Continuous Marine Tracking via Autonomous UAV Handoff

(2) School of Natural Sciences; Alice James (1); Avishkar Seth (1); Endrowednes Kuantama (1); Heegyeong Kim (1); Jane Williamson (2); Macquarie University; Macquarie University); Richard Han (1) ((1) School of Computing; Yimeng Feng (1)

REVIEW 6 major objections 5 minor 25 references

Continuous Marine Tracking via Autonomous UAV Handoff

T0 review · 6 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A two-UAV system uses ORB feature matching to hand off live shark tracking between drones, extending monitoring beyond a single battery.

desk verdict The paper is an honest engineering integration with an overstated abstract: the tracker comparison is solid, but the 'seamless handoff' claim is only backed by template-crop matching in HIL, not a real two-drone transfer. read the letter →

arxiv 2507.12763 v1 pith:NZGJUBLH submitted 2025-07-17 cs.CV cs.RO

Heegyeong Kim (1) , Alice James (1) , Avishkar Seth (1) , Endrowednes Kuantama (1) , Jane Williamson (2) , Yimeng Feng (1) , Richard Han (1) ((1) School of Computing , Macquarie University

show 2 more authors

(2) School of Natural Sciences Macquarie University)

This is my paper

classification cs.CVcs.RO

keywords UAVhandoffsharktrackingOSTrackORBfeaturematchingmarinemonitoringhardware-in-the-loopJetsonNanovisual

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

Continuous observation of sharks from a single drone ends when the battery does. This paper tries to establish that a pair of cooperating drones can remove that limit: the first drone tracks the shark with an onboard OSTrack visual tracker, and when its battery runs low, a second drone takes over through a handoff protocol based on ORB feature matching and ArUco-marker localization. The system is tested on a curated 5,200-frame aerial shark dataset in which OSTrack shows the strongest robustness (fewest failure frames and shortest failure streak), and the handoff's two-stage template matching achieves 82.9% target coverage with a longest continuous match of 236 frames (over 7 seconds). If these results hold, coordinated UAV teams become a practical way to sustain real-time marine monitoring beyond the endurance of any single aircraft.

What carries the argument

The load-bearing mechanism is the handoff protocol plus the tracker. OSTrack, an end-to-end transformer single-object tracker, is selected as the onboard tracker because it had the lowest failure count (467 frames) and the shortest longest-failure streak (251 frames) among five tested trackers on the shark dataset. The transfer itself uses ORB (Oriented FAST and Rotated BRIEF), a rotation- and scale-invariant feature extractor and matcher, applied in a two-stage template strategy: a broad region of interest is cropped with 300-pixel padding, then the shark is re-localized with tighter templates of 30, 50, or 70 pixels, with 70 pixels giving the best coverage. An ArUco marker on D1 and a Wi-Fi link provide the geometric and data connection that lets D2 know where and what to look for.

What would settle it

Run a real two-drone flight over water with D1 and D2 at different altitudes and yaw angles while the shark dives and resurfaces, and measure ORB-based target coverage through the live transfer; if coverage drops well below the reported 82.9% or the match breaks for longer than the battery handoff window, the central handoff claim is refuted.

Watch

Extended reading notes

Core claim

The paper's central claim is that continuous marine tracking can be achieved by two UAVs that pass a tracked shark from one to the other rather than by extending a single drone's flight time. The first drone (D1) tracks the target; the second (D2) navigates to D1, detects an ArUco marker to lock on to D1's pose, completes a handshake, and receives live tracking state over Wi-Fi. D2 then attempts target acquisition via ORB feature matching between its own view and the transmitted template, and once matched, tracking responsibility transfers. The authors report 82.9% target coverage for this handoff and stable end-to-end operation at 4–5 Hz on an NVIDIA Jetson Nano, concluding that coordinated UAV operations are viable for extended marine tracking.

Load-bearing premise

The load-bearing premise is that template crops labeled as D1 and D2 views in a hardware-in-the-loop setting faithfully reproduce a real simultaneous two-drone handoff; if that proxy misses true viewpoint mismatch, timing, or battery constraints, the central claim collapses.

