Pith. sign in

REVIEW 3 major objections 6 minor 85 references

A geometric and deep learning reproducible pipeline for monitoring floating anthropogenic debris in urban rivers using in situ cameras

T0 review · 3 major / 6 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read The study asserts that a fixed riverside camera, combining deep-learning detection with pinhole-camera geometry and regression corrections, can estimate floating debris size at near-resolution accuracy (about 1.3 cm per pixel).

desk verdict Useful applied pipeline for sizing river debris from bank cameras, with a genuinely informative data-leakage experiment — but the headline centimeter-level accuracy rests on regression corrections that appear to be fit and evaluated on the same data, so the claim is not yet established. read the letter →

arxiv 2510.23798 v2 pith:YXTEABV7 submitted 2025-10-27 cs.CV cs.AIcs.LG

Gauthier Grimmer , Romain Wenger , Clément Flint , Germain Forestier , Gilles Rixhon , Valentin Chardon This is my paper
classification cs.CVcs.AIcs.LG
keywords floatingdebrisurbanriversin-situcamerasdeeplearningobjectdetectionpinholecameramodelprojectivegeometrydataleakage
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 is trying to establish that low-cost, fixed in-situ cameras can continuously detect floating debris in urban rivers and estimate the real-world width and height of each detected object from a single 2D image. The size estimation uses a pinhole-camera geometric model that projects each detection bounding box into 3D using the camera's focal length, sensor size, mounting height, and tilt angle, then converts the box into metric dimensions. Two regression corrections, one for bounding-box shape and one for predicted dimensions, are added to absorb systematic biases. On the paper's leakage-free test setup, corrected RMSEs are about 2.14 cm for width and 1.68 cm for height, which the authors compare with a per-pixel sensitivity of about 1.34-1.36 cm and argue that errors are near what image resolution physically allows. If true, this offers a practical path to continuous, dimension-resolved debris monitoring that could support plastic flux and mass estimates in rivers.

What carries the argument

The central mechanism is the pinhole-camera ray-projection model: each pixel is converted into a 3D ray using focal lengths expressed in pixels, a pitch rotation places the ray in world coordinates, and the ray is intersected with a horizontal reference plane (the water surface). The four edges of a detected bounding box become four planes through the camera center, and object width and height are computed by summing point-to-plane distances from the projected box center. The second key component is a two-stage regression correction stack: a polynomial or linear fit corrects the predicted bounding-box shape against manual annotations, and a second fit corrects the resulting metric dimensions

What would settle it

Take the same detection and projection pipeline, fit the box-shape and dimension corrections on a training subset, then evaluate RMSE/MAE on a disjoint subset of scenes never used for fitting. If corrected errors on that held-out set exceed roughly 4 cm or clearly diverge from the reported 1.68-2.14 cm range, the near-resolution-limit claim is not supported.

Watch

Extended reading notes

Core claim

The central claim is that monocular bank-mounted cameras can recover centimetre-level physical dimensions of floating debris without any in-water measurement, provided the camera's intrinsic and extrinsic parameters are known and systematic biases are corrected. The pipeline works as follows: a YOLO-family detector supplies axis-aligned bounding boxes; each box is lifted into 3D via rays through a pinhole model; the central ray is intersected with a reference ground plane to locate the object; and width and height are recovered as sums of distances from that point to the four planes defined by the box edges. On a leakage-free dataset, the uncorrected RMSEs are roughly 3.16 cm (width) and 3.5

Load-bearing premise

The load-bearing premise is that the regression corrections fitted on the calibration objects also apply to unseen objects; because the paper does not report a held-out split or cross-validation for these coefficients, the headline centimetre-level errors could reflect in-sample training accuracy rather than predictive accuracy.

Editorial extensions

If this is right

  • If the claimed accuracy holds in deployment, a single riverside camera could yield continuous per-object width and height time series for floating debris, enabling plastic mass and flux estimates without manual sampling.
  • The cluster-based, leakage-free data split shows that random temporal splits substantially overstate detector performance; adopting similar split protocols would make reported metrics of river-debris models more comparable.
  • Because the geometric model depends only on camera calibration parameters, the same projection code can be redeployed at new fixed-camera sites, with the paper's qualitative test on a second river suggesting cross-site transfer is plausible.
  • Once errors approach the per-pixel sensitivity bound, further accuracy gains require higher-resolution sensors or more precise calibration rather than additional regression corrections.
  • The mixed effect of adding negative images (beneficial for a small model, not for a large one) indicates that dataset augmentation choices should be evaluated per architecture rather than assumed universally beneficial.

Reading between the lines

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

  • An implication the authors leave implicit is that the fitted regression corrections are likely site-specific: they absorb the particular camera's tilt, height, and lens behavior, so deploying the pipeline at a new location probably requires recalibration or a small on-site validation set.
  • A testable extension is to adopt the per-pixel sensitivity metric as a reporting standard for monocular sizing pipelines, since it converts a sensor property into a concrete lower bound on achievable error.
  • If prospective validation succeeds, the pipeline could feed object-level mass models, but the paper's kayak-misclassification caveat suggests a size-versus-label consistency check would be a practical safeguard before aggregating mass estimates.
  • Future detectors with rotated bounding boxes or instance segmentation could remove the axis-aligned-box distortion the paper identifies for tilted objects, while leaving the geometric projection itself unchanged.
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

3 major / 6 minor

Summary. The paper proposes a reproducible pipeline for monitoring floating anthropogenic debris in urban rivers using fixed in-situ cameras. It combines manual annotation of 6,013 images, training of several YOLO variants under different data-splitting protocols (random split, 30% negative images, and a t-SNE/DBSCAN cluster-based split intended to remove temporal leakage), and a pinhole-camera geometric model that reconstructs pixel rays and intersects them with a reference plane to estimate metric object widths and heights. Two regression-based corrections are then applied to the bounding-box shape and to the predicted dimensions, and end-to-end errors are reported for YOLOv11-m models with and without the leak-free split. The paper reports best corrected RMSEs of 2.14 cm (width) and 1.68 cm (height) for the leak-free model, and interprets these as approaching the per-pixel sensitivity limit of roughly 1.3 cm.

