Pith. sign in

REVIEW 1 major objections 5 minor 58 references

Learning to See Inside Opaque Liquid Containers using Speckle Vibrometry

T0 review · 1 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read This paper claims that a camera capturing laser speckle vibrations on a 2D grid of surface points, combined with a transformer, can remotely infer the fill level of opaque liquid containers to about 1% error within distribution, without…

desk verdict A genuinely new proof-of-concept for remote fill-level sensing via speckle vibrometry, with a real overclaim about source invariance that should be fixed before publication. read the letter →

arxiv 2507.20757 v1 pith:RNLQTEDK submitted 2025-07-28 cs.CV cs.AI

classification cs.CVcs.AI
keywords specklevibrometryliquidlevelestimationopaquecontainersnon-contactsensingvibrationtransformerlasercomputervisionobjectinspection
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to give computer vision the ability to see inside opaque liquid containers. Its central claim is that a container's tiny surface vibrations, captured remotely by imaging laser speckle at many points at once, encode its hidden liquid level accurately enough to be learned by a transformer. The authors build a first-of-a-kind 2D-grid speckle vibrometry system, record a dataset of everyday containers, and train a Vibration Transformer that reads fill level from Fourier magnitudes of the vibrations. They report errors as low as 1% for containers seen during training, 9% on unseen instances of the same class, and 4% under ambient sound. If correct, the method opens a non-contact way to inspect sealed containers in warehouses, factories, and settings where weighing or touching is impractical or unsafe.

What carries the argument

The mechanism is speckle-based vibrometry on a 2D grid: one laser is split into a 6x6 grid of points, projected onto the containers, and a defocused camera records each point as a patch of random interference (speckle) that shifts with surface tilt. The authors track these shifts with phase correlation followed by Lucas-Kanade (PCLK+) to get two-axis vibration signals per point at rates up to 57 kHz. The Vibration Transformer then takes the Fourier magnitudes of these signals over 100-2500 Hz, tokenizes frequency bands, processes each point with a PointTransformer, and fuses the points with a ShapeTransformer, trained with a SORD loss that respects the ordinal nature of fill levels.

What would settle it

Play a single pure tone near a container's resonant frequency as the only excitation, and evaluate the trained model on an unseen instance of that container: if the source-invariance claim is correct, the MAE should stay close to the 0.04 seen with ambient sound, while a large MAE increase would show that narrowband or notched excitations break the Fourier-magnitude proxy.

Watch

Extended reading notes

Core claim

The paper's central discovery is that a 2D grid of remotely sensed surface vibrations, converted to per-point Fourier magnitudes, is enough to regress an opaque container's liquid level across many everyday container types. On the sensing side, the authors build a laser-grid speckle system that captures vibrations of several containers simultaneously at up to 57 kHz, and on the inference side a 'Vibration Transformer' whose shared PointTransformer encodes each point's spectrum and whose ShapeTransformer fuses the points to produce container class and fill level. They report a 0.01 MAE (1%) for within-distribution fill levels, 0.09 for unseen instances of a known class, and 0.04 under ambient sound, and they show the model can interpolate to fill levels (25%, 50%, 75%) it never trained on.

Load-bearing premise

The load-bearing premise is that the Fourier magnitude of the recorded vibrations is a faithful proxy for the container's transfer function, which holds only when the excitation sound is sufficiently broadband and which the paper tests only with a chirp, a song segment, and one ambient noise type.

Editorial extensions

If this is right

  • Warehouse and factory inspection of sealed beverage or chemical containers could be done remotely and at once, with no physical handling or weighing.
  • Because the model discards phase and uses only Fourier magnitudes, the same trained network generalizes to different excitation sounds, including unseen ambient noise, as long as the sound is broadband.
  • The ordinal loss lets the model interpolate to fill levels never seen in training (e.g., 25%, 50%, 75%), so the method is not limited to the six discrete training levels.
  • Trained on several instances of a container class, the model can predict the fill level of a new instance of the same class, such as the sixth can of a six-pack.

