Pith. sign in

REVIEW 3 major objections 6 minor 41 references

Tracking the Flight: Exploring a Computational Framework for Analyzing Escape Responses in Plains Zebra (Equus quagga)

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Quarter-body-length error lets drone footage track zebra escapes

desk verdict A solid, useful comparison of drone-motion unwrapping methods for animal tracking, with an honest external validation; the behavioral patterns are preliminary and the mean error hides a heavy tail that deserves more analysis. read the letter →

arxiv 2505.16882 v2 pith:IDHRN4YE submitted 2025-05-22 cs.CV

Isla Duporge , Sofia Minano , Nikoloz Sirmpilatze , Igor Tatarnikov , Scott Wolf , Adam L. Tyson , Daniel Rubenstein This is my paper · ORCID
classification cs.CV
keywords dronevideoanalysisStructure-from-Motioncollectiveanimalbehaviorplainszebraescaperesponseposeestimationtrajectoryunwrappingopen-sourcepipeline
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 seeks to establish that a fully open-source pipeline can convert drone video of moving animals into world-coordinate trajectories with enough accuracy for behavioral analysis, and that doing so for a single escape event can produce biologically meaningful collective patterns. The authors compare three ways of 'unwrapping' a moving-camera video into fixed ground coordinates: frame-to-frame image registration, Structure-from-Motion with linearly interpolated camera poses, and SfM with registration-based rotation. Using static trees as ground references, they find the linearly interpolated SfM approach has the lowest error, an average dispersion of $0.275$ zebra body lengths—about a quarter of a body length. Applied to a 3.5-minute video of 44 zebras fleeing a simulated predator approach, the method yields measurable patterns: alignment increases during escape, spacing briefly widens just before the group stops, and zebras near the herd's center are more aligned than those on the periphery. This matters because it suggests a practical, low-cost route from raw drone footage to quantitative collective-behavior data in wild terrestrial mammals.

What carries the argument

The load-bearing element is a two-stage 'unwrapping' pipeline. First, Structure-from-Motion (SfM) estimates camera poses on a sparse set of keyframes (every 20th frame plus the first and last), with pixels belonging to zebras masked out so the moving animals do not corrupt the static scene reconstruction; the paper reports an average reprojection error of $0.58$ pixels. Second, the camera position at each missing frame is linearly interpolated between keyframes assuming constant velocity, while orientation is handled either by spherical linear interpolation or by registering each frame to its nearest keyframe. Each observed image point is then projected along its viewing ray onto the best-fitting plane to the reconstructed 2.5D mesh, yielding a world-coordinate trajectory. Static trees, tracked with a deep-learning detector and a multi-object tracker, serve as the objective error reference: since trees do not move, the dispersion of their unwrapped trajectories quantifies the accuracy of each unwrapping method. The behavioral metrics (polarization, speed, inter-individual distances, distance from centroid) are then computed from the head and tail keypoints supplied by a pose-estimation model.

What would settle it

Record a controlled test flight where a ground vehicle with RTK-GPS follows a known route across the same savannah terrain while the drone performs the same escape-event maneuvers, unwrap the vehicle's trajectory with the SfM-linear-interpolation method, and compare per-frame positions to the GPS ground truth; if the average error during accelerated or turning flight segments exceeds the $0.275$-body-length benchmark, the method's suitability for behavioral analysis of fast-moving zebras would be falsified.

Watch

Extended reading notes

Core claim

The paper's central claim is that the interpolated Structure-from-Motion approach—estimating camera poses on every 20th frame and then linearly interpolating the missing poses under a constant-velocity assumption—unwraps drone footage into world coordinates with an average tree-dispersion error of $0.275$ zebra body lengths, outperforming both pure image registration ($0.910$ body lengths) and SfM with registration-based rotation ($0.299$ body lengths). The authors argue that this error level makes the extracted individual trajectories trustworthy for behavioral analysis, and they demonstrate the claim by reconstructing the paths of 44 zebras during a four-wave escape event. From those trajectories they report that herd polarization rises with running speed, that the mean inter-individual distance briefly increases immediately before the group stops, and that centrally positioned zebras are more aligned with the herd's mean direction than peripheral ones. They also note that, contrary to the classic selfish-herd prediction, individuals do not consistently move toward the herd's center during the escape.