Significance. If the reported metric-sizing accuracy is genuinely predictive, the paper offers a valuable, low-cost monitoring tool: a fixed monocular camera, a standard geometric projection, and deep-learning detections could provide continuous debris-size estimates for flux and mass calculations. The manuscript is strong in its explicit treatment of temporal data leakage for the detection task, its clear presentation of the pinhole projection geometry (§2.6), and its commitment to reproducibility (code and model weights are public). However, the central claim of near-resolution-limit accuracy currently rests on regression corrections that are fitted and evaluated without a described held-out calibration split, and the per-pixel sensitivity benchmark used to justify 'near resolution limits' is not a formal lower bound. The significance is therefore real but conditional on additional validation.

major comments (3)
  1. [§2.7, §3.3, Tables 3–4] The two regression corrections are applied to the same objects whose ground-truth measurements are used both to fit the regression coefficients and to compute the reported RMSE/MAE. The manuscript describes the reference data for the first correction as 'dimensions predicted on the manually annotated bounding boxes' and for the second as 'manually measuring the dimensions of the objects,' but it never states that the coefficients were fit on a calibration set distinct from the evaluation set, nor describes any cross-validation, leave-one-object-out, or temporal split for this correction stage. As written, the headline values (e.g., 2.14 cm width, 1.68 cm height for the leak-free YOLOv11-m in Table 4) are in-sample calibration errors rather than predictive accuracy estimates. Because the conclusion in §4.2.3 and the abstract's 'image resolution limits' claim depend on these numbers, the a
  2. [§3.3.3, Eq. (18)] The per-pixel sensitivity metric S_pixel is defined as the average absolute change in estimated size after a deterministic one-pixel shift of the bounding box. This is not a lower bound on RMSE or MAE: a well-calibrated estimator can average out pixel-level noise and achieve errors below S_pixel, while RMSE can exceed S_pixel because it penalizes outliers and systematic bias. The statement that RMSEs of 2.14 cm and 1.68 cm are 'not substantially greater' than thresholds of 1.34 cm and 1.36 cm, and therefore the method is 'nearly as precise as what is physically achievable,' is not quantitatively supported. A proper error-propagation analysis or a localization-noise simulation is needed to justify the resolution-limit interpretation; otherwise the claim should be softened to an empirical reference value.
  3. [§2.3, §2.4, §3.2.3] The 'leak-free' dataset is constructed by applying t-SNE and DBSCAN to all 6,013 images before the 80/10/10 split into train/validation/test. Because the t-SNE embedding is estimated on the union of the three subsets, the test images influence the low-dimensional coordinates and hence the cluster assignments used to form the split. This unsupervised preprocessing step itself constitutes a potential source of leakage: information from the test set is used to define the scene groups. Consequently, Train 9 and Train 10 are not guaranteed to be leak-free, and the observed drop in mAP between the random split and the cluster split may overstate the effect of temporal leakage. A stricter protocol would define scene groups from acquisition timestamps or fit the embedding/clustering on training data only, then assign new scenes to the resulting clusters.
minor comments (6)
  1. [Eq. (1)] The t-SNE conditional probability notation is inconsistent: the subscript is 'a|b' but the denominator sums over k with 'a ≠ k', and the normalization is not clearly over the correct index. Please rewrite using standard t-SNE notation and ensure the formula is mathematically correct.
  2. [§3.3.1] For YOLOv11-m (Train 6) height, the text says linear regression 'reduces errors' but the reported MAE increases from 1.06 cm to 1.25 cm while RMSE decreases from 2.29 cm to 1.73 cm. Please clarify whether this is a typo or whether the claim is intended to refer to RMSE only.
  3. [§3.3.3] The sentence 'geometric-level correction on its one is more efficient of box-level correction only' is garbled. Also, the general statement that 'double level correction reaches better RMSE and MAE than one level correction' is contradicted by Table 3b (best double RMSE 1.86 cm vs. best single 1.83 cm) and by the instability of polynomial dimension correction alone in Table 4b (RMSE 5.80 cm). Please qualify the claim to the specific best configurations.
  4. [§2.2.3, §2.6, Table 1] Typos and wording: 'Daily insolution duration' should be 'insolation'; 'we compute the the ray direction' in §2.6; 'Technical drive characteristics' in the Table 1 caption. A light copyedit would improve readability.
  5. [§4.3] The Bruche River test is presented as a spatial generalization check but only qualitatively. If feasible, report quantitative results (e.g., precision/recall on a small labeled subset or detection counts) to support the claim of generalization to a different river.
  6. [§2.5.3] The training comparison would be easier to interpret if the hyperparameter search were described more explicitly. In particular, 'the number of epochs is set to 200' but several models stop earlier; reporting the selected epoch for each model would help reproducibility.

Circularity Check

1 steps flagged · score 6.0 of 10

Reported centimeter-level accuracy after regression corrections is in-sample: the manually measured sizes used to fit the corrections are the same sizes used to compute the reported RMSE/MAE.