Reading between the lines

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

  • The source-invariance argument suggests a practical recipe: in industrial settings one could deliberately play a short broadband probe (a chirp or noise burst) and then rely on ambient sound afterward; the paper does not test this mixed-excitation scenario.
  • Since multi-point data clearly helps on unseen instances, extending the 3 points per container to a denser grid (the hardware already captures a 6x6 grid) could be expected to further improve generalization beyond the paper's reported results.
  • The same Fourier-magnitude representation would presumably apply to other hidden contents whose presence changes acoustics, such as granular materials, powders, or spoilage gases in sealed food; the authors name these as open questions, not demonstrated claims.
  • A narrowband ambient environment (e.g., a room with a strong 50/60 Hz hum or a tonal alarm) could violate the broadband assumption; a robust system might need to estimate and whiten the excitation spectrum, which the current model does not do.
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

1 major / 5 minor

Summary. The paper proposes a speckle-based vibrometry system that captures vibrations on a 2D grid of laser points using a single defocused camera, enabling simultaneous multi-point measurement of multiple containers. The authors introduce a transformer-based architecture (the Vibration Transformer) that takes the Fourier magnitudes of vibration signals at three surface points per container and classifies container type and discrete liquid fill level. They collect a dataset of everyday containers and report MAE values of 0.01 (within-distribution), 0.09 (unseen instances), and 0.04 (ambient sound), with ablations showing that a CNN baseline fails at level prediction and that multi-point measurements help in harder generalization settings. The paper claims invariance to the vibration source, qualified in a footnote as holding for 'sufficiently broadband' excitation.

Significance. If the results hold, this is a novel non-contact sensing modality for inspecting sealed containers, with potential applications in warehouses, industrial monitoring, and hazardous-liquid storage. The paper's strengths include a genuine hardware contribution (2D-grid speckle vibrometry with ROI-based high-speed readout), a newly collected dataset, a physics-inspired architecture with sensible ablations, and a self-critical limitations section. The experimental support is appropriate for a proof of concept. The main weakness is that the source-invariance claim is broader than the evidence supports, since only a limited set of broadband excitations were tested and the conditioning assumption 'sufficiently broadband' is not quantitatively defined.

major comments (1)
  1. [Abstract; §1, §4, §6(d–f)] The claim that the architecture is 'invariant to the vibration source' is load-bearing and currently rests on an untested conditioning assumption. As Eq. (1) shows, the model input is |F{v_i}|, which for a linear structure equals |H_i(f)||X(f)|; separating the container transfer function from the excitation spectrum is impossible without a reference measurement of X(f) when X(f) has deep notches or is narrowband. Footnote 1 limits the claim to 'sufficiently broadband' excitation, but this condition is never defined quantitatively and no experiment uses a source with strong spectral notches or a narrowband hum (e.g., 50/120 Hz harmonics), which are common in real environments. The three tested excitations (chirp, song, supermarket noise) and the random smooth-filter augmentation all have broad, positive spectra. Please either (i) add experiments with held-out broadband sources (e.g., white noise, a different song) and with narrowband or notched sources, reporting MAE for each, and discuss what 'sufficiently broadband' means quantitatively, or (ii) remove the unconditional invariance wording from the abstract and Introduction and state the limitation explicitly in Sec. 7.
minor comments (5)
  1. [§5.2] The frequency set F_fixed = {100, 100.5, ..., 2500 Hz} contains 4801 values, but the text says the resulting V_i is a 2×4800 matrix; please correct the count or the frequency set.
  2. [§6, test (a)] Excluding one of six speakers yields about 16.7% of the data for testing, not 'about 20%' as stated; please adjust the text.
  3. [§3] The phrase 'Our GPU implementation is ×20 faster' lacks a baseline; please specify the comparison (e.g., a CPU implementation, or a previous version of PCLK) and the hardware.
  4. [Abstract; §1] The abstract states 'invariant to the vibration source' without the 'sufficiently broadband' qualification from Footnote 1; adding the qualifier would align the abstract with the technical content of the paper.
  5. [Table 1] The 'chance≈ 30%' figure for level MAE is not defined; a brief explanation (e.g., expected MAE of a uniform random guess on [0,1]) would improve clarity.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the fill-level prediction is empirically validated on genuinely held-out data with scale-based ground truth; self-citations are background and not load-bearing.

