Pith. sign in

REVIEW 3 major objections 5 minor 33 references

SINETRA: a Versatile Framework for Evaluating Single Neuron Tracking in Behaving Animals

T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read SINETRA generates annotated synthetic videos of neurons in deforming Hydra tissue, and its benchmark shows that all current trackers lose accuracy during fast, dense motion and in 3D.

desk verdict Useful simulator for a real gap, but the benchmark conclusions outrun the validation; deserves peer review. read the letter →

arxiv 2411.09462 v2 pith:3FIEWQPI submitted 2024-11-14 cs.CV

Raphael Reme , Alasdair Newson , Elsa Angelini , Jean-Christophe Olivo-Marin , Thibault Lagache This is my paper · ORCID
classification cs.CV
keywords singleneurontrackingsyntheticdatagenerationbehavinganimalsfluorescencemicroscopydeformablemotiondampedharmonicoscillatorsopticalflowbenchmark
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper introduces SINETRA, a simulator that generates annotated 2D and 3D fluorescence videos of neurons moving inside a deforming tissue, modeled on behaving Hydra vulgaris. The motivation is that tracking algorithms for behaving animals cannot currently be evaluated objectively, because real recordings lack ground-truth trajectories and manual annotation is impractical. SINETRA produces ground truth by construction, with particle positions, intensities, and shapes evolving through realistic deformation and shot noise. The paper uses this testbed to compare four tracking algorithms and reports that optical-flow-assisted Kalman tracking handles deformation best, while methods assuming near-constant velocity lose neurons during sudden contractions. If synthetic realism holds, this gives the community a standard way to measure and improve neuron trackers.

What carries the argument

The load-bearing mechanism is the generative model of fluorescence image formation and deformation. Image pixels combine particle and background Gaussian profiles with weights set by α, then pass through a Poisson shot-noise process with integration time Δ, so the simulator can dial the difficulty of detection. Motion is generated in two interchangeable ways: dense optical flow fields computed from an experimental Hydra video push every profile along measured deformation, or a lattice of control points connected by damped springs and driven by random contraction/elongation forces produces localized, smooth deformation, with particle and background positions interpolated by thin-plate splines. The oscillator equations use critical damping with relaxation time τ=10 frames so deformations are transient and physically plausible. This machinery turns arbitrary deformation fields into ground-truth tracks.

What would settle it

Run the same four trackers on real Hydra calcium-imaging videos with manual ground-truth tracks and compare their performance ranking to the SINETRA ranking; if the rankings differ, or if a classifier can reliably distinguish synthetic frames from real frames at matched α and Δ, the simulator's realism assumption fails.

Watch

Extended reading notes

Core claim

The central claim is that the difficulty of neuron tracking in behaving animals comes primarily from tissue deformation, and that a simulator can reproduce that deformation faithfully enough to benchmark trackers. SINETRA models each neuron as a weighted Gaussian spot whose position, orientation, and size evolve under damped harmonic dynamics, embedded in a background of large overlapping Gaussian profiles representing tissue auto-fluorescence; frames are formed by a linear mixture with contrast parameter α and Poisson shot noise with integration time Δ. Body motion is generated either by advecting all profiles with optical flow extracted from a real Hydra contraction video, or by solving a network of critically damped springs under random contraction/elongation forces and interpolating positions with thin-plate splines. On three scenarios the paper reports HOTA scores for u-track, eMHT, KOFT, and ZephIR, finding that KOFT exceeds 90 percent HOTA on 2D scenarios but every method degrades during fast dense motion and in 3D.

Load-bearing premise

The result transfers to real experiments only if SINETRA's synthetic videos reproduce both the appearance and the failure modes of genuine fluorescence recordings; the paper supports this visually and through the use of real optical flow, but not with a quantitative demonstration that tracker rankings match those on real annotated data.

Editorial extensions

If this is right

  • Neuron-tracking researchers can generate unlimited annotated 2D and 3D training and evaluation data without manual annotation.
  • The benchmark provides a concrete baseline: optical-flow-enhanced Kalman tracking is currently the strongest approach for 2D deforming tissue.
  • The results identify the open problem: fast, dense motion and 3D volumes, where all four trackers lose accuracy and ZephIR degrades sharply without many annotated frames.
  • Because α and Δ control contrast and noise, the simulator also allows systematic stress-testing of trackers across imaging conditions.