  1. fitted input called prediction [§2.7, §3.3.2, §3.3.3, Eq. 16]
    "The second correction (Predicted dimensions correction) aims to correct the dimensions (width and length) predicted by the geometric model ... the reference data is created by manually measuring the dimensions of the objects. ... polynomial regression yielded optimal results, reducing the RMSE from 4.02 cm to 1.72 cm and the MAE from 3.58 cm to 1.54 cm. ... For this final experiment, we adopt an end-to-end evaluation of the complete pipeline ... Table 3 and 4 summarizes the mean absolute error (MAE) and root mean square error (RMSE) obtained for each correction strategy."

    The correction stage fits linear/polynomial regressions to manually measured object dimensions (§2.7). The subsequent evaluation computes RMSE via Eq. 16 using the same manually measured dimensions as ground truth (Tables 3–4). No held-out split, cross-validation, or otherwise separate calibration set for the correction coefficients is described anywhere in §2.7 or §3.3. Thus the post-correction errors (e.g., RMSE 2.14 cm width, 1.68 cm height for the leak-free model) are in-sample residuals of the fitted regressions, not out-of-sample prediction errors. Presenting these as 'predictions' and comparing them with the per-pixel sensitivity lower bound (§3.3.3, 'nearly as precise as what is physically achievable') turns fitted values into the paper's headline accuracy claim.

full rationale

The pinhole-camera geometric projection itself (§2.6) is self-contained and non-circular: it derives ray directions from intrinsics/extrinsics and intersects them with a ground plane, with no fitted parameter smuggled into the geometry. The YOLO detection evaluation is also based on explicit train/validation/test protocols, including a leak-free cluster split described in §2.3–2.4. However, the paper's headline metric-size claim depends on the regression corrections introduced in §2.7 and evaluated in §3.3. The text never states that the linear/polynomial correction coefficients were fitted on a calibration set distinct from the evaluation set, and it reports RMSE/MAE before/after correction using the same manually measured dimensions that served as the regression targets. Under the stated protocol, the corrected errors are in-sample training residuals, so the 'near resolution limit' conclusion in §4.2.3 and the Conclusion is not established for unseen objects. The §2.3 data-leakage discussion concerns YOLO dataset splits (random vs. clustered), not the correction-fitting protocol, so it does not mitigate this statistical circularity. No load-bearing self-citation or imported uniqueness theorem is present; the issue is specifically that the 'predicted' corrected dimensions reduce to fitted values when evaluated.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

No new physical entities are introduced. The load-bearing input from outside the paper is a set of standard ML/geometry assumptions plus a calibration procedure that is never validated out-of-sample. The largest burden is the unverified generalization of the regression corrections.

free parameters (3)
  • Regression correction coefficients (linear/polynomial) for box shape and dimensions = not reported
    Fitted to manually measured ground-truth dimensions on the same objects whose errors are later reported; coefficients not listed. These corrections carry much of the claimed accuracy gain (e.g., width RMSE from 3.16 to 2.14 cm).
  • t-SNE/DBSCAN hyperparameters = perplexity=30, learning rate=200, eps=5, min_samples=10
    Chosen by hand with no sensitivity analysis. They define which images are considered 'same scene' and therefore control whether the leakage-free split is actually leakage-free.
  • Negative image fraction = 30%
    Arbitrary choice affecting YOLOv8-n-neg30 and YOLOv11m-neg30 results; no sweep over the fraction is reported.
assumptions (6)
  • domain assumption Pinhole camera model: each pixel corresponds to a ray through the optical center; lens distortion is ignored.
    Invoked in §2.6: 'We adopt the pinhole camera model... ignoring lens distortion and treating the projection as perfectly linear.'
  • domain assumption The water surface acts as a flat ground plane, and the object center lies on that plane.
    Used in §2.6: the central ray is intersected with the ground plane to obtain object center O. §4.2.1 concedes this is 'not theoretically justified' but operationally effective.
  • domain assumption Axis-aligned bounding boxes sufficiently represent the physical extents of objects.
    The whole sizing pipeline uses bounding-box corners. §4.2.2 acknowledges rotated objects are distorted and that evaluation was restricted to horizontally/vertically oriented objects.
  • domain assumption Manually annotated bounding boxes and manually measured object dimensions are error-free ground truth.
    All detection metrics and correction fits use these manual measurements as reference (§2.2.2, §2.7).
  • ad hoc to paper The regression corrections fitted to ground truth generalize to new data.
    Assumed implicitly in §3.3 where corrected RMSE/MAE are reported without any held-out calibration split or cross-validation.
  • ad hoc to paper t-SNE + DBSCAN clusters correspond to distinct temporal scenes, so no temporal leakage remains.
    Used to build the 'leak-free' datasets in §2.3–2.4. Only DBCV=0.98 is reported; no external validation that clusters are exactly scenes.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A geometric and deep learning reproducible pipeline for monitoring floating anthropogenic debris in urban rivers using in situ cameras." pith.science (2026). https://pith.science/paper/YXTEABV7

@misc{pith2026251023798,
  author       = {Pith},
  title        = {Pith review of: A geometric and deep learning reproducible pipeline for monitoring floating anthropogenic debris in urban rivers using in situ cameras},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YXTEABV7}},
  note         = {Machine review of arXiv:2510.23798}
}
read the original abstract

The proliferation of floating anthropogenic debris in rivers has emerged as a pressing environmental concern, exerting a detrimental influence on biodiversity, water quality, and human activities such as navigation and recreation. The present study proposes a novel methodological framework for the monitoring the aforementioned waste, utilising fixed, in-situ cameras. This study provides two key contributions: (i) the continuous quantification and monitoring of floating debris using deep learning and (ii) the identification of the most suitable deep learning model in terms of accuracy and inference speed under complex environmental conditions. These models are tested in a range of environmental conditions and learning configurations, including experiments on biases related to data leakage. Furthermore, a geometric model is implemented to estimate the actual size of detected objects from a 2D image. This model takes advantage of both intrinsic and extrinsic characteristics of the camera. The findings of this study underscore the significance of the dataset constitution protocol, particularly with respect to the integration of negative images and the consideration of temporal leakage. In conclusion, the feasibility of metric object estimation using projective geometry coupled with regression corrections is demonstrated. This approach paves the way for the development of robust, low-cost, automated monitoring systems for urban aquatic environments.

