Pith. sign in

REVIEW 3 major objections 6 minor 41 references

KAMERA: Enhancing Aerial Surveys of Ice-associated Seals in Arctic Environments

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 paper introduces KAMERA, a synchronized multi-camera, multi-spectral imaging and real-time detection system, and claims it cuts Arctic seal-survey dataset processing time by up to 80 percent relative to earlier semi-automated methods.

desk verdict A real, open-source multi-spectral aerial survey system with a useful late-fusion detection pipeline, but the headline 80% speedup is a confounded cross-survey comparison and the model validation is thinner than the abstract implies. read the letter →

arxiv 2509.19129 v2 pith:F7UBW2GH submitted 2025-09-23 cs.CV

Adam Romlein , Benjamin X. Hou , Yuval Boss , Cynthia L. Christman , Stacie Koslovsky , Erin E. Moreland , Jason Parham , Anthony Hoogs This is my paper
classification cs.CV
keywords aerialsurveysice-associatedsealsmulti-spectralimagingreal-timeobjectdetectioncamerasynchronizationthermalinfraredpolarbearopen-sourcesoftware
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

KAMERA is a complete hardware-and-software system for airborne wildlife surveys that captures RGB, thermal, and ultraviolet imagery from nine synchronized cameras, runs object detection in real time on the aircraft, and maps every image and detection onto a ground plane. The paper's central claim is that this integrated pipeline reduces the time from data collection to survey results by up to 80 percent relative to earlier large surveys, which relied on loosely coupled, asynchronous capture and semi-automated hotspot analysis. If true, million-image surveys covering roughly 20,000 km of transect lines could be processed in weeks instead of months or years, which matters because ice seal populations are legally protected and must be monitored as their sea-ice habitat changes. The system's software, models, and hardware designs are released open-source, so other aerial survey efforts could adopt the approach.

What carries the argument

The load-bearing mechanism is the late-fusion detection pipeline enabled by hardware synchronization and multi-camera calibration. First, a fast infrared hot-spot detector, a single-shot convolutional object detector, finds candidate animals in thermal imagery. Then, using the calibrated rigid transforms between cameras and the inertial navigation system, the system crops the corresponding full-resolution region from the paired color image and passes that chip to a species-specific color detector for classification. This two-stage design lets the low-resolution thermal spectrum do the cheap search while the high-resolution color spectrum does the fine identification; the same geometry projec

What would settle it

Take a dataset from the 2016 Chukchi survey and run it through the current KAMERA pipeline, and take 2021 Beaufort imagery and process it with the old semi-automated hotspot method. If the time reduction shrinks or disappears in either direction, the 80% reduction cannot be attributed to the new system alone.

Watch

Extended reading notes

Core claim

The paper introduces KAMERA, a synchronized multi-camera, multi-spectral imaging system for detecting ice-associated seals and polar bears in Arctic aerial surveys. All nine cameras (three RGB, three thermal, three ultraviolet) are triggered by a single external time pulse tied to GPS time, so each image sample is aligned in time and space; a calibration process estimates each camera's position relative to the inertial navigation system, allowing any detection to be projected onto a planar model of the world. Onboard GPUs run a two-stage detection pipeline: a thermal hot-spot detector finds candidate animals in infrared imagery, and the calibrated geometry crops the corresponding high-resolu

Load-bearing premise

The 80% speedup claim rests on comparing the 2021 Beaufort survey, which used KAMERA and a two-stage IR-to-RGB pipeline, with the 2016 Chukchi survey, which used different cameras and a semi-automated hotspot workflow; the assumption that those differences in hardware, area, and procedure do not materially affect the timing is untested.

Editorial extensions

If this is right

  • If the 80% time reduction holds, large Arctic seal surveys—roughly a million samples each—can go from data collection to final results in weeks, enabling faster management decisions for threatened and protected species.
  • Real-time detection means blank imagery can be discarded before storage, cutting data volume and downstream review for datasets where fewer than 1% of images contain animals.
  • Geolocated detections and per-camera footprints make surveyed-area calculations and post-flight quality checks routine, reducing reliance on manual processing.
  • Because the software stack, models, and hardware schematics are open-source, other wildlife survey programs can adopt synchronized multi-spectral capture without rebuilding proprietary systems.
  • The IR-triggered color classification approach may carry over to any survey where one spectrum offers reliable detection and another offers higher-resolution species identification.