Load-bearing premise

The accuracy of the reconstructed zebra trajectories depends on the ground being a single flat plane and on the drone moving at constant speed between the sparse reconstruction frames—if either assumption is violated, the reported quarter-body-length error may be too optimistic for animals in motion.

Editorial extensions

If this is right

  • The same pipeline can be applied to the rest of the collected videos to test whether the observed escape patterns hold across group sizes and contexts, without additional annotation or calibration.
  • The tree-based error metric provides a simple benchmarking protocol that can be reused to evaluate future unwrapping methods on conservation footage.
  • The behavioral findings—speed-correlated alignment, pre-stop spacing increase, and central-individual alignment—offer concrete, quantitative predictions about zebra coordination that could be compared with other terrestrial herd species.
  • Because the method uses only open-source tools and runs on a consumer GPU, it lowers the technical barrier for conservationists to move from raw drone footage to trajectory data.

Reading between the lines

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

  • The static-tree error likely underestimates the true trajectory error for zebras because trees have high-contrast, rigid shapes and do not suffer motion blur or partial occlusion; a field validation with objects of known motion would be needed to confirm the $0.275$-body-length accuracy for moving targets.
  • The constant-velocity interpolation assumption is the most fragile link: any drone acceleration or turn between keyframes will systematically bias positions, and because these maneuvers often coincide with the start of an escape wave, error may be concentrated exactly when behavioral metrics are most variable.
  • The finding that central zebras are more aligned could partly reflect a tracking artifact if peripheral animals are more often occluded or have lower pose confidence; this is testable by checking whether the correlation between centroid distance and alignment persists after excluding low-confidence tracks.
  • The pipeline creates a natural bridge to theoretical models of collective motion: the polarization-speed relationship and the pre-stop spacing peak could be fit to existing self-propelled particle models to infer interaction rules for zebras, something the paper does not do.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper presents a computational pipeline for unwrapping drone-captured video of moving animals into world coordinates, using three methods: frame-to-frame image registration (itk-elastix), Structure-from-Motion with linear pose interpolation, and SfM with registration-based rotation interpolation. The methods are validated by tracking 45 static trees and measuring their dispersion after unwrapping; the SfM with linear interpolation achieves the lowest error (weighted mean 0.275 zebra body lengths). The selected method is then applied to a single 3.5-minute drone video of 44 plains zebras during four escape waves, from which the authors extract individual trajectories and compute collective metrics: polarization, inter-individual spacing, speed, and center-periphery alignment. They report increased alignment during escape, a transient widening of spacing before stopping, and higher alignment near the herd center, framing these as a proof of concept rather than a generalizable behavioral finding.

Significance. The paper addresses a practical and timely problem: extracting reliable animal trajectories from drone footage without ground control points or GPS, using open-source tools. The validation against static trees is an external, non-circular benchmark and provides a concrete error estimate in ecologically meaningful units. The authors make their code publicly available as notebooks, which supports reproducibility and follow-up work. If the validation can be shown to support per-frame behavioral analysis, the framework would be a useful contribution to collective behavior research in the field. The behavioral results are explicitly presented as illustrative, and the paper discusses limitations such as the single-video basis and the inability to identify individuals. The method has clear potential to scale to the larger dataset (41 videos) mentioned in the paper.