Figures

Figures reproduced from arXiv: 2510.23798 by the authors.

Figure 1
Figure 1. Employed methodology : (1) Acquired data is manually annotated to constitute [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Location of Steingiessen River and inlet and outlet cameras (north of Greater [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Data acquired on March 4, 2025 upstream (left) and downstream (right) of the [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (13 more)
Figure 4
Figure 4. Figure 4: Evolution of Daily Insolation Duration (INST), Daily Global Radiation (GLOT) [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: General architecture of a YOLO object detection model. The structure is [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: Visualization of the object size estimation process. Subfigures 6a and 6b il [PITH_FULL_IMAGE:figures/full_fig_p021_6.png]
Figure 7
Figure 7. Figure 7: Regression-based correction pipeline 2.7.1. Regression-based correction For both corrections, two parametric regression models were used : simple linear regression and polynomial regression. In parametric regression, the number of parameters is finite and a specific fu…
Figure 8
Figure 8. Figure 8: Clusters applied to dimensionally reduced data, occurrence of object classes in [PITH_FULL_IMAGE:figures/full_fig_p025_8.png]
Figure 9
Figure 9. Figure 9: YOLOv8-n training results where 1 = Anthropogenic Debris ; 2 = Natural [PITH_FULL_IMAGE:figures/full_fig_p028_9.png]
Figure 10
Figure 10. Figure 10: YOLOv11-m training result where 1 = Anthropogenic Debris ; 2 = Natural [PITH_FULL_IMAGE:figures/full_fig_p030_10.png]
Figure 11
Figure 11. Figure 11: YOLOv11-m (Train 6) bounding box shape correction [PITH_FULL_IMAGE:figures/full_fig_p031_11.png]
Figure 12
Figure 12. Figure 12: Data leak-free YOLOv11-m (Train 10) bounding box shape correction [PITH_FULL_IMAGE:figures/full_fig_p033_12.png]
Figure 13
Figure 13. Figure 13: Bounding box dimensions correction 3.3.3. Object size prediction with corrections applied For this final experiment, we adopt an end-to-end evaluation of the com￾plete pipeline. The YOLOv11-m (Train 6 and 10) models, identify bounding boxes on which a shape correction…
Figure 14
Figure 14. Figure 14: Example of parallel (left) and inclined (right) bottle [PITH_FULL_IMAGE:figures/full_fig_p045_14.png]
Figure 15
Figure 15. Figure 15: Examples of mirror effect of water 46 [PITH_FULL_IMAGE:figures/full_fig_p046_15.png]
Figure 16
Figure 16. Figure 16: Spatial inference of YOLOv11-m (Train 10) model on Bruche River (South [PITH_FULL_IMAGE:figures/full_fig_p048_16.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

85 extracted references · 10 linked inside Pith

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  2. [2]

    , author Beucler, T

    author Aarnink, J. , author Beucler, T. , author Vuaridel, M. , author Ruiz-Villanueva, V. , year 2025 . title Automatic detection of floating instream large wood in videos using deep learning . journal Earth Surface Dynamics volume 13 , pages 167--189

    2025

  3. [3]

    , author Ologbonjaye, K.I

    author Alabi, O.A. , author Ologbonjaye, K.I. , author Awosolu, O. , author Alalade, O.E. , year 2019 . title Public and environmental health effects of plastic wastes disposal: a review . journal J Toxicol Risk Assess volume 5 , pages 1--13

    2019

  4. [4]

    , author Hussain, M

    author Alif, M.A.R. , author Hussain, M. , year 2024 . title Yolov1 to yolov10: A comprehensive review of yolo variants and their application in the agricultural domain . journal arXiv preprint arXiv:2406.10139

    arXiv 2024

  5. [5]

    , author Sadaoui, S

    author Anowar, F. , author Sadaoui, S. , author Selim, B. , year 2021 . title Conceptual and empirical comparison of dimensionality reduction algorithms (pca, kpca, lda, mds, svd, lle, isomap, le, ica, t-sne). comput. sci. rev. 40, 100378 (2021)

    2021

  6. [6]

    , author Isgr \`o , F

    author Apicella, A. , author Isgr \`o , F. , author Prevete, R. , year 2024 . title Don't push the button! exploring data leakage risks in machine learning and transfer learning . journal arXiv preprint arXiv:2401.13796

    arXiv 2024

  7. [7]

    , author Awty-Carroll, K

    author Armitage, S. , author Awty-Carroll, K. , author Clewley, D. , author Martinez-Vicente, V. , year 2022 . title Detection and classification of floating plastic litter using a vessel-mounted video camera and deep learning . journal Remote Sensing volume 14 , pages 3425

    2022

  8. [8]

    , author Pandey, S.K

    author Babu, M.A.A. , author Pandey, S.K. , author Durisic, D. , author Koppisetty, A.C. , author Staron, M. , year 2024 . title Improving image data leakage detection in automotive software . journal arXiv preprint arXiv:2410.23312

    arXiv 2024

Show all 85 references
  1. [9]

    , year 1998

    author Birchfield, S. , year 1998 . title An introduction to projective geometry (for computer vision) . journal Unpublished note, Stanford university volume 14

    1998

  2. [10]

    , author Staentzel, C

    author Bois, P. , author Staentzel, C. , year 2025 . title Rethinking our collective well-being by sliding from one to common health . journal BioScience , pages biaf013

    2025

  3. [11]

    , author Abdullah, A

    author Bouke, M.A. , author Abdullah, A. , year 2023 . title An empirical study of pattern leakage impact during data preprocessing on machine learning-based intrusion detection models reliability . journal Expert Systems with Applications volume 230 , pages 120715

    2023

  4. [12]

    , author Muhovi c , J

    author Bovcon, B. , author Muhovi c , J. , author Vranac, D. , author Mozeti c , D. , author Per s , J. , author Kristan, M. , year 2021 . title Mods—a usv-oriented object detection and obstacle segmentation benchmark . journal IEEE Transactions on Intelligent Transportation S...