Editorial extensions

If this is right

If the handoff works, a fleet of cooperating drones can monitor a moving marine target for longer than any single battery allows, with each drone cycling back to recharge.
OSTrack's robustness profile (467 failure frames, 251-frame worst streak) suggests it is a reliable long-term tracker for visually noisy marine environments.
The reported 4–5 Hz end-to-end loop at 7.4 W on a Jetson Nano indicates the full tracking-plus-handoff pipeline fits within small-UAV power budgets.
The handoff's 82.9% target coverage with a 236-frame continuous match (about 7 seconds) is enough to keep the target in view during a battery-limited transfer.

Reading between the lines

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

Beyond the paper's own tests, a live dual-UAV flight with real simultaneous viewpoints, radio latency, and battery swaps is the decisive validation the current hardware-in-the-loop setup cannot provide; the 82.9% figure is a proxy until then.
The two-stage template padding trade-off (30, 50, 70 pixels) suggests per-scene tuning could raise coverage further, but the paper does not explore it.
The protocol should transfer to other visually matchable targets, though sharks' contrast against water may make them an easier case than low-contrast or camouflaged species.
Combining OSTrack's robustness with MixFormer's higher raw accuracy might push success rates above the reported 81.9%, but that fusion is not tested here.

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.

Desk Editor's Note

read the letter

If you are looking for a validated demonstration of continuous multi-UAV marine tracking, this isn't it. But as an engineering integration paper with a useful tracker comparison on a realistic shark video, it's worth a look.

The paper compares five trackers (OSTrack, MixFormer, SiamAPN, AiATrack, Track-Anything) on a challenging aerial shark dataset. The dataset itself—6,928 frames of a shark in shallow water with glare, occlusion, and background clutter—is a legitimate contribution, even though it's a single video. The selection of OSTrack over MixFormer is well-argued: MixFormer has higher IoU but a 511-frame failure streak, while OSTrack recovers faster. That's a sensible robustness-first decision for long-duration monitoring. The hardware-in-the-loop setup with Jetson Nano, ROS, and Pixhawk is real engineering, and the power/GPU/CPU measurements are useful.

The central issue is that the 'seamless handoff' is never actually tested as a handoff. Section 5.2 reports ORB matching between D1-view and D2-view template crops, with target coverage of 82.98% and a longest streak of 236 frames. That is a template-matching proxy, not an end-to-end transfer where D2 acquires the target and follows it in closed loop. No real flight, no timing of the transfer, no battery state. So the 82.9% coverage validates the matching algorithm, not the claimed seamless handoff. The abstract makes it worse by saying '81.9% success during real-time flight control at 100 Hz'—the 81.9% is MixFormer's success rate on the dataset, not a flight result, and 100 Hz is the low-level Ardupilot loop, not the vision pipeline (4–5 Hz). The dataset size also varies: Section 3.2 says 6,928 frames, Section 5.1 says 5,200. And padding in Section 5.2 says 30/50/100, but Table 3 reports 30/50/70. These are the kind of inconsistencies that make a referee lose trust.

That said, the core engineering concept is plausible, and the tracker comparison is honest. The overclaiming is in the abstract and the interpretation of the handoff proxy. With a real flight test or at least a closed-loop simulation with two camera viewpoints, this could be a solid application paper.

My recommendation: send it to peer review, but flag the abstract and the handoff claim for major revision. The paper deserves referee time because the dataset and tracker comparison have value, and the handoff approach is worth discussing. A good referee could push the authors to either soften the claims or provide real two-drone validation.

Referee Report

6 major / 5 minor

Summary. The paper presents a two-UAV system for continuous, real-time tracking of sharks, combining an onboard Jetson Nano, a stabilized RGB-D camera, the OSTrack tracker, and an ORB-based inter-UAV handoff protocol. It evaluates five trackers on a single aerial shark video and reports hardware-in-the-loop results for the handoff and system resource usage. The abstract claims a tracking success rate of 81.9% at 100 Hz and 82.9% target coverage during handoff, but the body of the paper reports different numbers for the selected tracker and a much lower end-to-end pipeline rate.