major comments (3)
  1. [Section 2.3.3, Table 3] The validation reports a weighted average dispersion of 0.275 body lengths for the SfM linearly interpolated method, but the per-tree statistics in Table 3 reveal a heavy-tailed error distribution: tree 2 has a maximum dispersion of 3.58 body lengths, tree 68 reaches 1.85, tree 218 reaches 1.83, and several other trees exceed 1.5. The aggregate mean therefore does not bound the error at the frames and image locations that drive the per-frame behavioral metrics (polarization, spacing, alignment). Since behavioral patterns are computed from per-frame coordinates, a small number of high-error frames could create or mask the reported effects. The authors should report the error as a function of time (or drone speed) and image position, and ideally at the actual zebra keypoint locations, or demonstrate that the behavioral conclusions are robust to excluding frames where the tree-based error exceeds a threshold.
  2. [Section 2.3.2] The camera poses between SfM keyframes are obtained by linear interpolation assuming constant velocity motion, as stated in the text. This assumption is not validated against actual drone motion during the fast maneuvers that characterize the escape waves. The tree-dispersion error is averaged over the whole video and may understate the error during periods of rapid drone acceleration or large rotation, exactly when the zebras are moving fastest and when behavioral metrics are most important. The authors should either provide a time-resolved validation (e.g., error versus drone speed estimated from the SfM keyframes) or compare the linear interpolation against an alternative that uses the image content, such as the registration-based rotation method, to show that the choice does not materially affect the behavioral metrics.
  3. [Section 2.4 and Section 3] The behavioral metrics (polarization, mean/max inter-individual distance, center-periphery alignment) are reported without uncertainty quantification or sensitivity analysis. Given the heavy-tailed error distribution of the unwrapping validation, the authors should provide confidence intervals (e.g., via bootstrapping over individuals or frames) or show that the observed temporal patterns are stable under alternative preprocessing choices, such as different keypoint confidence thresholds (currently set at 0.9), different displacement exclusion thresholds (currently 2 body lengths per frame), or different Savitzky-Golay smoothing windows (currently 7). Without such analysis, it is difficult to assess whether the reported 'brief widening of spacing before stopping' and 'higher alignment near the center' are robust features or artifacts of a few outlier frames.
minor comments (6)
  1. [Section 2.2] The text states that 1963 frames were manually annotated, 'approximately 1/5 of the video', but for a 3.5-minute video at 29.97 fps (about 6294 frames), 1963 frames is closer to one third. Please clarify the total number of frames in the selected video or correct the fraction.
  2. [Table 1 and Table 3] The tables report 'Weighted average distance' without explicitly stating the units in the table headers; the units (zebra body lengths) are mentioned in the text but should be made explicit in the tables for self-contained reading.
  3. [Section 2.4] The paper does not report what fraction of keypoints are discarded by the 0.9 confidence threshold or by the two-body-length displacement exclusion. Reporting these fractions would help readers assess the completeness of the trajectory data used for the behavioral metrics.
  4. [Section 1 and References] Several references are incomplete or inconsistently formatted (e.g., [3] lacks a journal/year, [4] lacks a year, [12] and [13] have '30:R206–R210' without a journal name, and [35] lists only the author and year without a title). The reference list should be checked against the journal's style.
  5. [Figure 2] The scatter plot of speed versus polarization would benefit from a clear colorbar label and a statement about the time interval used for averaging (the text says 'in ~1-second intervals (30 frames)', which is clear, but the colorbar should indicate 'Time (s)' explicitly rather than relying on the caption alone).
  6. [Section 2.1] The phrase 'Experimental trials were ran' should be 'Experimental trials were run'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; tree-based validation is an external benchmark.

full rationale

The paper's central claim—that the interpolated SfM method unwraps drone footage with a mean tree-dispersion error of 0.275 body lengths—is validated against 45 stationary tree trajectories detected and tracked independently with DeepForest and BotSort. This is an external benchmark: the trees are not used in the SfM reconstruction or in fitting any parameter of the unwrapping pipeline, and the dispersion metric measures the temporal stability of known-static objects. The behavioral metrics (polarization, spacing, center-periphery alignment) are computed from the chosen unwrapped tracks using standard definitions from the 'movement' package; they are observations of the processed data, not predictions derived from fitted inputs. Self-citations to [21], [24], and [36] are references to datasets and software tools, not load-bearing arguments for the method's validity. The choice of SfM with linear interpolation is made by comparing three approaches on the tree benchmark before behavioral analysis, so the benchmark is not constructed to force the reported behavioral patterns. No equation or parameter reduces to another by definition, and no fitted quantity is renamed as a prediction.

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

The paper introduces no new physical entities or parameters beyond standard analysis thresholds. The free parameters are hand-chosen analysis settings, not fitted to force a result. The axioms are domain assumptions about the plane fit, the tree error proxy, the predation proxy, and the constant-velocity interpolation, all of which are reasonable but not independently validated. The central methodological claim rests on the tree-based validation being a faithful error model for moving zebras.