Reading between the lines

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

  • The 80% figure compares two different surveys with different cameras, geographic regions, and analysis procedures, so the speedup is an upper-bound estimate of what the system alone contributes; a controlled re-analysis of the same datasets would be needed to separate system effects from workflow and hardware changes.
  • The two-stage detection strategy suggests a general cost-geometry principle: spend cheap detection compute in the low-resolution spectrum where targets contrast strongly, and spend expensive classification compute only on small, calibrated crops. That principle could transfer to other remote-sensing tasks, such as searching thermal imagery for animals or objects and then verifying in high-resoluti
  • UV imagery is currently collected but not yet used in trained models; the paper notes white-coat seal pups and polar bears absorb UV, so a UV-based detector is a natural next step that could extend the system's species coverage.
  • The paper reports that the 2025 IR hot-spot model generalized poorly to new thermal cameras, so real-world adoption will likely depend on per-survey model recalibration or domain adaptation to new sensor hardware.
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. KAMERA is an open-source multi-camera, multi-spectral aerial imaging system for detecting ice-associated seals and polar bears in Arctic surveys. The paper describes the hardware (nine RGB/IR/UV cameras, INS, custom trigger box), the ROS-based software stack and GUI, COLMAP-based calibration, YOLOv3-based IR hot-spot and IR-to-RGB late-fusion detection models, and mapping/data products. It reports deployment in a 2021 Beaufort Sea survey with roughly 900,000 image triplets and a 2025 Bering/Chukchi/Beaufort survey with 1.5 million samples, claims up to an 80% reduction in dataset processing time relative to the 2016 Chukchi survey, and reports held-out detection metrics in Table 4.

Significance. The system's open release, full hardware schematics, published models, and real operational use are concrete and valuable contributions. The synchronized multi-spectral capture, rich metadata, and world-plane mapping are credible improvements over the proprietary, loosely coupled systems used in earlier surveys, and the 2025 deployment at 1.5 million samples demonstrates practical engineering maturity. If the speedup claim were supported by controlled measurements, the result would be significant for applied aerial wildlife surveys. As written, however, the headline speedup is not experimentally isolated from the simultaneous introduction of the deep-learning pipeline, and the reported model validation is too limited to carry the cross-survey claim.

major comments (3)
  1. [Section 2 / Table 1 / Abstract] The headline 'up to an 80% reduction in dataset processing time' is a cross-survey comparison, not a controlled evaluation. The 2021 Beaufort survey used KAMERA plus a two-stage YOLOv3 IR/RGB pipeline, while the 2016 Chukchi survey used semi-automated hot-spot detection with different cameras, different region, and different analysis procedures. Section 2 attributes the reduction to 'the integration of AI methods with the improved data produced by KAMERA,' which conflates the new synchronized acquisition system with the new AI pipeline. No timing breakdown per stage, no ablation, and no uncertainty or statistical analysis are provided. The claim should be removed or substantially weakened unless the comparison is controlled or component contributions are separated.
  2. [Section 4 / Table 4] The validation is reported on a single held-out 10% of human-labeled data, with no description of how the split was stratified, no confidence intervals, and no per-class error analysis. Table 4 reports per-model GPU inference FPS on a GTX 1080Ti, but no end-to-end processing-time measurements (acquisition, triggering, crop generation, secondary classification, manual review) under survey conditions. The 'real-time' assertion therefore applies to individual models, not to the claimed pipeline speedup. The paper's own Section 4 states that the IR model's precision dropped dramatically on 2025 data, so the 2021-era validation numbers are not a reliable proxy for operational performance across surveys; this strengthens the concern about relying on historical survey comparisons.
  3. [Section 3.3] The final models were trained with pseudo-labels formed by ensembling early models, while a separate 10% was held out for final evaluation. The text does not state whether the held-out set or its labels were used at any point in pseudo-label generation, model selection, or threshold tuning. If they were, Table 4 would be circular. Even if not, the missing labels in thermal imagery and the use of pseudo-labels make the validation metrics sensitive to label noise; this should be stated explicitly and, if possible, quantified.
minor comments (6)
  1. [Table 4] The polar bear model row reports input dimensions '412x412x3'; this is inconsistent with '416x416x3' in Table 3 and is presumably a typo.
  2. [Section 3.3] The notation 'P5/32-3A', 'P4/16-5A', and 'P3/8-1A' in Table 3 is not defined in the text. Please define P and A explicitly so the architecture table is self-contained.
  3. [Table 1] The 2025 row states '1.5 million samples, reduced' but does not explain what 'reduced' refers to. Clarify whether this is the number of archived samples after on-board filtering or the raw collected sample count.
  4. [Section 3.1] Synchronization is a central claimed capability, but no measured trigger-to-exposure latency or jitter is reported. A short quantitative characterization of the synchronization accuracy would strengthen the hardware description.
  5. [Section 3.2] The manual alignment step is described as ensuring that 'relative alignment is exact.' Please provide a tolerance or residual error estimate instead of 'exact,' since the later Figure 8 examples show visible calibration misalignment.
  6. [Figure 8] The examples of double classification and slight calibration misalignment are noted qualitatively but not quantified. It would be helpful to state how such cases are handled in the precision/recall metrics in Table 4.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the 80% speedup is a confounded empirical comparison, not a quantity forced by construction, and held-out validation prevents the pseudo-label loop from being circular.

