Pith. sign in

REVIEW 2 major objections 2 minor 20 references

ISOPoT adapts modern point tracking to sonar images to produce reliable underwater odometry that beats prior methods on real datasets.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.3

2026-06-26 08:46 UTC pith:RONFJGTB

load-bearing objection ISOPoT adapts camera point trackers to sonar with multi-frame tracks and reports consistent odometry gains on two real datasets. the 2 major comments →

arxiv 2606.23006 v1 pith:RONFJGTB submitted 2026-06-22 cs.RO

ISOPoT: Imaging Sonar Odometry by Point Tracking

Ja\v{s}a Samec , Vid Rijavec , Marko Peljhan , Aleksander Grm , Andrej Androjna , Danijel Sko\v{c}aj , Matej Dobrevski This is my paper
classification cs.RO
keywords imaging sonarodometrypoint trackingunderwater navigationmarine roboticsforward-looking sonarpose estimationsensor fusion
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper presents ISOPoT as a sonar odometry pipeline that treats multi-frame point tracks as the main way to match features across images. Because sonar images are noisy and lack the clear structure of camera photos, the authors add only lightweight adjustments to make standard point trackers work. Tests on the Aracati 2017 dataset and a separate real-world collection show the method delivers lower error than earlier techniques both when using sonar alone and when fused with other sensors. A reader would care because underwater robots need accurate motion estimates in conditions where cameras and GPS fail.

Core claim

ISOPoT is an imaging sonar odometry method whose core representation is multi-frame point tracks obtained from modern trackers, augmented by lightweight optimizations that improve robustness to sonar noise and artifacts; this pipeline yields lower trajectory error than previous state-of-the-art approaches on the Aracati 2017 dataset and an internal real-world sonar dataset, both in sonar-only and multi-sensor configurations.

What carries the argument

Multi-frame point tracks as the primary correspondence representation, augmented with lightweight optimizations for sonar imagery.

Load-bearing premise

Point-tracking techniques developed for ordinary camera images can be transferred to sonar with only lightweight optimizations and will still produce usable tracks despite noise, artifacts, and missing semantic structure.

What would settle it

A controlled experiment in which ISOPoT produces higher trajectory error than the previous best method on the Aracati 2017 dataset or the internal sonar dataset would falsify the performance claim.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Sonar-only navigation becomes more accurate without requiring rich semantic features in the images.
  • Multi-sensor fusion pipelines that include sonar gain a stronger odometry component.
  • The same tracking backbone can be reused across different forward-looking sonar hardware and environments.
  • Odometry estimation no longer depends on hand-crafted sonar-specific feature detectors.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Similar lightweight transfers of camera point trackers could be tried on other low-structure sensors such as radar or thermal imagers.
  • If the optimizations remain minimal, existing visual-odometry codebases might incorporate sonar tracks with little extra engineering.
  • Longer track lengths could further stabilize estimates in very turbid water where frame-to-frame matches are sparse.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The paper introduces ISOPoT, an imaging sonar odometry pipeline that adapts modern multi-frame point tracking techniques from computer vision with lightweight optimizations to handle sonar noise, artifacts, and low semantic content. It claims consistent outperformance versus prior state-of-the-art methods on the Aracati 2017 dataset and an internal real-world sonar dataset, in both sonar-only and multi-sensor fusion settings, supported by quantitative ATE/RPE results and qualitative track visualizations.

Significance. If the reported outperformance holds under scrutiny, the work would represent a meaningful advance in underwater robotics by demonstrating that camera-derived point trackers can be transferred to forward-looking sonar with modest changes, enabling more reliable long-range odometry in turbid conditions where traditional keypoint methods fail.

major comments (2)
  1. [Evaluation] Evaluation section: the manuscript supplies ATE/RPE tables showing gains but provides no ablation studies isolating the contribution of the multi-frame tracking component versus single-frame baselines, nor error bars or statistical tests on the metrics; this makes it impossible to verify whether the data support the headline outperformance claim over SOTA.
  2. [Method] Method section: the claim that only lightweight optimizations suffice to produce usable long tracks despite speckle, shadows, and lack of semantic structure is central to the pipeline, yet the description does not include quantitative analysis of track length, accuracy, or failure modes on sonar imagery to substantiate transferability from camera-based trackers.
minor comments (2)
  1. The abstract would be strengthened by including one or two concrete quantitative results (e.g., average ATE improvement) rather than the qualitative statement of 'consistent outperformance'.
  2. Dataset details such as sequence lengths, sensor specifications, and ground-truth acquisition method for the internal collection are referenced but could be expanded for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address the two major comments below and will revise the manuscript accordingly to strengthen the evaluation and method sections.

read point-by-point responses

  1. Referee: [Evaluation] Evaluation section: the manuscript supplies ATE/RPE tables showing gains but provides no ablation studies isolating the contribution of the multi-frame tracking component versus single-frame baselines, nor error bars or statistical tests on the metrics; this makes it impossible to verify whether the data support the headline outperformance claim over SOTA.