Significance. Coordinated multi-UAV tracking with seamless handoff would be a valuable capability for marine monitoring, and the paper's comparison of modern trackers on real marine footage is a useful data point. The hardware-in-the-loop measurement of CPU, GPU, and memory usage is also informative. However, the central claims are not currently supported: the handoff is validated only through template matching in a simulated setting, not through a closed-loop inter-UAV transfer, and several key numbers in the abstract conflict with the experimental sections. These issues substantially limit the paper's contribution in its present form.

major comments (6)

[Abstract vs. Section 5.1 and 5.2] The abstract states that the system achieves an 81.9% tracking success rate 'during real-time flight control at 100 Hz.' However, Table 1 reports that MixFormer, not the selected OSTrack, achieves the 81.9% success rate, while OSTrack achieves 74.8%. Moreover, Section 5.2 states that the end-to-end pipeline runs at 4–5 Hz, with 100+ Hz reserved for the low-level Ardupilot attitude loop. The abstract's headline number and rate are therefore not supported by the body.
[Section 5.2 (UAV Data Handover)] The handoff is evaluated by matching ORB features between D1-view and D2-view template crops in a hardware-in-the-loop environment. No experiment demonstrates D2 acquiring and following the same target after D1 hands off, and no handoff latency, viewpoint-difference, or battery-state measurements are reported. Consequently, the 82.9% target coverage and 236-frame streak validate a template-matching proxy rather than the claimed seamless inter-UAV handoff.
[Section 3.2 vs. Section 5.1] The dataset is described as containing 6,928 frames in Section 3.2, but Section 5.1 states that the evaluation uses 5,200 frames. All percentages in Tables 1 and 2 are computed against an unspecified denominator, making the reported metrics ambiguous and not reproducible.
[Section 5.2, Table 3] The text states that template paddings of 30, 50, and 100 pixels were evaluated, but Table 3 reports paddings of 30, 50, and 70 pixels, with the peak target coverage of 82.98% at 70 pixels. This discrepancy directly affects the central handoff result and must be resolved.
[Section 5.1 and 5.2 (in-sample evaluation)] The tracker is selected based on its performance on the same 5,200-frame evaluation set, and the template padding is tuned on the same video used to report the 82.98% peak coverage. This in-sample selection and tuning introduce optimistic bias; no held-out test set, cross-validation, or separate validation protocol is described.
[Tables 1 and 2 (continuous tracking claim)] The paper describes the system as providing 'continuous' tracking, but Table 2 shows OSTrack has 467 failure frames and a longest failure streak of 251 frames, which at the stated 50 fps frame rate corresponds to more than 5 seconds of sustained tracking loss. The 'continuous' characterization is not supported unless a recovery mechanism or an explicit definition of continuity is provided.

minor comments (5)

[Section 1 (Contributions)] There is a typo in 'evaulate' and an inconsistent spacing in 'UA V' in the contributions list.
[Section 3.2] The sentence beginning 'the dataset includes 6928 frames' should start with a capital letter, and the frame count should be reconciled with Section 5.1.
[Figure 4] The label 'OSTracker' in Figure 4 should be 'OSTrack' for consistency with the rest of the paper.
[Tables 1 and 2] The tracker name 'Track Anything' appears with and without a hyphen; please standardize the spelling.
[Section 4 and 4.1] The protocol description is internally inconsistent: Section 4 states the ArUco marker is on top of D2 and D1 estimates D2's pose, while Section 4.1 says D2 detects the ArUco marker on D1. Clarify which drone carries the marker and how the roles are assigned.

Circularity Check

1 steps flagged · score 6.0 of 10

The handoff coverage number is selected by construction: template padding is swept on the evaluation set and the GT-coverage metric rewards larger boxes, so 82.98% is a tuned maximum rather than an independent handoff result.

self definitional [Section 5.2, Table 3; metric definition in Section 5.1]
"With a 70-pixel template, the system achieved the highest target coverage of 82.98%, with the longest matching streak of 236 frames. ... Ground Truth Coverage metric (GT Cover %), which quantifies how often and how fully the predicted bounding box encompasses the ground truth object, regardless of its exact shape or scale."