full rationale

KAMERA is a systems/engineering description rather than a derivation, so the main circularity patterns do not apply. The headline claim of 'up to an 80% reduction in dataset processing time over previous methods' (Abstract) is supported in Section 2/Table 1 by comparing the 2021 Beaufort survey (KAMERA + YOLOv3 IR/RGB trigger model) with the 2016 Chukchi survey (semi-automated hot spot detection). The paper explicitly says: 'The integration of AI methods with the improved data produced by KAMERA enabled an 80% reduction in time-to-results compared to the Chukchi Sea surveys in 2016' (Section 2). This conflates two changes—the KAMERA acquisition/synchronization system and the new two-stage deep-learning pipeline—and is a confounded attribution/correctness risk, not a circularity: the 80% figure is an empirical, retrospective survey comparison, not a parameter fitted to those surveys and then renamed as a prediction, nor is it equal to an input by construction. The pseudo-label procedure (Section 3.3: 'multiple early models were ensembled and the resulting pseudolabels were used in training of the final models') is a self-training loop in model development, but Table 4's validation results are computed on a deliberately held-out 10% of human-labeled data, so the reported metrics are not predetermined by the training set. The paper's own negative result—'These results show weak generalization of the IR YOLOv3 hot spot model' (Section 4)—is an honest external limitation, not a circular step. Self-citations (KAMERA GitHub [16], VIAME [10], pipeline release [4]) are for reproducibility and open-source availability and are not load-bearing for the central claim. No load-bearing step reduces to its inputs by construction or to a self-citation chain.

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

The central performance and detection claims rest primarily on domain assumptions about thermal signatures and cross-survey comparability, plus model hyperparameters tuned on the data. No new physical entities are introduced. The calibration process depends on COLMAP and SIFT succeeding, which the paper itself identifies as fragile. The free parameters listed are tuned or calibrated quantities that directly affect the reported detection metrics and the claimed processing speedup.

free parameters (5)
  • YOLOv3 anchor boxes = not specified
    Anchor boxes were chosen to fit the object size distribution for each model (Section 3.3).
  • lambda_noobj (IR model) = reduced from default; table caption notes lambda_noobj = .5 but text says it was reduced
    The regularizer penalizing no-object boxes was tuned to improve validation accuracy (Section 3.3).
  • Focal loss parameters (alpha, gamma) = not specified
    Focal loss was used to balance seal classes for the RGB model; hyperparameters were presumably tuned (Section 3.3).
  • Detection confidence threshold = not specified
    Thresholds for accepting detections affect precision and recall but are not reported in the paper.
  • COLMAP matching parameters = not specified
    COLMAP parameters were tuned manually if the reconstruction failed (Section 3.2).
assumptions (4)
  • domain assumption Thermal hot spots in IR imagery are reliable indicators of ice seals against sea ice.
    The entire real-time archiving strategy depends on this; the paper reports that precision dropped dramatically in the 2025 survey, weakening this axiom (Section 4).
  • domain assumption Structure-from-motion with SIFT features produces accurate camera calibrations across RGB/UV and IR separately, and these calibrations remain valid throughout a survey.
    Calibration in Section 3.2 relies on COLMAP; the paper acknowledges failure cases and calibration drift, and shows an example of misalignment in Figure 8.
  • ad hoc to paper The 2021 Beaufort and 2016 Chukchi surveys are comparable in difficulty and scale, so the 80% time reduction can be attributed to KAMERA.
    This assumption is necessary for the headline claim but is not tested; the surveys differ in hardware, survey area, and analysis procedures (Section 2, Table 1).
  • domain assumption Held-out validation data is representative of operational survey imagery, despite missing labels and pseudo-label training.
    Model evaluation in Section 4 assumes the test set reflects real conditions; the 2025 precision drop suggests this assumption is fragile.