free parameters (5)
  • SfM keyframe interval = every 20th frame
    Chosen by the authors, affects the density of camera pose estimates and the accuracy of interpolation between them (Section 2.3.2).
  • SLEAP keypoint confidence threshold = 0.9
    Hand-chosen threshold for discarding low-confidence keypoints, affects which frames and individuals enter the behavioral analysis (Section 2.4).
  • Displacement exclusion threshold = 2 body lengths per frame
    Hand-chosen threshold to exclude implausible jumps in centroid position, which could remove genuine fast movements (Section 2.4).
  • Body vector outlier threshold = ±2 standard deviations
    Hand-chosen threshold for removing outlier body vectors, affects orientation and polarization estimates (Section 2.4).
  • Savitzky-Golay smoothing window = 7
    Hand-chosen window size for smoothing the polarization-speed scatter plot, affects visual interpretation (Figure 2 caption).
assumptions (4)
  • domain assumption The ground surface is well approximated by a single best-fitting plane to the OpenDroneMap 2.5D mesh across the entire surveyed area.
    Trajectories are obtained by intersecting camera rays with this plane (Section 2.3.2). If the terrain is not planar, the unwrapped positions will be systematically distorted.
  • domain assumption Trees are stationary in the world coordinate system, so their unwrapped trajectory dispersion is a valid proxy for camera-pose and plane-fitting error.
    The error metric (Table 1) is defined as the weighted average distance of each tree trajectory to its centroid. This assumes trees do not move and that the error computed on static trees transfers to moving zebras.
  • domain assumption The researchers' coordinated walking approach simulates a predation event and induces a genuine escape response in the zebras.
    The study labels the behavior as 'escape waves' initiated by researchers mimicking a lion coalition (Section 2.1). If the zebras were not actually fleeing from a perceived predator, the behavioral findings would not be about escape.
  • domain assumption Camera translation between SfM keyframes can be linearly interpolated assuming constant velocity.
    Section 2.3.2 states 'we linearly interpolated the position assuming constant velocity motion.' This is a modeling choice that can fail during accelerated drone motion, and the tree-based error may not fully capture the impact on moving animals.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Tracking the Flight: Exploring a Computational Framework for Analyzing Escape Responses in Plains Zebra (Equus quagga)." pith.science (2026). https://pith.science/paper/IDHRN4YE

@misc{pith2026250516882,
  author       = {Pith},
  title        = {Pith review of: Tracking the Flight: Exploring a Computational Framework for Analyzing Escape Responses in Plains Zebra (Equus quagga)},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IDHRN4YE}},
  note         = {Machine review of arXiv:2505.16882}
}
read the original abstract

Ethological research increasingly benefits from the growing affordability and accessibility of drones, which enable the capture of high-resolution footage of animal movement at fine spatial and temporal scales. However, analyzing such footage presents the technical challenge of separating animal movement from drone motion. While non-trivial, computer vision techniques such as image registration and Structure-from-Motion (SfM) offer practical solutions. For conservationists, open-source tools that are user-friendly, require minimal setup, and deliver timely results are especially valuable for efficient data interpretation. This study evaluates three approaches: a bioimaging-based registration technique, an SfM pipeline, and a hybrid interpolation method. We apply these to a recorded escape event involving 44 plains zebras, captured in a single drone video. Using the best-performing method, we extract individual trajectories and identify key behavioral patterns: increased alignment (polarization) during escape, a brief widening of spacing just before stopping, and tighter coordination near the group's center. These insights highlight the method's effectiveness and its potential to scale to larger datasets, contributing to broader investigations of collective animal behavior.

Figures

Figures reproduced from arXiv: 2505.16882 by the authors.

Figure 1
Figure 1. Stitched drone flight path overlaid with the movement trajectories of 44 individual zebra within the observed herd. The trajectories [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Scatter plot showing co-evolution of average group [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Plot showing individual zebra alignment with the herd’s [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Frame-by-frame Pearson correlation between each ze [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 31 canonical work pages

  1. [1]

    Risk of visual detection and pursuit by a predator and the selective advantage of flocking behaviour

    Ian Vine. Risk of visual detection and pursuit by a predator and the selective advantage of flocking behaviour. Journal of Theoretical Biology , 30(2):405–422, 1971. ISSN 0022-

    work page 1971

  2. [2]

    Levin, and Daniel I

    Shay Gueron, Simon A. Levin, and Daniel I. Rubenstein. The dynamics of herds: From individuals to aggrega- tions. Journal of Theoretical Biology , 182(1):85–98, 1996. ISSN 0022-5193. doi: https://doi.org/10.1006/jtbi.1996

    work page doi:10.1006/jtbi.1996 1996

  3. [3]