Reading between the lines

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

  • Editorial inference: Because the optical-flow scenario inherits the errors of the flow estimator, SINETRA may understate failures caused by motion discontinuities that optical flow smooths over, such as rapid somersaulting.
  • Editorial inference: The experiments fix particle fluorescence at constant intensity, so the benchmark does not yet test trackers against calcium-indicator flicker; adding time-varying intensities could plausibly change the ranking.
  • Editorial inference: The same spring-and-spline machinery could be applied to other deforming organisms by replacing the body mask and control-point grid, turning SINETRA into a general-purpose testbed for registration-based tracking.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. SINETRA is a simulator that generates synthetic fluorescence microscopy videos of particles (neurons) on a deformable tissue background. Particles are modeled as time-varying Gaussian spots, the background as large overlapping Gaussian profiles, and image noise as Poisson shot noise with user-controlled mixing and integration-time parameters. Motion is generated either by transporting particles along optical flow fields extracted from a real Hydra video (Hydra Flow scenario) or by interpolating the displacement of a damped spring system driven by random contraction/elongation forces (Springs 2D and 3D scenarios). The authors evaluate four tracking algorithms (u-track, eMHT, KOFT, ZephIR) on these synthetic datasets and report HOTA scores over five random runs. They conclude that KOFT performs best overall, that ZephIR requires well-chosen annotated frames, and that u-track and eMHT struggle with large contractions.

Significance. If the synthetic videos are representative of real fluorescence recordings, SINETRA would fill a genuine gap by providing annotated, customizable 2D and 3D benchmark data for neuron tracking in behaving animals. The paper is clearly written on the modeling side: the equations for particle appearance, background, noise, and damped-spring motion are explicit, and the code and datasets are publicly released. The authors also report mean and standard deviation over five random seeds, and they include a scenario based on real optical flow. However, the central claim that the simulations 'closely mimic live animal recordings' is currently supported only by visual inspection and a single flow-extraction scenario. Because the benchmark's conclusions about algorithm limitations are meant to transfer to real experiments, this validation gap is load-bearing. The paper also contains an internal inconsistency in the reported ZephIR@3 Hydra Flow score.

major comments (3)
  1. [Sections 2.2.1 and 3.1] The central claim that SINETRA 'closely mimics live animal recordings' is supported only by qualitative side-by-side images (Figure 1c,d) and by using Farneback optical flow from one experimental video in the Hydra Flow scenario. There is no quantitative validation that the simulated image statistics, particle density, detection difficulty, or motion characteristics are representative of real fluorescence videos of behaving animals. As a result, it is not established that the ranking of tracking algorithms on SINETRA transfers to real experimental data. Please add a validation experiment—for example, comparing image statistics (noise level, background power spectrum, particle signal-to-noise ratio) between real and simulated frames, or running the benchmark on a small set of manually annotated real sequences—or alternatively temper the abstract and conclusions to state that realism is assumed rather than demonstrated.
  2. [Section 3.3 vs Table 1] The text states that 'Using 3 annotated frames ... it outperforms u-track and eMHT on Hydra Flow scenario, reaching 80.9% HOTA,' but Table 1 lists ZephIR@3 on Hydra Flow as 76.4 ± 6.0%. The reported standard deviations of u-track (67.7 ± 5.7%) and eMHT (73.4 ± 8.1%) overlap substantially with this value, so the comparative claim is not supported without a significance test. Please correct the numerical inconsistency and either add statistical tests or explicitly state which differences are not significant.
  3. [Sections 2.2.1 and 3.3] The Hydra Flow scenario generates ground-truth motion by applying optical flow computed from an experimental video, while KOFT explicitly integrates optical-flow estimates into its Kalman filter. This creates a confound: the winning method in that scenario is given access to the same motion cue used to generate the data, which may inflate its measured advantage. Although KOFT also performs well on the Springs scenarios, the Hydra Flow result should be interpreted with this caveat. Please discuss this confounding factor explicitly, and ideally include a control condition, such as running KOFT without optical flow or generating optical-flow-based motion with a different (e.g., coarse-to-fine) method.
minor comments (5)
  1. [Section 3.3] The word 'sligthly' should be 'slightly'.
  2. [Section 2.2.3] The choice of critical time τ = 10 frames is described as 'reasonable,' but no sensitivity analysis or justification is provided. Please add a brief motivation or a reference to experimental timescales.
  3. [Section 3.2] The pre-detection F1 score of 'around 80% across our synthetic dataset' is not detailed. Please specify which detection method was used (the reference to wavelet thresholding is given) and how the F1 was measured, including whether it varies by scenario and by particle density.
  4. [Section 3.1] The HOTA tolerance η = 2 pixels is defined, but the sensitivity of the rankings to this choice is not examined. A short discussion or supplementary experiment varying η would strengthen the benchmarking claims.
  5. [Table 1] The notation 'ZephIR@3' and 'ZephIR@10' is not defined in the table caption or in the text where the table is introduced. Please define these abbreviations.