how reviews work

0 comments
Cite this review

Pith. "Pith review of KAMERA: Enhancing Aerial Surveys of Ice-associated Seals in Arctic Environments." pith.science (2026). https://pith.science/paper/F7UBW2GH

@misc{pith2026250919129,
  author       = {Pith},
  title        = {Pith review of: KAMERA: Enhancing Aerial Surveys of Ice-associated Seals in Arctic Environments},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/F7UBW2GH}},
  note         = {Machine review of arXiv:2509.19129}
}
read the original abstract

We introduce KAMERA: a comprehensive system for multi-camera, multi-spectral synchronization and real-time detection of seals and polar bears. Utilized in aerial surveys for ice-associated seals in the Bering, Chukchi, and Beaufort seas around Alaska, KAMERA provides up to an 80% reduction in dataset processing time over previous methods. Our rigorous calibration and hardware synchronization enable using multiple spectra for object detection. All collected data are annotated with metadata so they can be easily referenced later. All imagery and animal detections from a survey are mapped onto a world plane for accurate surveyed area estimates and quick assessment of survey results. We hope KAMERA will inspire other mapping and detection efforts in the scientific community, with all software, models, and schematics fully open-sourced.

Figures

Figures reproduced from arXiv: 2509.19129 by the authors.

Figure 1
Figure 1. KAMERA’s hardware, software, and area of operations [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Camera mount containing the nine-camera sensor sys [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Analysis of the imagery on a calibration flight over Anchorage, AK using COLMAP. (Left) Example of a successful COLMAP [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: (Top) Overview of the calibration process for the nine [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 6
Figure 6. Figure 6: Visualization of late fusion pipeline. (Left) IR imagery [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 5
Figure 5. Figure 5: Late fusion 2-stage pipeline, utilizing thermal detections [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 7
Figure 7. Figure 7: Example visualized feature grid P3/8 representing the scale of detection for the thermal model architecture. layer was used as shown in [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 8
Figure 8. Figure 8: Example detection results, IR on top, matching RGB on bottom. Purple boxes are classified as bearded seals, yellow boxes are [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: GUI in flight operation. All cameras are shown, with [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

41 extracted references · 8 linked inside Pith

  1. [1]

    Yolo v3- tiny: Object detection and recognition using one stage im- proved model

    Pranav Adarsh, Pratibha Rathi, and Manoj Kumar. Yolo v3- tiny: Object detection and recognition using one stage im- proved model. In2020 6th International Conference on Ad- vanced Computing and Communication Systems (ICACCS), pages 687–694, 2020. 6

    2020

  2. [2]

    The great grevy’s rally: The need, methods, findings, implications and next steps

    Tanya Berger-Wolf, Jon Crall, Jason Holberg, Jason Parham, Chuck Stewart, Belinda Low Mackey, Paula Kahumbu, and Dan Rubenstein. The great grevy’s rally: The need, methods, findings, implications and next steps. InTechnical report. Grevy’s Zebra Trust, 2016. 1

    2016

  3. [3]

    Nasa-noaa feasibility report

    Abhay Borade and Jeffrey Homola. Nasa-noaa feasibility report. Technical report, NASA-NOAA, 2021. 3

    2021

  4. [4]

    Polar ecosystems program inflight kamera detection pipelines.https://doi.org/10.5281/ zenodo.5765673, 2021

    Yuval Boss. Polar ecosystems program inflight kamera detection pipelines.https://doi.org/10.5281/ zenodo.5765673, 2021. Version 2021.03.19; Accessed August 6, 2025. 8

    2021

  5. [5]

    Boveng, Vladimir I

    Peter L. Boveng, Vladimir I. Chernook, Erin E. Moreland, Paul B. Conn, Irina S. Trukhanova, Michael F. Cameron, Cynthia L. Christman, Justin A. Crawford, Dmitry M. Glazov, Lois Harwood, Benjamin X. Hou, Stacie M. Koslovsky, Jessica M. Lindsay, Denis I. Litovka, Josh M. London, Brett T. McClintock, Nikita Platonov, Lori Quak- enbush, Erin L. Richmond, Alex...