The handoff headline metric is the GT-coverage metric, which by definition counts how often the predicted bounding box merely contains the ground truth object, regardless of alignment or scale. In Section 5.2 the paper then varies template padding (30, 50, and 100 pixels in the text; 70 pixels in Table 3) and reports the largest coverage value, 82.98%, as the handoff performance. Since additional padding directly enlarges the template and hence the matched bounding box around the target, the observed increase from 68.37% to 82.98% is the metric being maximized by a hyperparameter choice, not evidence of correct target transfer.

full rationale

The paper contains no derivation chain in the usual sense; most results are direct empirical measurements, and there is no load-bearing self-citation or imported uniqueness theorem. The tracker comparison in Section 5.1 is an independent benchmark with real content. The circular step is localized to Section 5.2: the GT-coverage metric is defined to reward predicted boxes that encompass the ground truth regardless of shape or scale, and then template padding is swept as if evaluating handoff quality. Larger padding mechanically enlarges the matched box, so the reported 82.98% peak coverage is an artifact of the metric definition combined with in-sample padding selection. The abstract's '82.9% target coverage' therefore does not independently validate the seamless handoff claim; it is a tuned maximum over the same footage used to measure it. This is partial circularity of the central handoff result, giving a score of 6, while the remainder of the paper's tracking evaluation remains non-circular. Additional inconsistencies, such as the abstract attributing 81.9% success to real-time flight at 100 Hz while Table 1 assigns 81.9% to MixFormer and Section 5.2 reports the end-to-end pipeline at 4-5 Hz, are correctness concerns rather than circularity.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

No new physical entities are introduced. The main results rest on one curated video and a HIL proxy for handoff; the matching template padding is tuned on the same data used to report the headline number.

free parameters (2)

Template padding for handoff matching = 70 px
Chosen by comparing 30/50/70 pixel paddings on the same evaluation footage; the best value is reported as 82.98% target coverage (Table 3).
Broad ROI padding = 300 px
Hand-set cropping margin in the two-stage matching pipeline (Section 5.2); no sensitivity analysis is provided.

assumptions (3)

domain assumption A single 2m25s sunny coastal video of one shark represents diverse dynamic marine tracking conditions.
Robustness claims about occlusion, illumination, and clutter are drawn from this one curated sequence (Sections 3.2, 5.1); no cross-site or multi-condition validation is provided.
domain assumption Template crops labeled as D1 and D2 views in a hardware-in-the-loop setup reproduce the geometry and timing of a real two-UAV handoff.
The handoff evaluation (Section 5.2, Figure 5) uses these crops and HIL metrics; no real simultaneous two-drone flight or battery-driven transfer is reported.
domain assumption ArUco marker detection plus GPS gives D1 an accurate estimate of D2's pose (X, Y, alpha).
Section 4 relies on this for inter-drone alignment; no error analysis of marker or GPS accuracy is given.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Continuous Marine Tracking via Autonomous UAV Handoff." pith.science (2026). https://pith.science/paper/NZGJUBLH

@misc{pith2026250712763,
  author       = {Pith},
  title        = {Pith review of: Continuous Marine Tracking via Autonomous UAV Handoff},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NZGJUBLH}},
  note         = {Machine review of arXiv:2507.12763}
}

read the original abstract

This paper introduces an autonomous UAV vision system for continuous, real-time tracking of marine animals, specifically sharks, in dynamic marine environments. The system integrates an onboard computer with a stabilised RGB-D camera and a custom-trained OSTrack pipeline, enabling visual identification under challenging lighting, occlusion, and sea-state conditions. A key innovation is the inter-UAV handoff protocol, which enables seamless transfer of tracking responsibilities between drones, extending operational coverage beyond single-drone battery limitations. Performance is evaluated on a curated shark dataset of 5,200 frames, achieving a tracking success rate of 81.9\% during real-time flight control at 100 Hz, and robustness to occlusion, illumination variation, and background clutter. We present a seamless UAV handoff framework, where target transfer is attempted via high-confidence feature matching, achieving 82.9\% target coverage. These results confirm the viability of coordinated UAV operations for extended marine tracking and lay the groundwork for scalable, autonomous monitoring.