    Geometry for the selfish herd

    W D Hamilton. Geometry for the selfish herd. 31:295–311. 1, 5

    work page

  4. [4]

    The dilemma of the selfish herd: the search for a realistic move- ment rule

    Steven V Viscido, Matthew Miller, and David S Wethey. The dilemma of the selfish herd: the search for a realistic move- ment rule. 217:183–194. 1

    work page

  5. [5]

    D. C. Heard. The effect of wolf predation and snow cover on musk-ox group size. American Naturalist, 139:190–204,

    work page

  6. [6]

    Spieler and K

    M. Spieler and K. E. Linsenmair. Aggregation behaviour of Bufo maculatus as an antipredator mechanism. Ethology, 105:665–686, 1999

    work page 1999

  7. [7]

    P. J. Watt, S. F. Nottingham, and S. Young. Toad tadpole ag- gregation behaviour: evidence for a predator avoidance func- tion. Animal Behaviour, 54:865–872, 1997

    work page 1997

  8. [8]

    B. J. Ens, M. Klaassen, and L. Zwarts. Flocking and feeding in the fiddler crab (Uca tangeri): prey availability as risk- taking behaviour. Netherlands Journal of Sea Research, 31: 477–494, 1993. 1

    work page 1993

Show all 41 references
  1. [9]

    Collec- tive responses of flocking sheep (Ovis aries) to a herding dog (border collie)

    Vivek Jadhav, Romain Pasqua, Chiara Zanon, et al. Collec- tive responses of flocking sheep (Ovis aries) to a herding dog (border collie). Communications Biology, 7:1543, 2024. doi: 10.1038/s42003-024-07245-7. 1

    2024 doi

  2. [10]

    Fischhoff, and Daniel I

    Anping Chen, Leslie Reperant, Ilya R. Fischhoff, and Daniel I. Rubenstein. Increased vigilance of plains ze- bras (equus quagga) in response to more bush cover- age in a kenyan savanna. Climate Change Ecology , 1: 100001, 2021. ISSN 2666-9005. doi: https://doi.org/ 10.1016/j.e...

    2021

  3. [11]

    Rubenstein

    Daniel I. Rubenstein. Equids and predators. In Herbert H. T. Prins and Iain J. Gordon, editors, The Equids: A Suite of Splendid Species, chapter 12, pages 323–345. Springer Inter- national Publishing, Cham, 2023. ISBN 978-3-031-27144-

    2023

  4. [12]

    Bird flocks

    Steven J Portugal. Bird flocks. 30:R206–R210. 1

  5. [13]

    Self- organization of collective escape in pigeon flocks

    Marina Papadopoulou, Hanno Hildenbrandt, Daniel W E Sankey, Steven J Portugal, and Charlotte K Hemelrijk. Self- organization of collective escape in pigeon flocks. 18: e1009772, . 1

  6. [14]

    URL https: //doi.org/10.1007/978- 3- 031- 27144- 1_12

    doi: 10.1007/978-3-031-27144-1 12. URL https: //doi.org/10.1007/978- 3- 031- 27144- 1_12 . 1, 2, 5

    doi

  7. [15]

    Oddity and the ‘confu- sion effect’ in predation

    Laurie Landeau and John Terborgh. Oddity and the ‘confu- sion effect’ in predation. 34:1372–1380. 1

  8. [16]

    Scale-free correlations in starling flocks

    Andrea Cavagna, Andrea Cimarelli, Irene Giardina, Gior- gio Parisi, Roberto Santagati, Francesco Stefanini, and Mas- similiano Viale. Scale-free correlations in starling flocks. Proceedings of the National Academy of Sciences of the United States of America, 107(26):11865–1187...

    2010 doi

  9. [17]

    From schooling to shoal- ing: patterns of collective motion in zebrafish (Danio re- rio)

    Nathan Miller and Robert Gerlai. From schooling to shoal- ing: patterns of collective motion in zebrafish (Danio re- rio). PLoS One, 7(11):e48865, 2012. doi: 10.1371/journal. pone.0048865. URL https://doi.org/10.1371/ journal.pone.0048865. 1

    2012 doi

  10. [18]

    Initiation and spread of escape waves within animal groups

    James E Herbert-Read, Jerome Buhl, Feng Hu, Ashley J W Ward, and David J T Sumpter. Initiation and spread of escape waves within animal groups. 2:140355. 1

  11. [19]

    Collective behaviour: Leadership and learning in flocks

    James Herbert-Read. Collective behaviour: Leadership and learning in flocks. Current Biology , 25(23):R1127– R1129, 2015. ISSN 0960-9822. doi: https://doi.org/ 10 . 1016 / j . cub. 2015 . 10 . 031. URL https : / / www . sciencedirect . com / science / article / pii / S09609822...