full rationale

This paper's central claim — that a transformer can classify container type and fill level from multi-point speckle-vibration spectra — is supported by a genuinely held-out experimental protocol rather than by construction. The model is trained on measured vibration signals with fill-level labels obtained from external kitchen scales, and is tested on novel speaker positions (test a), unseen container instances (test b), unseen intermediate fill levels (tests c and e), and a training-unseen excitation type, supermarket ambient noise (tests d-f). No equation in Sec. 4 defines the output in terms of a fitted parameter: Eq. (1) is only the input representation, Eqs. (2)-(5) define the ordinal loss and the two estimators, and none of them constrain the held-out predictions. The network's 0.01-0.16 MAE values are empirical results on excluded data, not re-statements of fit values. The self-citations (Sheinin et al. [46] and Zhang et al. [58], both sharing author Sheinin) are background references for the speckle-vibrometry measurement principle and prior modal analysis; they are externally published, independently validated CVPR works, and the present contribution — 2D-grid sensing and learned fill-level inference — is validated by this paper's own experiments. No uniqueness theorem from the authors' prior work is invoked to force the architecture or the result, so these citations are not load-bearing. Two flagged concerns are scope limitations rather than circularity. First, the abstract's unconditional 'invariant to the vibration source' is qualified in Sec. 4 by footnote 1: 'As long as the excitation signal is sufficiently broadband.' Since for a linear system V_i(f) = H_i(f)X(f), the Eq. (1) input retains |X|, a narrowband or notch-heavy source could mask container resonances; the evaluation covers only broadband excitations (chirp, song, supermarket noise), so the unconditional phrasing is stronger than the evidence. Second, Sec. 7 honestly acknowledges container-material limitations and the small dataset for novel-class generalization. Neither concern reduces a claimed prediction to its inputs, so the circularity score is 1.

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

The central claim depends on standard optical and mechanical physics assumptions (speckle tilt sensitivity, linear elasticity), on a source-invariance assumption stated in a footnote, and on several design choices (SORD temperature, frequency range, point count, level set). No new physical entities are introduced. The paper contributes a hardware system and a learned model, not a new conserved quantity or mediator.

free parameters (4)
  • SORD temperature beta = 50
    Eq. (2) sets the soft target distribution width for the ordinal loss. Chosen by hand with no sensitivity analysis; it controls how the model interpolates to intermediate liquid levels.
  • Frequency set F_fixed = 100-2500 Hz, 0.5 Hz resolution
    Sec. 5.2. Chosen based on the observation that most containers have little energy above 2000 Hz. This determines the input representation and the model's operational bandwidth.
  • Number of surface points per container = 3
    Sec. 4 (WLOG). The ablation (Tab. 2) shows single-point models are worse on unseen instances, so this design choice affects the central generalization claim.
  • Discrete liquid level set L = {0.0, 0.2, 0.4, 0.6, 0.8, 1.0}
    Sec. 4. The model is trained only on these six levels; intermediate levels are predicted via the expectation estimator (Eq. 5). This representation choice shapes the reported accuracy and MAE.
assumptions (5)
  • domain assumption Speckle patterns shift in the image in response to surface tilts, enabling vibration recovery from defocused laser spots.
    Sec. 2.2 and Sec. 3 rely on prior speckle vibrometry results (Zalevsky et al. [57], Sheinin et al. [46]). This is the physical basis for the entire measurement system.
  • domain assumption Linear elasticity and modal analysis: object vibrations decompose into resonant modes, and material properties can be inferred from modes.
    Sec. 2.1 and Sec. 4 motivate the Fourier-domain input and the PointTransformer/ShapeTransformer design by analogy to mode frequencies and mode shapes.
  • domain assumption The Fourier magnitude of the vibration response is approximately independent of the excitation content when the excitation is sufficiently broadband.
    Footnote 1 states this condition is needed for source invariance. It is load-bearing for tests (d)-(f) and the 'invariant to the vibration source' claim in the abstract.
  • domain assumption Three surface points per container capture enough mode-shape information to generalize to unseen instances of the same class.
    Sec. 4 and the ablation in Tab. 2 assume that multi-point data encodes mode shapes; the performance gain in unseen instances (0.09 vs 0.11 MAE) supports this, but it remains an empirical assumption.
  • domain assumption The ground-truth fill level measured by kitchen scales is accurate.
    Sec. 6: the dataset uses scales to measure added liquid. This is a standard experimental assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning to See Inside Opaque Liquid Containers using Speckle Vibrometry." pith.science (2026). https://pith.science/paper/RNLQTEDK