Figures

Figures reproduced from arXiv: 2507.12763 by the authors.

**Figure 2.** Proposed system architecture of the continu [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Seamless Data Transfer Flowchart With D2 now handling the tracking, Drone D1 can safely return to its base station. This transfer method ensures continuous target monitoring, allowing the drones to alternate tracking duties - meaning that D2 is now referred to as D1. By distributing the workload, each drone conserves battery life and extends mission duration, resulting in a more efficient and effective monitoring s… view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: OSTracker performance In terms of drift, OSTrack maintained a low drift of 5.6 pixels, confirming its spatial consistency when tracking correctly. SiamAPN and AiATrack have very low drift values, 2.8 and 2.2 pixels, respectively. These values are considered unreliable…

**Figure 5.** Figure 5: Handover: (a) Feature matching; System per [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

25 extracted references · 22 canonical work pages

[1]

Australia, T.C.S.: Australian shark incident database (2025), https://taronga.org.au/conservation-and-science/australian-shark- incident-database, accessed: April 11, 2025

work page 2025
[2]

Drones 5(1), 8 (2021)

Butcher, P.A., Colefax, A.P., Gorkin III, R.A., Kajiura, S.M., López, N.A., Mourier, J., Purcell, C.R., Skomal, G.B., Tucker, J.P., Walsh, A.J., et al.: The drone revolution of shark science: A review. Drones 5(1), 8 (2021)

work page 2021
[3]

SiamAPN++: Siamese Attentional Aggregation Network for Real-Time UAV Tracking

Cao, Z., Fu, C., Ye, J., Li, B., Li, Y.: Siamapn++: Siamese attentional aggregation network for real-time uav tracking (2021), https://arxiv. org/abs/2106.08816

work page Pith review arXiv 2021
[4]

Sensors 23(22), 9193 (2023)

Colefax, A.P., Walsh, A.J., Purcell, C.R., Butcher, P.: Utility of spectral filtering to improve the reliability of marine fauna detections from drone-based monitoring. Sensors 23(22), 9193 (2023)

work page 2023
[5]

IEEE Transactions on Pat- tern Analysis and Machine Intelligence 46(6), 4129–4146 (2024)

Cui, Y., Jiang, C., Wu, G., Wang, L.: Mixformer: End-to-end track- ing with iterative mixed attention. IEEE Transactions on Pat- tern Analysis and Machine Intelligence 46(6), 4129–4146 (2024). https://doi.org/10.1109/TPAMI.2024.3349519

arXiv 2024
[6]

Remote Sensing in Ecology and Conservation 7(3), 341–354 (2021)

Dujon, A.M., Ierodiaconou, D., Geeson, J.J., Arnould, J.P., Allan, B.M., Katselidis, K.A., Schofield, G.: Machine learning to detect marine ani- mals in uav imagery: effect of morphology, spacing, behaviour and habitat. Remote Sensing in Ecology and Conservation 7(3), 341–354 (2021)

work page 2021
[7]

Marine Ecology Progress Series 689, 77–94 (2022)

Elliott, R.G., Montgomery, J.C., Della Penna, A., Radford, C.A.: Satellite tags describe movement and diving behaviour of blue sharks prionace glauca in the southwest pacific. Marine Ecology Progress Series 689, 77–94 (2022)

work page 2022
[8]

Environmental Monitoring and Assessment 197(5), 503 (2025)

Ganie, P.A., Khatei, A., Posti, R., Sidiq, M.J., Pandey, P.K.: Unmanned aerial vehicles in fisheries and aquaculture: a comprehensive overview. Environmental Monitoring and Assessment 197(5), 503 (2025)

work page 2025

Show all 25 references

[9]