    2021

  5. [13]

    , author Kazmierski, M.R

    author Brown, C.F. , author Kazmierski, M.R. , author Pasquarella, V.J. , author Rucklidge, W.J. , author Samsikova, M. , author Zhang, C. , author Shelhamer, E. , author Lahera, E. , author Wiles, O. , author Ilyushchenko, S. , et al., year 2025 . title Alphaearth foundations...

    2025 arXiv

  6. [14]

    , author Gonz \'a lez-Fern \'a ndez, D

    author Castro-Jim \'e nez, J. , author Gonz \'a lez-Fern \'a ndez, D. , author Fornier, M. , author Schmidt, N. , author Semp \'e r \'e , R. , year 2019 . title Macro-litter in surface waters from the rhone river: Plastic pollution and loading to the nw mediterranean sea . jou...

    2019

  7. [15]

    , author Wu, Y.D

    author Chang, Y.J. , author Wu, Y.D. , author Liao, Y.L. , author Wang, C.H. , author Wu, Y.C. , year 2024 . title Modified yolo network model for metaphase cell detection in antinuclear antibody images . journal Engineering Applications of Artificial Intelligence volume 127 ,...

    2024

  8. [16]

    , author Grimmer, G

    author Chardon, V. , author Grimmer, G. , author Wenger, R. , year 2025 . title A simple and low-cost method using active radio frequency identification to survey the mobility of plastic debris in rivers . journal Science of the Total Environment volume 999 , pages 180336

    2025

  9. [17]

    , author Egger, M

    author De Vries, R. , author Egger, M. , author Mani, T. , author Lebreton, L. , year 2021 . title Quantifying floating plastic debris at sea using vessel-based optical data and artificial intelligence . journal Remote Sensing volume 13 , pages 3401

    2021

  10. [18]

    , year 2016

    author Do Carmo, M.P. , year 2016 . title Differential geometry of curves and surfaces: revised and updated second edition . publisher Courier Dover Publications

    2016

  11. [19]

    , author Kamarudin, L

    author Donal, O. , author Kamarudin, L. , author Zakaria, A. , author Azmi, N. , year 2023 . title Automated detection of floating wood logs on river surfaces using yolov5 model for flood warning system , in: booktitle Journal of Physics: Conference Series , organization IOP P...

    2023

  12. [20]

    , author Mendes, H.A

    author Figueiredo, R.B. , author Mendes, H.A. , year 2024 . title Analyzing information leakage on video object detection datasets by splitting images into clusters with high spatiotemporal correlation . journal IEEE Access

    2024

  13. [21]

    , author He, Y

    author Gao, L. , author He, Y. , author Sun, X. , author Jia, X. , author Zhang, B. , year 2019 . title Incorporating negative sample training for ship detection based on deep learning . journal Sensors volume 19 , pages 684

    2019

  14. [22]

    , author Barbe, V

    author Ghiglione, J.F. , author Barbe, V. , author Bruzaud, S. , author Burgaud, G. , author Cachot, J. , author Eyheraguibel, B. , author Lartaud, F. , author Ludwig, W. , author Meistertzheim, A.L. , author Paul-Pont, I. , et al., year 2023 . title Mission tara microplastics...

    2023

  15. [23]

    , author Bellinger, C

    author Ghosh, K. , author Bellinger, C. , author Corizzo, R. , author Branco, P. , author Krawczyk, B. , author Japkowicz, N. , year 2024 . title The class imbalance problem in deep learning . journal Machine Learning volume 113 , pages 4845--4901

    2024

  16. [24]

    , author Donahue, J

    author Girshick, R. , author Donahue, J. , author Darrell, T. , author Malik, J. , year 2014 . title Rich feature hierarchies for accurate object detection and semantic segmentation , in: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition ,...

    2014

  17. [25]

    , author Scandolo, L

    author G \'o mez, \`A .S. , author Scandolo, L. , author Eisemann, E. , year 2022 . title A learning approach for river debris detection . journal International Journal of Applied Earth Observation and Geoinformation volume 107 , pages 102682

    2022

  18. [26]

    , author Liu, G

    author Gong, Y. , author Liu, G. , author Xue, Y. , author Li, R. , author Meng, L. , year 2023 . title A survey on dataset quality in machine learning . journal Information and Software Technology volume 162 , pages 107268

    2023

  19. [27]

    , author C \'o zar, A

    author Gonz \'a lez-Fern \'a ndez, D. , author C \'o zar, A. , author Hanke, G. , author Viejo, J. , author Morales-Caselles, C. , author Bakiu, R. , author Barcel \'o , D. , author Bessa, F. , author Bruge, A. , author Cabrera, M. , et al., year 2021 . title Floating macrolit...

    2021

  20. [28]

    , author Chen, T

    author Gui, J. , author Chen, T. , author Zhang, J. , author Cao, Q. , author Sun, Z. , author Luo, H. , author Tao, D. , year 2024 . title A survey on self-supervised learning: Algorithms, applications, and future trends . journal IEEE Transactions on Pattern Analysis and Mac...