@misc{pith2026250720757,
  author       = {Pith},
  title        = {Pith review of: Learning to See Inside Opaque Liquid Containers using Speckle Vibrometry},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RNLQTEDK}},
  note         = {Machine review of arXiv:2507.20757}
}
read the original abstract

Computer vision seeks to infer a wide range of information about objects and events. However, vision systems based on conventional imaging are limited to extracting information only from the visible surfaces of scene objects. For instance, a vision system can detect and identify a Coke can in the scene, but it cannot determine whether the can is full or empty. In this paper, we aim to expand the scope of computer vision to include the novel task of inferring the hidden liquid levels of opaque containers by sensing the tiny vibrations on their surfaces. Our method provides a first-of-a-kind way to inspect the fill level of multiple sealed containers remotely, at once, without needing physical manipulation and manual weighing. First, we propose a novel speckle-based vibration sensing system for simultaneously capturing scene vibrations on a 2D grid of points. We use our system to efficiently and remotely capture a dataset of vibration responses for a variety of everyday liquid containers. Then, we develop a transformer-based approach for analyzing the captured vibrations and classifying the container type and its hidden liquid level at the time of measurement. Our architecture is invariant to the vibration source, yielding correct liquid level estimates for controlled and ambient scene sound sources. Moreover, our model generalizes to unseen container instances within known classes (e.g., training on five Coke cans of a six-pack, testing on a sixth) and fluid levels. We demonstrate our method by recovering liquid levels from various everyday containers.

Figures

Figures reproduced from arXiv: 2507.20757 by the authors.

Figure 1
Figure 1. Learning to ‘see’ the fill level of opaque containers. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Capturing vibrations on a 2D grid. (a) Our system con￾sists of a laser projecting a 2D grid of points on the scene, and a single defocused camera. The camera image, shown in (b), cap￾tures the speckle interference for all grid points. (b) We define a set of regions of interest (ROIs) centered on the middle of each de￾focused row of laser points (ROI centers marked in yellow), with each ROI having a height of a few p… view at source ↗
Figure 2
Figure 2. Variations in container responses. (a) Setup showing vi￾bration measurements from a resonant wine glass (clear for illus￾tration) and a standard beverage container. We record responses from two identical items at different fill levels using a single sur￾face point per object. (b) The wine glass exhibits a distinct res￾onance that shifts with fill level. In contrast, the coconut water container exhibits a complex fre… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Vibration Transformer network architecture. Our model processes the signals after conversion to the Fourier domain. The [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Experimental setup. (a) We acquire vibrations for multi￾ple containers, at multiple points per container at once, for various fill levels. The containers are excited by six different speakers po￾sitioned on a beam separated from the container set to avoid trans￾ferring…
Figure 6
Figure 6. Figure 6: We explore the learned latent space by projecting [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Challenging inference examples. (a) Two visually simi￾lar empty containers that exhibit different resonant frequency pro￾files, suggesting subtle manufacturing differences. (b) A different container shows nearly identical frequency responses across fill levels, indicat…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

58 extracted references · 55 canonical work pages

  1. [1]

    Imaging with local speckle intensity correlations: the- ory and practice

    Marina Alterman, Chen Bar, Ioannis Gkioulekas, and Anat Levin. Imaging with local speckle intensity correlations: the- ory and practice. ACM Transactions on Graphics (TOG), 40 (3):1–22, 2021. 3

  2. [2]

    Deep speech 2: End-to-end speech recognition in english and man- darin

    Dario Amodei, Sundaram Ananthanarayanan, Rishita Anub- hai, Jingliang Bai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Qiang Cheng, Guoliang Chen, et al. Deep speech 2: End-to-end speech recognition in english and man- darin. In International Conference on Machine Learning (ICML), 2016. 4