    2025

  6. [6]

    Methodologies and preliminary results of the noaa na- tional marine fisheries service aerial survey program for right whales (eubalaena glacialis) in the northeast us, 1998-2006

    Timothy VN Cole, Patricia Gerrior, and Richard L Mer- rick. Methodologies and preliminary results of the noaa na- tional marine fisheries service aerial survey program for right whales (eubalaena glacialis) in the northeast us, 1998-2006. Technical report, NOAA, 2007. 1

    1998

  7. [7]

    Hotspot- ter—patterned species instance recognition

    Jonathan P Crall, Charles V Stewart, Tanya Y Berger-Wolf, Daniel I Rubenstein, and Siva R Sundaresan. Hotspot- ter—patterned species instance recognition. In2013 IEEE workshop on applications of computer vision (WACV), pages 230–237. IEEE, 2013. 1

    2013

  8. [8]

    Adapt: an open-source suas payload for real-time disaster predic- tion and response with ai.arXiv preprint arXiv:2201.10366,

    Daniel Davila, Joseph VanPelt, Alexander Lynch, Adam Romlein, Peter Webley, and Matthew S Brown. Adapt: an open-source suas payload for real-time disaster predic- tion and response with ai.arXiv preprint arXiv:2201.10366,

    arXiv

Show all 41 references
  1. [9]

    Automatic scallop detection in benthic environ- ments

    Matthew Dawkins, Charles Stewart, Scott Gallager, and Am- ber York. Automatic scallop detection in benthic environ- ments. InProceedings of the IEEE Workshop on Applica- tions of Computer Vision, pages 160–167, Clearwater Beach, FL, USA, 2013. IEEE. 1

    2013

  2. [10]

    An open-source platform for underwater im- age and video analytics

    Matthew Dawkins, Linus Sherrill, Keith Fieldhouse, An- thony Hoogs, Benjamin Richards, David Zhang, Laksh- man Prasad, Kresimir Williams, Nathan Lauffenburger, and Gaoang Wang. An open-source platform for underwater im- age and video analytics. InProceedings of the IEEE Winter...

    2017

  3. [11]

    Unmanned aerial vehicles in agri- culture: A survey.Agronomy, 11(2):203, 2021

    Jaime Del Cerro, Christyan Cruz Ulloa, Antonio Barrientos, and Jorge de Le´on Rivas. Unmanned aerial vehicles in agri- culture: A survey.Agronomy, 11(2):203, 2021. 1

    2021

  4. [12]

    Superpoint: Self-supervised interest point detection and description.arXiv preprint arXiv:1712.07629, 2018

    Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabi- novich. Superpoint: Self-supervised interest point detection and description.arXiv preprint arXiv:1712.07629, 2018. 4

    2018 arXiv

  5. [13]

    Improving the precision and accuracy of animal population estimates with aerial image object detection.Methods in Ecology and Evolution, 10(11):1875–1887, 2019

    Jasper AJ Eikelboom, Johan Wind, Eline Van de Ven, Lek- ishon M Kenana, Bradley Schroder, Henrik J De Knegt, Frank van Langevelde, and Herbert HT Prins. Improving the precision and accuracy of animal population estimates with aerial image object detection.Methods in Ecology an...

    2019

  6. [14]

    Noaa uncrewed aircraft systems report for fiscal year 2023

    Ashley Hann, Max Anderson, Laura Dwyer, Justin Blancher, Chesna Cox, Lisa Nakamura, William Mowitt, Allan Knox, Jesse Reich, Melissa Cook, et al. Noaa uncrewed aircraft systems report for fiscal year 2023. Technical report, NOAA,

    2023

  7. [15]

    Vision-based modal survey of civil infrastructure using unmanned aerial vehicles.Journal of Structural Engineering, 145(7):04019062, 2019

    Vedhus Hoskere, Jong-Woong Park, Hyungchul Yoon, and Billie F Spencer Jr. Vision-based modal survey of civil infrastructure using unmanned aerial vehicles.Journal of Structural Engineering, 145(7):04019062, 2019. 1

    2019

  8. [16]

    KAMERA: Knowledge-guided image acqui- sition manager and archiver.https://github.com/ Kitware/kamera/tree/v0.4.0, 2025