Circularity Check

0 steps flagged · score 2.0 of 10

No load-bearing circularity; the SINETRA construction and benchmark do not reduce to their inputs, though the KOFT/optical-flow overlap and unvalidated realism claim are validity concerns.

full rationale

Walking the paper's chain: Eqs. (1)-(3) define particles, background, and noise; Section 2.2 defines motion (optical-flow and springs); Section 3.1 defines HOTA evaluation on generated videos. No step defines a derived quantity in terms of the quantity it later predicts. The 'closely mimicking live animal recordings' claim is qualitative, supported only by side-by-side crops (Fig. 1c/d) and by injecting Farneback optical flow from a real video into the Hydra Flow scenario; this is weak external validation, not circularity. The strongest confound is that KOFT (ref [6], from the same authors) wins on Hydra Flow, where ground-truth motion is generated from the same optical-flow cue KOFT uses; this makes the benchmark model-matched in KOFT's favor, but KOFT still has to detect particles and estimate flow from the rendered frames, so its score is not identical to the generator's input by construction. There is no fitted parameter renamed as a prediction and no self-citation used to forbid alternatives. The paper is also open about code/data availability. One internal inconsistency exists: Section 3.3 reports ZephIR@3 at 80.9% HOTA on Hydra Flow while Table 1 lists 76.4 ± 6.0%; this is an error to correct, not circular reasoning. Overall, no derivation-level circularity; score 2 reflects one minor self-referential benchmarking choice, not a circular derivation.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

No new physical entities are introduced; control points and springs are computational devices used to generate motion, not claims about the world. The simulator relies on several hand-chosen parameters, especially alpha, Delta, amax, and tau, and on domain assumptions about Gaussian spot models and optical flow representing true tissue deformation. The most consequential free parameter is the choice of motion model for each scenario, because it directly shapes which tracker wins the benchmark.

free parameters (7)
  • alpha (particle/background mixing) = 0.20 (default)
    Controls the mixing of particle signal and background fluorescence; chosen by hand, not fit to data; affects all generated images and therefore benchmark difficulty.
  • Delta (integration time) = 50
    Poisson shot-noise integration time; controls noise level; chosen by hand; affects detection and tracking difficulty.
  • amax (random force amplitude) = 4 pixels (2D springs), 3 pixels (3D springs)
    Maximum random force amplitude in the springs motion model; chosen by hand; controls deformation magnitude.
  • tau (critical damping time) = 10 frames
    Decay time for critical damping of harmonic oscillators; chosen by hand; sets temporal smoothness of motion.
  • HOTA tolerance eta = 2 pixels
    Matching tolerance for evaluation; chosen by hand; affects HOTA scores.
  • minimal distance between particles = not specified
    Controls particle sparsity; named in Section 2.1.1 but value not reported.
  • ZephIR annotation count = 3 and 10 frames
    Number of annotated frames given to ZephIR; chosen by hand; strongly affects ZephIR scores.
assumptions (5)
  • domain assumption Gaussian profiles adequately model fluorescent particles and background tissue.
    Central to the image generation (Eqs. 1-3); no validation against real PSFs or background statistics.
  • domain assumption The 1D single-spring damped harmonic oscillator analysis extrapolates to the n-body spring system.
    Section 2.2.3 states the extrapolation without proof; stability and damping of the multi-spring system are assumed.
  • domain assumption Farneback optical flow estimates of real Hydra videos approximate true tissue deformation.
    Section 2.2.1 uses these flows as ground-truth motion; errors in the flow become motion errors in the synthetic data.
  • ad hoc to paper Critically damped settings (lambda=2/tau, k=1/tau^2) with tau=10 frames make motion realistic.
    Values chosen for plausibility; no data support.
  • standard math Poisson shot noise with integration time Delta models camera noise.
    Common model in microscopy simulation, standard assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SINETRA: a Versatile Framework for Evaluating Single Neuron Tracking in Behaving Animals." pith.science (2026). https://pith.science/paper/3FIEWQPI