  3. [3]

    Video from stills: Lensless imaging with rolling shutter

    Nick Antipa, Patrick Oare, Emrah Bostan, Ren Ng, and Laura Waller. Video from stills: Lensless imaging with rolling shutter. In 2019 IEEE International Conference on Computational Photography (ICCP) , pages 1–8. IEEE,

  4. [4]

    Piyush Bagad, Makarand Tapaswi, Cees G. M. Snoek, and Andrew Zisserman. The sound of water: Inferring physical properties from pouring liquids, 2025. 2

  5. [5]

    Long-range detection of acoustic vibrations by speckle tracking

    S Bianchi and E Giacomozzi. Long-range detection of acoustic vibrations by speckle tracking. Applied optics, 58 (28):7805–7809, 2019. 2

  6. [6]

    Estimating the material properties of fabric from video

    Katherine L Bouman, Bei Xiao, Peter Battaglia, and William T Freeman. Estimating the material properties of fabric from video. In Proceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR) ,

  7. [7]

    Yates, and Laura Waller

    Mingxuan Cai, Dekel Galor, Amit Pal Singh Kohli, Jacob L. Yates, and Laura Waller. Event2audio: Event-based optical vibration sensing. In Proceedings of the IEEE International Conference on Computational Photography (ICCP), 2025. 2

  8. [8]

    Listen, attend and spell

    William Chan, Navdeep Jaitly, Quoc V Le, and Oriol Vinyals. Listen, attend and spell. arXiv preprint arXiv:1508.01211, 2015. 4

Show all 58 references
  1. [9]

    Modal identification of simple structures with high-speed video us- ing motion magnification

    Justin G Chen, Neal Wadhwa, Young-Jin Cha, Fr ´edo Du- rand, William T Freeman, and Oral Buyukozturk. Modal identification of simple structures with high-speed video us- ing motion magnification. Journal of Sound and Vibration, 345:58–71, 2015. 2, 3

  2. [10]

    Deep thermal imaging: Prox- imate material type recognition in the wild through deep learning of spatial surface temperature patterns

    Youngjun Cho, Nadia Bianchi-Berthouze, Nicolai Mar- quardt, and Simon J Julier. Deep thermal imaging: Prox- imate material type recognition in the wild through deep learning of spatial surface temperature patterns. In Proceed- ings of the 2018 CHI conference on human factors i...

  3. [11]

    Sapphire lpx 532 nm 500 mw laser

    Coherent. Sapphire lpx 532 nm 500 mw laser. https:// www.coherent.com/lasers/cw- solid- state/ sapphire. Accessed: 2025-02-27. 5

  4. [12]

    Thermal spread functions (tsf): Physics-guided material classification

    Aniket Dashpute, Vishwanath Saragadam, Emma Alexan- der, Florian Willomitzer, Aggelos Katsaggelos, Ashok Veer- araghavan, and Oliver Cossairt. Thermal spread functions (tsf): Physics-guided material classification. In Proceedings of the IEEE/CVF Conference on Computer Vision a...

  5. [13]

    Visual vibrometry: Estimating material properties from small mo- tion in video

    Abe Davis, Katherine L Bouman, Justin G Chen, Michael Rubinstein, Fredo Durand, and William T Freeman. Visual vibrometry: Estimating material properties from small mo- tion in video. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2015. 2, 3

  6. [14]

    Mechanical vibrations

    Jacob Pieter Den Hartog. Mechanical vibrations. Courier Corporation, 1985. 2

  7. [15]

    BERT: Pre-training of deep bidirectional trans- formers for language understanding

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep bidirectional trans- formers for language understanding. In Proceedings of the Conference of the North American Chapter of the Associa- tion for Computational Linguistics: Human Langua...