    2024

  21. [29]

    , author Galgani, F

    author Hanke, G. , author Galgani, F. , author Werner, S. , author Oosterbaan, L. , author Nilsson, P. , author Fleet, D. , author Kinsey, S. , author Thompson, R. , author Van Franeker, J.A. , author Vlachogianni, T. , et al., year 2013 . title Guidance on monitoring of marin...

    2013

  22. [30]

    , author Braaten, H.F.V

    author Hurley, R. , author Braaten, H.F.V. , author Nizzetto, L. , author Steindal, E.H. , author Lin, Y. , author Clayer, F. , author van Emmerik, T. , author Buenaventura, N.T. , author Eidsvoll, D.P. , author kelsrud, A. , et al., year 2023 . title Measuring riverine macrop...

    2023

  23. [31]

    , author Babu, A.R

    author Jaiswal, A. , author Babu, A.R. , author Zadeh, M.Z. , author Banerjee, D. , author Makedon, F. , year 2020 . title A survey on contrastive self-supervised learning . journal Technologies volume 9 , pages 2

    2020

  24. [32]

    , author Herrault, P.A

    author Jautzy, T. , author Herrault, P.A. , author Chardon, V. , author Schmitt, L. , author Rixhon, G. , year 2020 . title Measuring river planform changes from remotely sensed data--a monte carlo approach to assessing the impact of spatially variable error . journal Earth Su...

    2020

  25. [33]

    , author Kapelan, Z

    author Jia, T. , author Kapelan, Z. , author de Vries, R. , author Vriend, P. , author Peereboom, E.C. , author Okkerman, I. , author Taormina, R. , year 2023 . title Deep learning for detecting macroplastic litter in water bodies: A review . journal Water Research volume 231 ...

    2023

  26. [34]

    , author Saurette, D.D

    author John, K. , author Saurette, D.D. , author Heung, B. , year 2025 . title The problematic case of data leakage: A case for leave-profile-out cross-validation in 3-dimensional digital soil mapping . journal Geoderma volume 455 , pages 3

    2025

  27. [35]

    , author Khoshgoftaar, T.M

    author Johnson, J.M. , author Khoshgoftaar, T.M. , year 2019 . title Survey on deep learning with class imbalance . journal Journal of big data volume 6 , pages 1--54

    2019

  28. [36]

    , author Yoshida, T

    author Kataoka, T. , author Yoshida, T. , author Yamamoto, N. , year 2024 . title Instance segmentation models for detecting floating macroplastic debris from river surface images . journal Frontiers in Earth Science volume 12 , pages 1427132

    2024

  29. [37]

    , author Hussain, M

    author Khanam, R. , author Hussain, M. , year 2024 . title Yolov11: An overview of the key architectural enhancements . journal arXiv preprint arXiv:2410.17725

    2024 arXiv

  30. [38]

    , author Manna, C

    author Kumar, R. , author Manna, C. , author Padha, S. , author Verma, A. , author Sharma, P. , author Dhar, A. , author Ghosh, A. , author Bhattacharya, P. , year 2022 . title Micro (nano) plastics pollution and human health: How plastics can induce carcinogenesis to humans? ...

    2022

  31. [39]

    , author Kyriakides, I

    author Kylili, K. , author Kyriakides, I. , author Artusi, A. , author Hadjistassou, C. , year 2019 . title Identifying floating plastic marine debris using a deep learning approach . journal Environmental Science and Pollution Research volume 26 , pages 17091--17099

    2019

  32. [40]

    , author Sanchez, R

    author Landebrit, L. , author Sanchez, R. , author Soccalingame, L. , author Palazot, M. , author Kedzierski, M. , author Bruzeau, S. , author Albignac, M. , author Ludwig, W. , author Ghiglione, J.F. , author Ter Halle, A. , year 2024 . title Small microplastics have much hig...

    2024

  33. [41]

    , author Van Der Zwet, J

    author Lebreton, L.C. , author Van Der Zwet, J. , author Damsteeg, J.W. , author Slat, B. , author Andrady, A. , author Reisser, J. , year 2017 . title River plastic emissions to the world’s oceans . journal Nature communications volume 8 , pages 15611

    2017

  34. [42]

    , author Bengio, Y

    author LeCun, Y. , author Bengio, Y. , author Hinton, G. , year 2015 . title Deep learning . journal Nature volume 521 , pages 436--444

    2015

  35. [43]

    , author Maire, M

    author Lin, T.Y. , author Maire, M. , author Belongie, S. , author Hays, J. , author Perona, P. , author Ramanan, D. , author Doll \'a r, P. , author Zitnick, C.L. , year 2014 . title Microsoft coco: Common objects in context , in: booktitle Computer vision--ECCV 2014: 13th Eu...

    2014

  36. [44]

    , author Zielonka, A

    author Liro, M. , author Zielonka, A. , author van Emmerik, T.H. , year 2023 . title Macroplastic fragmentation in rivers . journal Environment International volume 180 , pages 108186

    2023

  37. [45]

    , author Han, W

    author Luo, W. , author Han, W. , author Fu, P. , author Wang, H. , author Zhao, Y. , author Liu, K. , author Liu, Y. , author Zhao, Z. , author Zhu, M. , author Xu, R. , et al., year 2022 . title A water surface contaminants monitoring method based on airborne depth reasoning...