In: European Conference on Computer Vision

Gao, S., Zhou, C., Ma, C., Wang, X., Yuan, J.: Aiatrack: Attention in attention for transformer visual tracking. In: European Conference on Computer Vision. pp. 146–164. Springer (2022)

2022
[10]

Drones 4(2), 18 (2020)

Gorkin III, R., Adams, K., Berryman, M.J., Aubin, S., Li, W., Davis, A.R., Barthelemy, J.: Sharkeye: Real-time autonomous personal shark alerting via aerial surveillance. Drones 4(2), 18 (2020)

2020
[11]

Endangered Species Research 47, 171–203 (2022)

Jorgensen, S.J., Micheli, F., White, T.D., Van Houtan, K.S., Alfaro- Shigueto, J., Andrzejaczek, S., Arnoldi, N.S., Baum, J.K., Block, B., Britten, G.L., et al.: Emergent research and priorities for shark and ray conservation. Endangered Species Research 47, 171–203 (2022)

2022
[12]

arXiv preprint arXiv:1710.02726 (2017)

Karami, E., Prasad, S., Shehata, M.: Image matching using sift, surf, brief and orb: performance comparison for distorted images. arXiv preprint arXiv:1710.02726 (2017)

2017 arXiv
[13]

Kristan, M., Leonardis, A., Matas, J., Felsberg, M., Pfugfelder, R., Če- hovin Zajc, L., Vojir, T., Bhat, G., Lukezic, A., Eldesokey, A., Fernandez, G., et al.: The sixth visual object tracking vot2018 challenge results (2018)

2018
[14]

arXiv preprint arXiv:2501.05717 (2025)

Lalgudi, C.K., Leone, M.E., Clark, J.V., Madrigal-Mora, S., Espinoza, M.: Zero-shot shark tracking and biometrics from aerial imagery. arXiv preprint arXiv:2501.05717 (2025)

2025 arXiv
[15]

Computers and Electronics in Agriculture 224, 109161 (2024)

Liu, Y., Li, W., Liu, X., Li, Z., Yue, J.: Deep learning in multiple animal tracking: A survey. Computers and Electronics in Agriculture 224, 109161 (2024)

2024
[16]

Science 372(6537), 84–87 (2021)

O’Hara, C.C., Frazier, M., Halpern, B.S.: At-risk marine biodiversity faces extensive, expanding, and intensifying human impacts. Science 372(6537), 84–87 (2021)

2021
[17]

Frontiers in Marine Science 9, 981897 (2022)

Purcell, C.R., Walsh, A.J., Colefax, A.P., Butcher, P.: Assessing the ability of deep learning techniques to perform real-time identification of shark species in live streaming video from drones. Frontiers in Marine Science 9, 981897 (2022)

2022
[18]

In: 2011 International conference on computer vision

Rublee, E., Rabaud, V., Konolige, K., Bradski, G.: Orb: An efficient al- ternative to sift or surf. In: 2011 International conference on computer vision. pp. 2564–2571. Ieee (2011)

2011
[19]

In: Humanity Driven AI: Productivity, Well-being, Sustain- ability and Partnership, pp

Sharma, N., Saqib, M., Scully-Power, P., Blumenstein, M.: Sharkspotter: Shark detection with drones for human safety and environmental protection. In: Humanity Driven AI: Productivity, Well-being, Sustain- ability and Partnership, pp. 223–237. Springer (2021)

2021
[20]

arXiv preprint arXiv:2407.20623 (2024)

Varini, F., Gayford, J.H., Jenrette, J., Witt, M.J., Garzon, F., Ferretti, F., Wilday, S., Bond, M.E., Heithaus, M.R., Robinson, D., et al.: Sharktrack: an accurate, generalisable software for streamlining shark and ray underwater video analysis. arXiv preprint arXiv:2407.20623 (2024)

2024 arXiv
[21]

Ecological Informatics 80, 102499 (2024)

Villon, S., Iovan, C., Mangeas, M., Vigliola, L.: Toward an artificial intelligence-assisted counting of sharks on baited video. Ecological Informatics 80, 102499 (2024)