@misc{pith2026241109462,
  author       = {Pith},
  title        = {Pith review of: SINETRA: a Versatile Framework for Evaluating Single Neuron Tracking in Behaving Animals},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3FIEWQPI}},
  note         = {Machine review of arXiv:2411.09462}
}
read the original abstract

Accurately tracking neuronal activity in behaving animals presents significant challenges due to complex motions and background noise. The lack of annotated datasets limits the evaluation and improvement of such tracking algorithms. To address this, we developed SINETRA, a versatile simulator that generates synthetic tracking data for particles on a deformable background, closely mimicking live animal recordings. This simulator produces annotated 2D and 3D videos that reflect the intricate movements seen in behaving animals like Hydra Vulgaris. We evaluated four state-of-the-art tracking algorithms highlighting the current limitations of these methods in challenging scenarios and paving the way for improved cell tracking techniques in dynamic biological systems.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages

  1. [1]

    These animals undergo significant body deforma- tions during behaviors such as worm crawling or Hydra’s contrac- tion and complex somersaulting [1]

    INTRODUCTION Numerous algorithms have been developed to track neurons in fluo- rescence live imaging of behaving animals, particularly in the worm Caenorhabditis (C.) Elegans, Zebrafish or the freshwater cnidarian Hydra Vulgaris. These animals undergo significant body deforma- tions during behaviors such as worm crawling or Hydra’s contrac- tion and compl...

    work page

  2. [2]

    SINETRA: a Versatile Framework for Evaluating Single Neuron Tracking in Behaving Animals

    METHOD 2.1. Modeling fluorescent particles and background 2.1.1. Particles We first modeled realistic fluorescent particles (neurons). A parti- cle is characterized by its position, intensity and shape, all of which change over time. Formally, each particle 1 ≤ i ≤ N in frame arXiv:2411.09462v2 [cs.CV] 15 Nov 2024 This work has been submitted to the IEEE ...

    work page Pith review arXiv 2024

  3. [3]

    Evaluation Using our simulator, we generated a synthetic dataset of fluorescent neurons in a behaving animal

    RESULTS AND DISCUSSIONS 3.1. Evaluation Using our simulator, we generated a synthetic dataset of fluorescent neurons in a behaving animal. We simulated 800 neurons over 200 frames, using ∆ = 50 and α = 0.2. We focused on three different scenarios:

    work page

  4. [4]

    The flows are extracted from 200 selected frames of an experimental video where the Hydra Vulgaris is contracting [14]

    Hydra Flow: we generated 1024x1024 images using optical- flow based motion. The flows are extracted from 200 selected frames of an experimental video where the Hydra Vulgaris is contracting [14]

    work page

  5. [5]

    Springs 2D: we used springs-based motion (with the amplitude of random forces amax = 4 pixels) to generate 1024x1024 im- ages

    work page

  6. [6]

    Springs 3D: we simulated 200x200x200 volumes with springs- based motion (with amax = 3 pixels). For a set of ground-truth and computed tracks, we measure the ac- curacy of the tracking algorithm with the HOTA score [18] that esti- mates both localization, detection and association performance. We used a tolerance distance η = 2 pixels (i.e. a predicted pa...

    work page

  7. [7]

    Fast deep neural cor- respondence for tracking and identifying neurons in c. elegans using semi-synthetic training,