    Kitware, Inc. KAMERA: Knowledge-guided image acqui- sition manager and archiver.https://github.com/ Kitware/kamera/tree/v0.4.0, 2025. Version 0.4.0, released July 28, 2025; Accessed July 30, 2025. 8

    2025

  9. [17]

    Feature pyramid networks for object detection.arXiv preprint arXiv:1612.03144, 2017

    Tsung-Yi Lin, Piotr Doll ´ar, Ross Girshick, Kaiming He, Bharath Hariharan, and Serge Belongie. Feature pyramid networks for object detection.arXiv preprint arXiv:1612.03144, 2017. 5, 6

    2017 arXiv

  10. [18]

    Focal loss for dense object detection.arXiv preprint arXiv:1708.02002, 2018

    Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Doll ´ar. Focal loss for dense object detection.arXiv preprint arXiv:1708.02002, 2018. 6

    2018 arXiv

  11. [19]

    Ssd: Single shot multibox detector

    Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, and Alexander C Berg. Ssd: Single shot multibox detector. InEuropean con- ference on computer vision, pages 21–37. Springer, 2016. 6

    2016

  12. [20]

    David G. Lowe. Distinctive image features from scale- invariant keypoints.International Journal of Computer Vi- sion, 60(2):91–110, 2004. 4

    2004

  13. [21]

    Robot operating system 2: Design, architecture, and uses in the wild.Science Robotics, 7(66):eabm6074, 2022

    Steven Macenski, Tully Foote, Brian Gerkey, Chris Lalancette, and William Woodall. Robot operating system 2: Design, architecture, and uses in the wild.Science Robotics, 7(66):eabm6074, 2022. 4 9

    2022

  14. [22]

    Noaa’s use of remote sensing technology and the coastal mapping program

    Brian Madore, Gretchen Imahori, Jamie Kum, Stephen White, and Aleah Worthem. Noaa’s use of remote sensing technology and the coastal mapping program. InOCEANS 2018 mts/ieee charleston, pages 1–7. IEEE, 2018. 1

    2018

  15. [23]

    Docker: lightweight linux containers for con- sistent development and deployment.Linux journal, 2014 (239):2, 2014

    Dirk Merkel. Docker: lightweight linux containers for con- sistent development and deployment.Linux journal, 2014 (239):2, 2014. 4

    2014

  16. [24]

    Erin Moreland, Michael Cameron, and Peter Boveng. Bering okhotsk seal surveys (boss), joint us-russian aerial sur- veys for ice-associated seals, 2012-13.Alaska Fisheries Science Center Quarterly Report (July-August-September 2013), pages 1–6, 2013. 2

    2012

  17. [25]

    Individual identi- fication of the endangered wyoming toad anaxyrus baxteri and implications for monitoring species recovery.Journal of Herpetology, 50(1):44–49, 2016

    Thomas A Morrison, Douglas Keinath, Wendy Estes-Zumpf, Jonathan P Crall, and Charles V Stewart. Individual identi- fication of the endangered wyoming toad anaxyrus baxteri and implications for monitoring species recovery.Journal of Herpetology, 50(1):44–49, 2016. 1

    2016

  18. [26]

    An animal detection pipeline for identification

    Jason Parham, Charles Stewart, Jonathan P Crall, Daniel Rubenstein, Jason Holmberg, and Tanya Berger-Wolf. An animal detection pipeline for identification. InProceedings of the IEEE Winter Conference on Applications of Computer Vision, pages 1075–1083, Lake Tahoe, NV , 2018. IEEE

    2018

  19. [27]

    Animal popula- tion censusing at scale with citizen science and photographic identification

    Jason Remington Parham, Jonathan Crall, Charles Stewart, Tanya Berger-Wolf, and Daniel Rubenstein. Animal popula- tion censusing at scale with citizen science and photographic identification. In2017 AAAI Spring Symposium Series, 2017. 1

    2017

  20. [28]

    YOLOv3: An Incremental Improvement.arXiv preprint arXiv:1804.02767, 2018

    Joseph Redmon and Ali Farhadi. YOLOv3: An Incremental Improvement.arXiv preprint arXiv:1804.02767, 2018. 6

    2018 arXiv

  21. [29]