  8. [16]

    Soft labels for ordinal regres- sion

    Raul Diaz and Amit Marathe. Soft labels for ordinal regres- sion. In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), 2019. 4

  9. [17]

    Solving sub-pixel image registration prob- lems using phase correlation and lucas-kanade optical flow method

    Youssef Douini, Jamal Riffi, Mohamed Adnane Mahraz, and Hamid Tairi. Solving sub-pixel image registration prob- lems using phase correlation and lucas-kanade optical flow method. In 2017 Intelligent Systems and Computer Vision (ISCV), pages 1–5, 2017. 4

  10. [18]

    Looking to listen at the cocktail party: a speaker-independent audio-visual model for speech sepa- ration

    Ariel Ephrat, Inbar Mosseri, Oran Lang, Tali Dekel, Kevin Wilson, Avinatan Hassidim, William T Freeman, and Michael Rubinstein. Looking to listen at the cocktail party: a speaker-independent audio-visual model for speech sepa- ration. ACM Transactions on Graphics (TOG). Proc. ...

  11. [19]

    Visual vibration tomography: Esti- mating interior material properties from monocular video

    Berthy T Feng, Alexander C Ogren, Chiara Daraio, and Katherine L Bouman. Visual vibration tomography: Esti- mating interior material properties from monocular video. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022. 2, 3

  12. [20]

    Integrating domain knowl- edge into deep networks for lung ultrasound with applica- tions to COVID-19

    Oz Frank, Nir Schipper, Mordehay Vaturi, Gino Soldati, An- drea Smargiassi, Riccardo Inchingolo, Elena Torri, Tiziano Perrone, Federico Mento, Libertario Demi, Meirav Galun, Yonina C Eldar, and Shai Bagon. Integrating domain knowl- edge into deep networks for lung ultrasound w...

  13. [21]

    Non-intrusive tank-filling sensor based on sound resonance

    Adri ´an Garc ´ıa, V ´ıctor Toral, ´Alvaro M ´arquez, Antonio Garc´ıa, Encarnaci ´on Castillo, Luis Parrilla, and Diego P Morales. Non-intrusive tank-filling sensor based on sound resonance. Electronics, 7(12):378, 2018. 2

  14. [22]

    Cnn archi- tectures for large-scale audio classification

    Shawn Hershey, Sourish Chaudhuri, Daniel PW Ellis, Jort F Gemmeke, Aren Jansen, R Channing Moore, Manoj Plakal, Devin Platt, Rif A Saurous, Bryan Seybold, et al. Cnn archi- tectures for large-scale audio classification. In IEEE interna- tional conference on acoustics, speech a...

  15. [23]

    Diffractive beam splitters

    HOLO-OR. Diffractive beam splitters. https://holo- or.com/products/beam- splitters/ . Accessed: 2025-02-27. 5

  16. [24]

    Batch normalization: Accelerating deep network training by reducing internal co- variate shift

    Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep network training by reducing internal co- variate shift. In International Conference on Machine Learn- ing (ICML). pmlr, 2015. 7

  17. [25]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 ,

  18. [26]

    The phase correlation image alignment method

    Charles D Kuglin. The phase correlation image alignment method. In IEEE International Conference on Cybernetics and Society, 1975. 4

  19. [27]

    Spectral sub- surface scattering for material classification

    Haejoon Lee and Aswin C Sankaranarayanan. Spectral sub- surface scattering for material classification. In Proceedings of the European Conference on Computer Vision (ECCV) ,

  20. [28]

    Multimodal material segmentation

    Yupeng Liang, Ryosuke Wakaki, Shohei Nobuhara, and Ko Nishino. Multimodal material segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR), 2022. 1

  21. [29]

    Local-area damage detection in composite structures using piezoelec- tric transducers

    Peter F Lichtenwalner and Donald A Sofge. Local-area damage detection in composite structures using piezoelec- tric transducers. In Smart Structures and Materials 1998: Industrial and Commercial Applications of Smart Structures Technologies, pages 509–515. SPIE, 1998. 3

  22. [30]

    Discriminative illumination: Per- pixel classification of raw materials based on optimal projec- tions of spectral brdf

    Chao Liu and Jinwei Gu. Discriminative illumination: Per- pixel classification of raw materials based on optimal projec- tions of spectral brdf. IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 36(1), 2013. 1

  23. [31]