    X. Yu, M. S. Creamer, F. Randi, et al., “Fast deep neural cor- respondence for tracking and identifying neurons in c. elegans using semi-synthetic training,”Elife, vol. 10, pp. e66410, 2021

    work page 2021

  8. [8]

    Auto- mated neuron tracking inside moving and deforming c. elegans using deep learning and targeted augmentation,

    C. F. Park, M. Barzegar-Keshteli, K. Korchagina, et al., “Auto- mated neuron tracking inside moving and deforming c. elegans using deep learning and targeted augmentation,” Nature Meth- ods, vol. 21, no. 1, pp. 142–149, 2024

    work page 2024

Show all 33 references
  1. [9]

    CONCLUSION In this paper, we presented SINETRA, a versatile framework to gen- erate synthetic annotated single-particle-tracking datasets with a re- alistic imaging noise and complex motions such as those experienced by neurons in behaving animals like Hydra Vulgaris or C. ele...

  2. [10]

    COMPLIANCE WITH ETHICAL STANDARDS This is a numerical simulation study for which no ethical approval was required

  3. [11]

    R.R and T.L

    ACKNOWLEDGMENTS This work is supported by the Institut Pasteur and France-BioImaging Infrastructure (ANR-10-INBS-04). R.R and T.L. are supported by the ANR (ANR-21-CE45-0020-01 REBIRTH). None of the authors declare to have a financial conflict of interest in the results of this study

  4. [12]

    Peptide-driven control of som- ersaulting in hydra vulgaris,

    W. Yamamoto and R. Yuste, “Peptide-driven control of som- ersaulting in hydra vulgaris,” Current Biology, vol. 33, no. 10, pp. 1893–1905, 2023

    1905

  5. [13]

    Robust single- particle tracking in live-cell time-lapse sequences,

    K. Jaqaman, D. Loerke, M. Mettlen, et al., “Robust single- particle tracking in live-cell time-lapse sequences,” Nature Methods, vol. 5, no. 8, pp. 695–702, 2008

    2008

  6. [14]

    Multiple hypothesis tracking for cluttered biological image sequences,

    N. Chenouard, I. Bloch, and J.-C. Olivo-Marin, “Multiple hypothesis tracking for cluttered biological image sequences,” IEEE Transactions on Pattern Analysis and Machine Intelli- gence, vol. 35, no. 11, pp. 2736–3750, 2013

    2013

  7. [15]

    Tracking cal- cium dynamics from individual neurons in behaving animals,

    T. Lagache, A. Hanson, J. P ´erez-Ortega, et al., “Tracking cal- cium dynamics from individual neurons in behaving animals,” PLoS Computational Biology, vol. 17, pp. e1009432, 10 2021

    2021

  8. [16]

    Tracking intermittent particles with self-learned visual features,

    R. Reme, V . Piriou, A. Hanson, et al., “Tracking intermittent particles with self-learned visual features,” in IEEE Interna- tional Symposium on Biomedical Imaging (ISBI). IEEE, 2023, pp. 1–5

    2023

  9. [17]

    Particle tracking in biological images with optical-flow enhanced kalman filter- ing,

    R. Reme, A. Newson, E. Angelini, et al., “Particle tracking in biological images with optical-flow enhanced kalman filter- ing,” in 2024 IEEE International Symposium on Biomedical Imaging (ISBI), 2024, pp. 1–5. This work has been submitted to the IEEE for possible publication....

    2024

  10. [18]

    Hota: A higher or- der metric for evaluating multi-object tracking,

    J. Luiten, A. Osep, P. Dendorfer, et al., “Hota: A higher or- der metric for evaluating multi-object tracking,” International Journal of Computer Vision, vol. 129, pp. 548–578, 2021

    2021

  11. [19]

    Fiji: an open-source platform for biological-image analysis,

    J. Schindelin, I. Arganda-Carreras, E. Frise, et al., “Fiji: an open-source platform for biological-image analysis,” Nature Methods, vol. 9, no. 7, pp. 676–682, 2012

    2012

  12. [20]

    Versatile mul- tiple object tracking in sparse 2d/3d videos via deformable im- age registration,

    J. Ryu, A. Nejatbakhsh, M. Torkashvand, et al., “Versatile mul- tiple object tracking in sparse 2d/3d videos via deformable im- age registration,” PLoS Computational Biology, vol. 20, no. 5, pp. e1012075, 2024

    2024

  13. [21]

    (3) KOFT [6] implemented in ByoTrack [13] is a probabilistic algorithm that exploits optical flow within Kalman This work has been submitted to the IEEE for possible publication

    is a probabilistic algorithm based on multiple motion models, with a probabilistic handling of tracks and a multiple hypothesis association method. (3) KOFT [6] implemented in ByoTrack [13] is a probabilistic algorithm that exploits optical flow within Kalman This work has bee...