    Rubenstein, Charles V

    Daniel I. Rubenstein, Charles V . Stewart, Tanya Y . Berger- Wolf, Jason Parham, Jon Crall, Clara Machogu, Paula Kahumbu, and Njambi Maingi. The Great Zebra and Giraffe Count: The Power and Rewards of Citizen Science. Tech- nical report, Kenya Wildlife Service, Nairobi, Kenya, 2015. 1

    2015

  22. [30]

    Superglue: Learning feature matching with graph neural networks.arXiv preprint arXiv:1911.11763, 2020

    Paul-Edouard Sarlin, Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabinovich. Superglue: Learning feature matching with graph neural networks.arXiv preprint arXiv:1911.11763, 2020. 4

    1911 arXiv

  23. [31]

    Structure-from-motion revisited

    Johannes Lutz Sch ¨onberger and Jan-Michael Frahm. Structure-from-motion revisited. InConference on Com- puter Vision and Pattern Recognition (CVPR), 2016. 4

    2016

  24. [32]

    Pixelwise view selection for un- structured multi-view stereo

    Johannes Lutz Sch ¨onberger, Enliang Zheng, Marc Pollefeys, and Jan-Michael Frahm. Pixelwise view selection for un- structured multi-view stereo. InEuropean Conference on Computer Vision (ECCV), 2016. 4

    2016

  25. [33]

    Mobile 3d mapping for surveying earthwork projects using an unmanned aerial vehicle (uav) system.Automation in construction, 41:1–14,

    Sebastian Siebert and Jochen Teizer. Mobile 3d mapping for surveying earthwork projects using an unmanned aerial vehicle (uav) system.Automation in construction, 41:1–14,

  26. [34]

    Advances in methods for marine mammal and fish stock assessments: Thermal imagery and camtrawl.Marine Technology Society Journal, 49(2):99–106, 2015

    Mike Sigler, Doug DeMaster, Peter Boveng, Michael Cameron, Erin Moreland, Kresimir Williams, and Rick Towler. Advances in methods for marine mammal and fish stock assessments: Thermal imagery and camtrawl.Marine Technology Society Journal, 49(2):99–106, 2015. 2

    2015

  27. [35]

    Unmanned aerial ve- hicle (uav) based mapping in engineering geological surveys: Considerations for optimum results.Engineering Geology, 232:12–21, 2018

    O Tziavou, S Pytharouli, and J Souter. Unmanned aerial ve- hicle (uav) based mapping in engineering geological surveys: Considerations for optimum results.Engineering Geology, 232:12–21, 2018. 1

    2018

  28. [36]

    Marine Mammal Protection Act of 1972.https://www.govinfo.gov/app/details/ COMPS-1679, 1972

    United States Congress. Marine Mammal Protection Act of 1972.https://www.govinfo.gov/app/details/ COMPS-1679, 1972. Public Law 92-522; Effective Decem- ber 21, 1972; Accessed 2025-06-28. 2

    1972

  29. [37]

    Department of Commerce, National Marine Fisheries Service, U.S

    U.S. Department of Commerce, National Marine Fisheries Service, U.S. Department of the Interior, Fish and Wildlife Service, and Indigenous Peoples Council for Marine Mam- mals. Memorandum of agreement for negotiation of ma- rine mammal protection act section 119 agreements. Me...

    2006

  30. [38]

    Department of Commerce, National Oceanic and At- mospheric Administration, National Marine Fisheries Ser- vice

    U.S. Department of Commerce, National Oceanic and At- mospheric Administration, National Marine Fisheries Ser- vice. Polar Ecosystems Program.https : / / www . fisheries . noaa . gov / inport / item / 23108,

  31. [39]

    Yolov9: Learning what you want to learn us- ing programmable gradient information.arXiv preprint arXiv:2402.13616, 2024

    Chien-Yao Wang, I-Hau Yeh, and Hong-Yuan Mark Liao. Yolov9: Learning what you want to learn us- ing programmable gradient information.arXiv preprint arXiv:2402.13616, 2024. 7

    2024 arXiv

  32. [40]

    Xiaoming Zhao, Xingming Wu, Weihai Chen, Peter C. Y . Chen, Qingsong Xu, and Zhengguo Li. Aliked: A lighter keypoint and descriptor extraction network via deformable transformation.arXiv preprint arXiv:2304.03608, 2023. 4 10

    2023 arXiv

  33. [2024]

    Accessed: 2025-06-28. 3

    2025

Pith tools

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