    2015

  12. [20]

    Boswell, Christos C

    Nils Olav Handegard, Kevin M. Boswell, Christos C. Ioan- nou, Scott P. Leblanc, Dag B. Tjøstheim, and Iain D. Couzin. The dynamics of coordinated group hunting and collective information transfer among schooling prey.Current Biology, 22(13):1213–1217, 2012. doi: 10.1016/j.cub....

    2012 doi

  13. [21]

    Baboonland dataset: Tracking primates in the wild and automating behaviour recognition from drone videos, 2024

    Isla Duporge, Maksim Kholiavchenko, Roi Harel, Scott Wolf, Dan Rubenstein, Meg Crofoot, Tanya Berger-Wolf, Stephen Lee, Julie Barreau, Jenna Kline, Michelle Ramirez, and Charles Stewart. Baboonland dataset: Tracking primates in the wild and automating behaviour recognition fro...

    2024

  14. [22]

    Z. Wu, C. Zhang, X. Gu, et al. Deep learning enables satellite-based monitoring of large populations of terrestrial mammals across heterogeneous landscape. Nature Commu- nications, 14:3072, 2023. doi: 10.1038/s41467-023-38901- y. 2

    2023 doi

  15. [23]

    Emer- gence of splits and collective turns in pigeon flocks under predation

    Marina Papadopoulou, Hanno Hildenbrandt, Daniel W E Sankey, Steven J Portugal, and Charlotte K Hemelrijk. Emer- gence of splits and collective turns in pigeon flocks under predation. 9:211898, . 2

  16. [24]

    Spiegel, Eleanor R

    Isla Duporge, Marcus P. Spiegel, Eleanor R. Thomson, Ta- tiana Chapman, Curt Lamberth, Caroline Pond, David W. Macdonald, Tiejun Wang, and Holger Klinck. Determina- tion of optimal flight altitude to minimise acoustic drone disturbance to wildlife using species audiograms. Met...

    2021 doi

  17. [25]

    Kerby, Ja- cob M

    Benjamin Koger, Adwait Deshpande, Jeffrey T. Kerby, Ja- cob M. Graving, Blair R. Costelloe, and Iain D. Couzin. Quantifying the movement, behaviour and environmental context of group-living animals using drones and computer vision. Journal of Animal Ecology, 92(7):1357–1371, 2...

    2023

  18. [26]

    Pereira, Nicholas Tabris, Amos Matsliah, Daniel M

    Talmo D. Pereira, Nicholas Tabris, Amos Matsliah, Daniel M. Turner, Jing Li, Shyam Ravindranath, Eleni S. Pa- padoyannis, Elise Normand, David S. Deutsch, Zhong Ying Wang, Georgia C. McKenzie-Smith, Catalin C. Mitelut, Mario D. Castro, Julia D’Uva, Michael Kislin, Dan H. Sanes...

    2022 doi

  19. [27]

    itk-elastix: Medical image registration in python

    Konstantinos Ntatsis, Niels Dekker, Viktor van der Valk, Tom Birdsong, D ˇzenan Zuki ´c, Stefan Klein, Marius Star- ing, and Matthew McCormick. itk-elastix: Medical image registration in python. In Proceedings of the Python in Sci- ence Conference, pages 101–105. SciPy, 1 June...

    2023 doi

  20. [28]

    Shamonin, Esther E

    Denis P. Shamonin, Esther E. Bron, Boudewijn P. Lelieveldt, Marion Smits, Stefan Klein, and Marius Staring. Fast par- allel image registration on cpu and gpu for diagnostic clas- sification of alzheimer’s disease. Frontiers in Neuroinfor- matics, 7, 2014. ISSN 1662-5196. doi: ...