    An iterative image reg- istration technique with an application to stereo vision

    Bruce D Lucas and Takeo Kanade. An iterative image reg- istration technique with an application to stereo vision. In IJCAI’81: 7th international joint conference on Artificial in- telligence, pages 674–679, 1981. 4

  24. [32]

    Eosens2.0mcx12-cm machine vision cam- era

    Mikrotron. Eosens2.0mcx12-cm machine vision cam- era. https : / / mikrotron . de / en / high - speed - cameras / mik - camera - detail . php ? id=EoSens2.0MCX12-CM. Accessed: 2025-02-27. 5

  25. [33]

    Shape from heat conduction

    Sriram Narayanan, Mani Ramanagopal, Mark Sheinin, Aswin C Sankaranarayanan, and Srinivasa G Narasimhan. Shape from heat conduction. In Proceedings of the Euro- pean Conference on Computer Vision (ECCV), 2024. 1

  26. [34]

    Ultralite-mk5 sound card

    Mark of the Unicorn (MOTU). Ultralite-mk5 sound card. https : / / motu . com / en - us / products / gen5 / ultralite-mk5/. Accessed: 2025-02-27. 5

  27. [35]

    Ogren, Berthy T

    Alexander C. Ogren, Berthy T. Feng, Jihoon Ahn, Kather- ine L. Bouman, and Chiara Daraio. Visual surface wave elas- tography: Revealing subsurface physical properties via visi- ble surface waves. arXiv preprint arXiv:2507.09207, 2025. 2

  28. [36]

    Video-rate image correlation processor

    JJ Pearson, DC Hines Jr, S Golosman, and CD Kuglin. Video-rate image correlation processor. In Applications of digital image processing, pages 197–205. SPIE, 1977. 4

  29. [37]

    A theory of joint light and heat transport for lambertian scenes

    Mani Ramanagopal, Sriram Narayanan, Aswin C Sankara- narayanan, and Srinivasa G Narasimhan. A theory of joint light and heat transport for lambertian scenes. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024. 1

  30. [38]

    Speaker recognition from raw waveform with sincnet

    Mirco Ravanelli and Yoshua Bengio. Speaker recognition from raw waveform with sincnet. In IEEE spoken language technology workshop (SLT), 2018. 4

  31. [39]

    Laser vi- brometry: pseudo-vibrations

    Steve Rothberg, JR Baker, and Neil A Halliwell. Laser vi- brometry: pseudo-vibrations. 1989. 3

  32. [40]

    Subhankar Roy, Willi Menapace, Sebastiaan Oei, Ben Luijten, Enrico Fini, Cristiano Saltori, Iris Huijben, Nishith Chennakeshava, Federico Mento, Alessandro Sen- telli, Emanuele Peschiera, Riccardo Trevisan, Giovanni Maschietto, Elena Torri, Riccardo Inchingolo, Andrea Smar- gi...

  33. [41]

    Material classification with ther- mal imagery

    Philip Saponaro, Scott Sorensen, Abhishek Kolagunda, and Chandra Kambhamettu. Material classification with ther- mal imagery. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 4649–4656, 2015. 1

  34. [42]

    Krism—krylov subspace-based optical computing of hyper- spectral images

    Vishwanath Saragadam and Aswin C Sankaranarayanan. Krism—krylov subspace-based optical computing of hyper- spectral images. ACM Transactions on Graphics (TOG), 38 (5):1–14, 2019. 1

  35. [43]

    Programmable spectrometry: Per-pixel material classifica- tion using learned spectral filters

    Vishwanath Saragadam and Aswin C Sankaranarayanan. Programmable spectrometry: Per-pixel material classifica- tion using learned spectral filters. In Proceedings of the IEEE International Conference on Computational Photog- raphy (ICCP), 2020

  36. [44]

    Sassi—super-pixelated adaptive spatio-spectral imaging

    Vishwanath Saragadam, Michael DeZeeuw, Richard G Baraniuk, Ashok Veeraraghavan, and Aswin C Sankara- narayanan. Sassi—super-pixelated adaptive spatio-spectral imaging. IEEE Transactions on Pattern Analysis and Ma- chine Intelligence (PAMI), 43(7):2233–2244, 2021. 1

  37. [45]