  14. [22]

    to detect particles for which we measured a detection f1 score of around 80% across our synthetic dataset. 3.3. Results Table 1 summarizes the performance of the different tracking algo- rithms on the three simulation scenarios. u-track [2, 19, 20] and eMHT [3, 21] model a nea...

  15. [23]

    Mitogen: a framework for gen- erating 3d synthetic time-lapse sequences of cell populations in fluorescence microscopy,

    D. Svoboda and V . Ulman, “Mitogen: a framework for gen- erating 3d synthetic time-lapse sequences of cell populations in fluorescence microscopy,” IEEE Transactions on Medical Imaging, vol. 36, no. 1, pp. 310–321, 2016

    2016

  16. [24]

    Bsim 2.0: an advanced agent-based cell simulator,

    A. Matyjaszkiewicz, G. Fiore, F. Annunziata, et al., “Bsim 2.0: an advanced agent-based cell simulator,” ACS Synthetic Biology, vol. 6, no. 10, pp. 1969–1972, 2017

    1969

  17. [25]

    An objective comparison of cell-tracking algorithms,

    V . Ulman, M. Maˇska, K. E. Magnusson, et al., “An objective comparison of cell-tracking algorithms,” Nature Methods, vol. 14, no. 12, pp. 1141–1152, 2017

    2017

  18. [26]

    Automatic moni- toring of neural activity with single-cell resolution in behaving hydra,

    A. Hanson, R. Reme, N. Telerman, et al., “Automatic moni- toring of neural activity with single-cell resolution in behaving hydra,” Scientific Reports, vol. 14, no. 1, pp. 5083, 2024

    2024

  19. [27]

    Non-overlapping neural networks in hydra vulgaris,

    C. Dupre and R. Yuste, “Non-overlapping neural networks in hydra vulgaris,” Current Biology, vol. 27, no. 8, pp. 1085– 1097, 2017

    2017

  20. [28]

    A high signal-to-noise ca2+ probe composed of a single green fluorescent protein,

    J. Nakai, M. Ohkura, and K. Imoto, “A high signal-to-noise ca2+ probe composed of a single green fluorescent protein,” Nature Biotechnology, vol. 19, no. 2, pp. 137–141, 2001

    2001

  21. [29]

    Fluorescence microscopy today,

    R. Yuste, “Fluorescence microscopy today,” Nature Methods, vol. 2, no. 12, pp. 902–904, 2005

    2005

  22. [30]

    Two-frame motion estimation based on poly- nomial expansion,

    G. Farneb ¨ack, “Two-frame motion estimation based on poly- nomial expansion,” in Image Analysis: 13th Scandinavian Conference, SCIA. Springer, 2003, pp. 363–370

    2003

  23. [31]

    Trackmate: An open and extensible platform for single-particle tracking,

    J.-Y . Tinevez, N. Perry, J. Schindelin, et al., “Trackmate: An open and extensible platform for single-particle tracking,” Methods, vol. 115, pp. 80–90, 2017

    2017

  24. [32]

    Icy: an open bioimage informatics platform for extended repro- ducible research,

    F. De Chaumont, S. Dallongeville, N. Chenouard, et al., “Icy: an open bioimage informatics platform for extended repro- ducible research,” Nature Methods, vol. 9, no. 7, pp. 690–696, 2012

    2012

  25. [33]

    Extraction of spots in biological images using multiscale products,

    J.-C. Olivo-Marin, “Extraction of spots in biological images using multiscale products,” Pattern Recognition, vol. 35, no. 9, pp. 1989–1996, 2002

    1989

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.