2024
[22]

Commu- nications Biology 7(1), 1018 (2024)

Williamson, M.J., Tebbs, E.J., Curnick, D.J., Ferretti, F., Carlisle, A.B., Chapple, T.K., Schallert, R.J., Tickler, D.M., Block, B.A., Jacoby, D.M.: Environmental stress reduces shark residency to coral reefs. Commu- nications Biology 7(1), 1018 (2024)

2024
[23]

Remote Sensing 14(10), 2432 (2022)

Winsen, M., Denman, S., Corcoran, E., Hamilton, G.: Automated de- tection of koalas with deep learning ensembles. Remote Sensing 14(10), 2432 (2022)

2022
[24]

arXiv preprint arXiv:2304.11968 (2023)

Yang, J., Gao, M., Li, Z., Gao, S., Wang, F., Zheng, F.: Track anything: Segment anything meets videos. arXiv preprint arXiv:2304.11968 (2023)

2023 arXiv
[25]

Ye, B., Chang, H., Ma, B., Shan, S., Chen, X.: Joint feature learning and relation modeling for tracking: A one-stream framework (2022), https://arxiv.org/abs/2203.11991 6

2022 arXiv

Pith tools

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

[1] [1]

Australia, T.C.S.: Australian shark incident database (2025), https://taronga.org.au/conservation-and-science/australian-shark- incident-database, accessed: April 11, 2025

work page 2025

[2] [2]

Drones 5(1), 8 (2021)

Butcher, P.A., Colefax, A.P., Gorkin III, R.A., Kajiura, S.M., López, N.A., Mourier, J., Purcell, C.R., Skomal, G.B., Tucker, J.P., Walsh, A.J., et al.: The drone revolution of shark science: A review. Drones 5(1), 8 (2021)

work page 2021

[3] [3]

SiamAPN++: Siamese Attentional Aggregation Network for Real-Time UAV Tracking

Cao, Z., Fu, C., Ye, J., Li, B., Li, Y.: Siamapn++: Siamese attentional aggregation network for real-time uav tracking (2021), https://arxiv. org/abs/2106.08816

work page Pith review arXiv 2021

[4] [4]

Sensors 23(22), 9193 (2023)

Colefax, A.P., Walsh, A.J., Purcell, C.R., Butcher, P.: Utility of spectral filtering to improve the reliability of marine fauna detections from drone-based monitoring. Sensors 23(22), 9193 (2023)

work page 2023

[5] [5]

IEEE Transactions on Pat- tern Analysis and Machine Intelligence 46(6), 4129–4146 (2024)

Cui, Y., Jiang, C., Wu, G., Wang, L.: Mixformer: End-to-end track- ing with iterative mixed attention. IEEE Transactions on Pat- tern Analysis and Machine Intelligence 46(6), 4129–4146 (2024). https://doi.org/10.1109/TPAMI.2024.3349519

arXiv 2024

[6] [6]

Remote Sensing in Ecology and Conservation 7(3), 341–354 (2021)

Dujon, A.M., Ierodiaconou, D., Geeson, J.J., Arnould, J.P., Allan, B.M., Katselidis, K.A., Schofield, G.: Machine learning to detect marine ani- mals in uav imagery: effect of morphology, spacing, behaviour and habitat. Remote Sensing in Ecology and Conservation 7(3), 341–354 (2021)

work page 2021

[7] [7]

Marine Ecology Progress Series 689, 77–94 (2022)

Elliott, R.G., Montgomery, J.C., Della Penna, A., Radford, C.A.: Satellite tags describe movement and diving behaviour of blue sharks prionace glauca in the southwest pacific. Marine Ecology Progress Series 689, 77–94 (2022)

work page 2022

[8] [8]

Environmental Monitoring and Assessment 197(5), 503 (2025)

Ganie, P.A., Khatei, A., Posti, R., Sidiq, M.J., Pandey, P.K.: Unmanned aerial vehicles in fisheries and aquaculture: a comprehensive overview. Environmental Monitoring and Assessment 197(5), 503 (2025)

work page 2025