    Reddy, Matthew O’Toole, and Srinivasa G

    Mark Sheinin, Dinesh N. Reddy, Matthew O’Toole, and Srinivasa G. Narasimhan. Diffraction line imaging. In Euro- pean Conference on Computer Vision (ECCV) , pages 1–16,

  38. [46]

    Narasimhan

    Mark Sheinin, Dorian Chan, Matthew O’Toole, and Srini- vasa G. Narasimhan. Dual-shutter optical vibration sensing. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022. 2, 3, 4, 8

  39. [47]

    Colux: Multi-object 3d micro-motion analysis us- ing speckle imaging

    Brandon M Smith, Pratham Desai, Vishal Agarwal, and Mo- hit Gupta. Colux: Multi-object 3d micro-motion analysis us- ing speckle imaging. ACM Transactions on Graphics (TOG), 36(4):1–12, 2017. 3

  40. [48]

    Pebble v2 speakers pair

    Creative Technology. Pebble v2 speakers pair. https: / / us . creative . com / p / speakers / creative - pebble-v2. Accessed: 2025-02-27. 5

  41. [49]

    Anamorphic prism pairs

    Thorlabs. Anamorphic prism pairs. https://www. thorlabs.com/thorproduct.cfm?partnumber= PS873-A. Accessed: 2025-02-27. 5

  42. [50]

    Metal-dielectric object classification by polarization degree map

    Shoji Tominaga and Tetsuya Yamamoto. Metal-dielectric object classification by polarization degree map. InProceed- ings of the IEEE International Conference on Pattern Recog- nition (ICPR), 2008. 1

  43. [51]

    Wavenet: A gener- ative model for raw audio

    A ¨aron van den Oord, Sander Dieleman, Heiga Zen, Karen Simonyan, Oriol Vinyals, Alex Graves, Nal Kalchbrenner, Andrew Senior, and Koray Kavukcuoglu. Wavenet: A gener- ative model for raw audio. In The 9th ISCA Speech Synthesis Workshop (SSW), 2016. 4

  44. [52]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszko- reit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in Neural Information Processing Systems (NeurIPS), 2017. 4

  45. [53]

    100,000 frames-per-second compressive imaging with a conventional rolling-shutter camera by random point-spread-function engineering

    Gil Weinberg and Ori Katz. 100,000 frames-per-second compressive imaging with a conventional rolling-shutter camera by random point-spread-function engineering. Op- tics Express, 28(21):30616–30625, 2020. 7

  46. [54]

    Justin Wilson, Auston Sterling, and Ming C. Lin. Analyzing liquid pouring sequences via audio-visual neural networks. In 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 7702–7709, 2019. 2

  47. [55]

    Fast motion estimation of one-dimensional laser speckle image and its application on real-time audio signal acquisition

    Nan Wu and Shinichiro Haruyama. Fast motion estimation of one-dimensional laser speckle image and its application on real-time audio signal acquisition. In 2020 the 6th In- ternational Conference on Communication and Information Processing, pages 128–134, 2020. 2

  48. [56]

    The 20k samples-per- second real time detection of acoustic vibration based on dis- placement estimation of one-dimensional laser speckle im- ages

    Nan Wu and Shinichiro Haruyama. The 20k samples-per- second real time detection of acoustic vibration based on dis- placement estimation of one-dimensional laser speckle im- ages. Sensors, 21(9):2938, 2021

  49. [57]

    Simultaneous remote extraction of multiple speech sources and heart beats from secondary speckles pattern.Op- tics express, 17(24):21566–21580, 2009

    Zeev Zalevsky, Yevgeny Beiderman, Israel Margalit, Shimshon Gingold, Mina Teicher, Vicente Mico, and Javier Garcia. Simultaneous remote extraction of multiple speech sources and heart beats from secondary speckles pattern.Op- tics express, 17(24):21566–21580, 2009. 2, 3

  50. [58]

    Narasimhan

    Tianyuan Zhang, Mark Sheinin, Dorian Chan, Mark Rau, Matthew O’Toole, and Srinivasa G. Narasimhan. Analyz- ing physical impacts using transient surface wave imaging. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023. 2

Pith tools

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