    2022

  38. [46]

    , author Miyazaki, H

    author Maharjan, N. , author Miyazaki, H. , author Pati, B.M. , author Dailey, M.N. , author Shrestha, S. , author Nakamura, T. , year 2022 . title Detection of river plastic using uav sensor data and deep learning . journal Remote Sensing volume 14 , pages 3049

    2022

  39. [47]

    , year 2019

    author Mahmoud, H.F. , year 2019 . title Parametric versus semi and nonparametric regression models . journal arXiv preprint arXiv:1906.10221

    2019 arXiv

  40. [48]

    , author Fuentes, O

    author Malagon-Borja, L. , author Fuentes, O. , year 2009 . title Object detection using image reconstruction with pca . journal Image and Vision Computing volume 27 , pages 2--9

    2009

  41. [49]

    , author Hensman, P

    author Masko, D. , author Hensman, P. , year 2015 . title The impact of imbalanced training data for convolutional neural networks

    2015

  42. [50]

    , author Triggs, B

    author Mohr, R. , author Triggs, B. , year 1996 . title Projective geometry for image analysis , in: booktitle XVIIIth International Symposium on Photogrammetry & Remote Sensing (ISPRS'96)

    1996

  43. [51]

    , author Jaskowiak, P.A

    author Moulavi, D. , author Jaskowiak, P.A. , author Campello, R.J. , author Zimek, A. , author Sander, J. , year 2014 . title Density-based clustering validation , in: booktitle Proceedings of the 2014 SIAM international conference on data mining , organization SIAM . pp. pag...

    2014

  44. [52]

    , author Bosch, J

    author Munappy, A.R. , author Bosch, J. , author Olsson, H.H. , author Arpteg, A. , author Brinne, B. , year 2022 . title Data management for production quality deep learning models: Challenges and solutions . journal Journal of Systems and Software volume 191 , pages 111359

    2022

  45. [53]

    , author Sobral, P

    author Neves, D. , author Sobral, P. , author Ferreira, J.L. , author Pereira, T. , year 2015 . title Ingestion of microplastics by commercial fish off the portuguese coast . journal Marine pollution bulletin volume 101 , pages 119--126

    2015

  46. [54]

    , author Miyamoto, H

    author Nunkhaw, M. , author Miyamoto, H. , year 2024 . title An image analysis of river-floating waste materials by using deep learning techniques . journal Water volume 16 , pages 1373

    2024

  47. [55]

    , author Machado, A.R

    author Oliveira, F.H.M. , author Machado, A.R. , author Andrade, A.O. , year 2018 . title On the use of t-distributed stochastic neighbor embedding for data visualization and classification of individuals with parkinson’s disease . journal Computational and mathematical method...

    2018

  48. [56]

    , author Cau, A

    author Palmas, F. , author Cau, A. , author Podda, C. , author Musu, A. , author Serra, M. , author Pusceddu, A. , author Sabatini, A. , year 2022 . title Rivers of waste: Anthropogenic litter in intermittent sardinian rivers, italy (central mediterranean) . journal Environmen...

    2022

  49. [57]

    , author da Costa, J.P

    author Prata, J.C. , author da Costa, J.P. , author Lopes, I. , author Andrady, A.L. , author Duarte, A.C. , author Rocha-Santos, T. , year 2021 . title A one health perspective of the impacts of microplastics on animal, human and environmental health . journal Science of the ...

    2021

  50. [58]

    , author Rivera, A.J

    author Pulgar, F.J. , author Rivera, A.J. , author Charte, F. , author del Jesus, M.J. , year 2017 . title On the impact of imbalanced data in convolutional neural networks performance , in: booktitle Hybrid Artificial Intelligent Systems: 12th International Conference, HAIS 2...

    2017

  51. [59]

    , author Macaya-Caquilp \'a n, V

    author Rech, S. , author Macaya-Caquilp \'a n, V. , author Pantoja, J. , author Rivadeneira, M. , author Madariaga, D.J. , author Thiel, M. , year 2014 . title Rivers as a source of marine litter--a study from the se pacific . journal Marine pollution bulletin volume 82 , pages 66--75

    2014

  52. [60]

    , author Divvala, S

    author Redmon, J. , author Divvala, S. , author Girshick, R. , author Farhadi, A. , year 2016 . title You only look once: Unified, real-time object detection , in: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition , pp. pages 779--788

    2016

  53. [61]

    title Réserve naturelle nationale du massif forestier de la robertsau et de la wantzenau

    author Réserves Naturelles de France , year 2023 . title Réserve naturelle nationale du massif forestier de la robertsau et de la wantzenau . https://reserves-naturelles.org/reserves/massif-forestier-de-la-robertsau-et-de-la-wantzenau. note accessed April 2025

    2023

  54. [62]

    , author Nicolaisen-Sobesky, E

    author Sasse, L. , author Nicolaisen-Sobesky, E. , author Dukart, J. , author Eickhoff, S. , author G \"o tz, M. , author Hamdan, S. , author Komeyer, V. , author Kulkarni, A. , author Lahnakoski, J. , author Love, B. , et al., year 2025 . title Overview of leakage scenarios i...

    2025

  55. [63]

    , author Dau, K

    author Sch \"o neich-Argent, R.I. , author Dau, K. , author Freund, H. , year 2020 . title Wasting the north sea?--a field-based assessment of anthropogenic macrolitter loads and emission rates of three german tributaries . journal Environmental Pollution volume 263 , pages 114367

    2020

  56. [64]

    , author Meisinger, L

    author Schwindt, S. , author Meisinger, L. , author Negreiros, B. , author Schneider, T. , author Nowak, W. , year 2024 . title Transfer learning achieves high recall for object classification in fluvial environments with limited data . journal Geomorphology volume 455 , pages 109185

    2024

  57. [65]

    , author Li, G

    author Seide, F. , author Li, G. , author Chen, X. , author Yu, D. , year 2011 . title Feature engineering in context-dependent deep neural networks for conversational speech transcription , in: booktitle 2011 IEEE Workshop on Automatic Speech Recognition & Understanding , org...