    2014

  21. [29]

    Elastix: a toolbox for intensity-based medical image registration

    Stefan Klein, Marius Staring, Keelin Murphy, Max A Viergever, and Josien PW Pluim. Elastix: a toolbox for intensity-based medical image registration. IEEE transac- tions on medical imaging, 29(1):196–205, 2009. 3

    2009

  22. [30]

    Optimization of mu- tual information for multiresolution image registration.IEEE transactions on image processing, 9(12):2083–2099, 2000

    Philippe Th ´evenaz and Michael Unser. Optimization of mu- tual information for multiresolution image registration.IEEE transactions on image processing, 9(12):2083–2099, 2000. 3

    2000

  23. [31]

    Odm—a command line toolkit to generate maps, point clouds, 3d models and dems from drone, balloon or kite images

    OpenDroneMap Authors. Odm—a command line toolkit to generate maps, point clouds, 3d models and dems from drone, balloon or kite images. OpenDroneMap/ODM GitHub Page, 2020. 3

    2020

  24. [32]

    Luis Ibanez, Bill Lorensen, Matthew McCormick, Brad King, Hans Johnson, Daniel Blezek, Bradley Lowekamp, Julien Jomier, Jim Miller, Ga ¨etan Lehmann, Josh Cates, Niels Dekker, Jon Haitz Legarreta, D ˇzenan Zuki ´c, Lydia Ng, Jisung Kim, Arnaud Gelas, Mathieu Malaterre, Karthik...

    2025

  25. [33]

    Deepforest: A python package for rgb deep learning tree crown delineation

    Ben G Weinstein, Sergio Marconi, M ´elaine Aubry-Kientz, Gregoire Vincent, Henry Senyondo, and Ethan P White. Deepforest: A python package for rgb deep learning tree crown delineation. Methods in Ecology and Evolution , 11 (12):1743–1751, 2020. 3

    2020

  26. [34]

    Individual tree-crown detec- tion in rgb imagery using semi-supervised deep learning neu- ral networks

    Ben G Weinstein, Sergio Marconi, Stephanie Bohlman, Alina Zare, and Ethan White. Individual tree-crown detec- tion in rgb imagery using semi-supervised deep learning neu- ral networks. Remote Sensing, 11(11):1309, 2019. 3

    2019

  27. [35]

    Animating rotation with quaternion curves

    Ken Shoemake. Animating rotation with quaternion curves. In Proceedings of the 12th annual conference on Computer graphics and interactive techniques, pages 245–254, 1985. 3

    1985

  28. [36]

    Peri, Dhruv Sharma, Laura Porta, Iv ´an Varela, and Adam L

    Nikoloz Sirmpilatze, Chang Huan Lo, Sof ´ıa Mi ˜nano, Bran- don D. Peri, Dhruv Sharma, Laura Porta, Iv ´an Varela, and Adam L. Tyson. neuroinformatics-unit/movement: v0.4.0, March 2025. URL https://doi.org/10.5281/ zenodo.15026249. 4

    2025

  29. [37]

    Robots mediating interactions between ani- mals for interspecies collective behaviors

    Franc ¸ois Bonnet, Francesco Mondada, Frank Bonnet, Frank Bonnet, Frank Bonnet, Frank Bonnet, Frank Bonnet, and Frank Bonnet. Robots mediating interactions between ani- mals for interspecies collective behaviors. Science Robotics, 4(30):eaau7897, 2019. doi: 10.1126/scirobotics...

    2019 doi

  30. [38]

    Boxmot: A collection of sota real-time, multi- object trackers for object detectors, June 2023

    Mikel Brostr ¨om. Boxmot: A collection of sota real-time, multi- object trackers for object detectors, June 2023. URL https://doi.org/10.5281/zenodo.8132989. 3

    2023 doi

  31. [41]

    Vggt: Visual geometry grounded transformer

    Jianyuan Wang, Minghao Chen, Nikita Karaev, Andrea Vedaldi, Christian Rupprecht, and David Novotny. Vggt: Visual geometry grounded transformer. arXiv preprint arXiv:2503.11651, 2025. 5 Tracking the Flight: Exploring a Computational Framework for Analyzing Escape Responses in P...

    2025 arXiv

  32. [144]

    URL https://www.sciencedirect.com/ science/article/pii/S0022519396901447. 1

  33. [5193]

    URL https: //www.sciencedirect.com/science/article/ pii/0022519371900610

    doi: 10.1016/0022-5193(71)90061-0. URL https: //www.sciencedirect.com/science/article/ pii/0022519371900610. 1

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.