024

  19. [27]

    A Realis- tic Radar Ray Tracing Simulator for Large MIMO-Arrays in Automotive Environments

    Christian Sch ¨ußler, Marcel Hoffmann, Johanna Br¨aunig, In- grid Ullmann, Randolf Ebelt, and Martin V ossiek. A Realis- tic Radar Ray Tracing Simulator for Large MIMO-Arrays in Automotive Environments. IEEE Journal of Microwaves, 1 (4):962–974, 2021. Conference Name: IEEE Jou...

    2021

  20. [28]

    Unre- stricted Facial Geometry Reconstruction Using Image-to- Image Translation, 2017

    Matan Sela, Elad Richardson, and Ron Kimmel. Unre- stricted Facial Geometry Reconstruction Using Image-to- Image Translation, 2017. arXiv:1703.10131 [cs]. 2

    2017 arXiv

  21. [29]

    MoFA: Model-based Deep Convolutional Face Autoencoder for Unsupervised Monocular Reconstruction

    Ayush Tewari, Michael Zollh ¨ofer, Hyeongwoo Kim, Pablo Garrido, Florian Bernard, Florian Bernard, Patrick P ´erez, Patrick P´erez, and Christian Theobalt. MoFA: Model-based Deep Convolutional Face Autoencoder for Unsupervised Monocular Reconstruction. arXiv: Computer Vision a...

    2017

  22. [30]

    Regressing Robust and Discriminative 3D Mor- phable Models with a Very Deep Neural Network

    Anh Tuan Tran, Tal Hassner, Iacopo Masi, and Gerard Medioni. Regressing Robust and Discriminative 3D Mor- phable Models with a Very Deep Neural Network. In 2017 IEEE Conference on Computer Vision and Pattern Recogni- tion (CVPR), pages 1493–1502, Honolulu, HI, 2017. IEEE. 2

    2017

  23. [31]

    Maroon: A framework for the joint characterization of near-field high-resolution radar and optical depth imaging techniques

    Vanessa Wirth, Johanna Br ¨aunig, Martin V ossiek, Tim Weyrich, and Marc Stamminger. Maroon: A framework for the joint characterization of near-field high-resolution radar and optical depth imaging techniques. arxiv: Electrical En- gineering and Systems Science - Image and Vid...

    2024

  24. [32]

    mm3dface: Nonintru- sive 3d facial reconstruction leveraging mmwave signals

    Jiahong Xie, Hao Kong, Jiadi Yu, Yingying Chen, Linghe Kong, Yanmin Zhu, and Feilong Tang. mm3dface: Nonintru- sive 3d facial reconstruction leveraging mmwave signals. In Proceedings of the 21st Annual International Conference on Mobile Systems, Applications and Services , pag...

    2023

  25. [33]

    BodyCompass: Monitoring Sleep Posture with Wireless Signals

    Shichao Yue, Yuzhe Yang, Hao Wang, Hariharan Rahul, and Dina Katabi. BodyCompass: Monitoring Sleep Posture with Wireless Signals. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, 4(2):1–25,

  26. [34]

    mmFER: Millimetre-wave Radar based Facial Expression Recognition for Multimedia IoT Applications

    Xi Zhang, Yu Zhang, Zhenguo Shi, and Tao Gu. mmFER: Millimetre-wave Radar based Facial Expression Recognition for Multimedia IoT Applications. In Proceedings of the 29th Annual International Conference on Mobile Computing and Networking, pages 1–15, Madrid Spain, 2023. ACM. 1, 2

    2023

  27. [35]

    Through-Wall Human Pose Estimation Using Ra- dio Signals

    Mingmin Zhao, Tianhong Li, Mohammad Abu Alsheikh, Yonglong Tian, Hang Zhao, Antonio Torralba, and Dina Katabi. Through-Wall Human Pose Estimation Using Ra- dio Signals. In 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7356–7365, Salt Lake City, UT,...

    2018

  28. [36]

    Through-Wall Human Mesh Recovery Using Radio Signals

    Mingmin Zhao, Yingcheng Liu, Aniruddh Raghu, Hang Zhao, Tianhong Li, Antonio Torralba, and Dina Katabi. Through-Wall Human Mesh Recovery Using Radio Signals. In 2019 IEEE/CVF International Conference on Computer Vision (ICCV), pages 10112–10121, Seoul, Korea (South),

    2019

  29. [2019]

    1, 2 10 3D Face Reconstruction From Radar Images Supplementary Material A

    IEEE. 1, 2 10 3D Face Reconstruction From Radar Images Supplementary Material A. Additional Results Additional Quantitative Results. In addition to the cosine similarity plots in Section 4 of the paper, we provide additional comparisons of the ground truth parameters, the enco...

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.