    2011

  58. [66]

    , author Vo, H.V

    author Sim \'e oni, O. , author Vo, H.V. , author Seitzer, M. , author Baldassarre, F. , author Oquab, M. , author Jose, C. , author Khalidov, V. , author Szafraniec, M. , author Yi, S. , author Ramamonjisoa, M. , et al., year 2025 . title Dinov3 . journal arXiv preprint arXiv...

    2025 arXiv

  59. [67]

    , author Ockelford, A

    author Skalska, K. , author Ockelford, A. , author Ebdon, J.E. , author Cundy, A.B. , year 2020 . title Riverine microplastics: Behaviour, spatio-temporal variability, and recommendations for standardised sampling and monitoring . journal Journal of Water Process Engineering v...

    2020

  60. [68]

    , author Parwati, E

    author Suteja, Y. , author Parwati, E. , author Budhiman, S. , author Radjawane, I.M. , author Hartuti, M. , author Afgatiani, P.M. , author Ulfa, A. , author Dewi, E.K. , author Purwiyanto, A.I.S. , et al., year 2025 . title Spatial patterns and types of marine anthropogenic ...

    2025

  61. [69]

    , author Eklund, A

    author Tampu, I.E. , author Eklund, A. , author Haj-Hosseini, N. , year 2022 . title Inflation of test accuracy due to data leakage in deep learning-based classification of oct images . journal Scientific Data volume 9 , pages 580

    2022

  62. [70]

    , author Zheng, Y

    author Tao, H. , author Zheng, Y. , author Wang, Y. , author Qiu, J. , author Stojanovic, V. , year 2024 . title Enhanced feature extraction yolo industrial small object detection algorithm based on receptive-field attention and multi-scale features . journal Measurement Scien...

    2024

  63. [71]

    , author Royer, S.J

    author Tata, G. , author Royer, S.J. , author Poirion, O. , author Lowe, J. , year 2021 . title A robotic approach towards quantifying epipelagic bound plastic using deep visual models . journal arXiv preprint arXiv:2105.01882

    2021 arXiv

  64. [72]

    , author Gasperi, J

    author Tramoy, R. , author Gasperi, J. , author Dris, R. , author Colasse, L. , author Fisson, C. , author Sananes, S. , author Rocher, V. , author Tassin, B. , year 2019 . title Assessment of the plastic inputs from the seine basin to the sea using statistical and field appro...

    2019

  65. [73]

    title LabelImg [computer software]

    author Tzutalin , year 2015 . title LabelImg [computer software] . howpublished https://github.com/tzutalin/labelImg . note GitHub repository

    2015

  66. [74]

    title Tasks – oriented object detection (obb)

    author Ultralytics , year 2024 . title Tasks – oriented object detection (obb) . howpublished https://docs.ultralytics.com/fr/tasks/obb/ . note Accessed on June 23, 2025

    2024

  67. [75]

    , year 2019

    author Ultralytics Inc. , year 2019 . title Model training with ultralytics yolo . howpublished https://docs.ultralytics.com/modes/train/ . note Accessed April 11, 2019

    2019

  68. [76]

    , author Schwarz, A

    author Van Emmerik, T. , author Schwarz, A. , year 2020 . title Plastic debris in rivers . journal Wiley Interdisciplinary Reviews: Water volume 7 , pages e1398

    2020

  69. [77]

    , author van Oeveren, K

    author Van Lieshout, C. , author van Oeveren, K. , author van Emmerik, T. , author Postma, E. , year 2020 . title Automated river plastic monitoring using deep learning and cameras . journal Earth and space science volume 7 , pages e2019EA000960

    2020

  70. [78]

    , author Zhao, H

    author Wang, J. , author Zhao, H. , year 2024 . title Improved yolov8 algorithm for water surface object detection . journal Sensors volume 24 , pages 5059

    2024

  71. [79]

    , author Harris, R

    author Waring, R.H. , author Harris, R. , author Mitchell, S. , year 2018 . title Plastic contamination of the food chain: A threat to human health? journal Maturitas volume 115 , pages 64--68

    2018

  72. [80]

    , author Zipperling, D

    author Wegmann, M. , author Zipperling, D. , author Hillenbrand, J. , author Fleischer, J. , year 2021 . title A review of systematic selection of clustering algorithms and their evaluation . journal arXiv preprint arXiv:2106.12792

    2021 arXiv

  73. [81]

    , author Ekenel, H.K

    author Yal c n, S. , author Ekenel, H.K. , year 2024 . title Impact of surface reflections in maritime obstacle detection . journal arXiv preprint arXiv:2410.08713

    2024 arXiv

  74. [82]

    , author Zhang, Y

    author Ye, J. , author Zhang, Y. , author Li, P. , author Guo, Z. , author Zeng, S. , author Wei, T. , year 2025 . title Real-time dense small object detection model for floating litter detection and removal on water surfaces . journal Marine Pollution Bulletin volume 218 , pa...

    2025

  75. [83]

    , year 2019

    author Ying, X. , year 2019 . title An overview of overfitting and its solutions , in: booktitle Journal of physics: Conference series , organization IOP Publishing . p. pages 022022

    2019

  76. [84]

    , author Min, C

    author Zhang, X. , author Min, C. , author Luo, J. , author Li, Z. , year 2023 . title Yolov5-ff: detecting floating objects on the surface of fresh water environments . journal Applied Sciences volume 13 , pages 7367

    2023

  77. [85]

    2015, , 579, A101

    Aladro, R., Martín, S., Riquelme, D., et al. 2015, , 579, A101

    2015

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.