    Authors: We agree that the current evaluation would be strengthened by explicit ablations and statistical analysis. In the revised manuscript we will add ablation studies that isolate the multi-frame point tracking component against single-frame baselines, include error bars on the ATE/RPE metrics, and report statistical tests to better substantiate the outperformance claims. revision: yes

  2. Referee: [Method] Method section: the claim that only lightweight optimizations suffice to produce usable long tracks despite speckle, shadows, and lack of semantic structure is central to the pipeline, yet the description does not include quantitative analysis of track length, accuracy, or failure modes on sonar imagery to substantiate transferability from camera-based trackers.

    Authors: We acknowledge that quantitative characterization of the point tracks would better support the transferability argument. In the revision we will add a quantitative analysis subsection reporting track lengths, accuracy, and observed failure modes on the sonar datasets to substantiate the effectiveness of the lightweight optimizations. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper introduces ISOPoT as an engineering pipeline adapting modern point-tracking methods to sonar imagery with lightweight optimizations, then evaluates it empirically on the Aracati 2017 dataset and an internal real-world collection. No mathematical derivations, fitted parameters renamed as predictions, self-definitional equations, or load-bearing self-citations appear in the abstract or described claims. Performance results (ATE/RPE) are presented as direct experimental outcomes against external benchmarks rather than reductions to the method's own inputs. The central claim therefore remains self-contained and falsifiable via the reported dataset comparisons.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no equations, parameters, or modeling choices; the ledger is therefore empty.

pith-pipeline@v0.9.1-grok · 5716 in / 1054 out tokens · 17671 ms · 2026-06-26T08:46:19.350892+00:00 · methodology

0 comments
read the original abstract

Reliable navigation in underwater environments remains a key challenge in marine robotics. In such scenarios, forward-looking sonars are a natural choice for long-range perception, offering wide coverage even in turbid, low-visibility conditions. However, sonar images are inherently noisy, contain artifacts, and lack rich semantic structure, causing standard computer vision methods for keypoint detection and matching to perform poorly. In this paper, we introduce ISOPoT, an imaging sonar odometry method based on modern point tracking techniques. We propose a sonar odometry pipeline that uses multi-frame point tracks as its primary correspondence representation, augmented with lightweight optimizations to improve robustness. We evaluated the proposed method on the Aracati 2017 dataset, as well as on an internal sonar dataset collected in real-world underwater environments. Our results show that ISOPoT outperforms previous state-of-the-art methods consistently in both sonar-only scenarios and in multi-sensor settings.

Figures

Figures reproduced from arXiv: 2606.23006 by Aleksander Grm, Andrej Androjna, Danijel Sko\v{c}aj, Ja\v{s}a Samec, Marko Peljhan, Matej Dobrevski, Vid Rijavec.

Figure 1
Figure 1. Figure 1: Overview of the proposed ISOPoT pipeline. Given a [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Image formation for a single sonar beam. For a fixed [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of ISOPoT. A sliding window containing [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Match refinement. Query points from the first frame [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Typical failure modes in the Aracati 2017 dataset. [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Example sonar image from the Portoroz 2025 dataset. [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative comparison on Aracati 2017. ISOPoT without correction follows the overall trajectory but accumulates [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative comparison of DISO and ISOPoT on the [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative comparison of point trackers on Por [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

20 extracted references · 2 canonical work pages

  1. [1]

    S. Gode, A. Hinduja, and M. Kaess, ”SONIC: Sonar Image Correspon- dence using Pose Supervised Learning for Imaging Sonars,” inProc. IEEE Intl. Conf. on Robotics and Automation (ICRA), Yokohama, Japan, May 2024

    2024

  2. [2]

    S. Xu, K. Zhang, Z. Hong, Y . Liu, and S. Wang, ”DISO: Direct Imaging Sonar Odometry,” in2024 IEEE Intl. Conf. on Robotics and Automation (ICRA), 2024, pp. 8573–8579

    2024

  3. [3]

    A. W. Harley et al., ”AllTracker: Efficient Dense Point Tracking at High Resolution,” inICCV, 2025

    2025

  4. [4]

    Karaev, Y

    N. Karaev, Y . Makarov, J. Wang, N. Neverova, A. Vedaldi, and C. Rupprecht, ”CoTracker3: Simpler and Better Point Tracking by Pseudo-Labeling Real Videos,” inProc. IEEE/CVF Intl. Conf. on Computer Vision (ICCV), 2025, pp. 1–10

    2025

  5. [5]

    Zholus et al., ”TAPNext: Tracking Any Point (TAP) as Next Token Prediction,” inProc

    A. Zholus et al., ”TAPNext: Tracking Any Point (TAP) as Next Token Prediction,” inProc. IEEE/CVF Intl. Conf. on Computer Vision (ICCV), Oct. 2025, pp. 9693–9703

    2025

  6. [6]

    M. M. dos Santos, ”Aracati 2017,” 2017. [Online]. Available: https://github.com/matheusbg8/aracati2017

    2017

  7. [7]

    J. Wang, F. Chen, Y . Huang, J. McConnell, T. Shan, and B. Englot, ”Virtual Maps for Autonomous Exploration of Cluttered Underwater Environments,”IEEE Journal of Oceanic Engineering, vol. 47, no. 4, pp. 916–935, 2022

    2022

  8. [8]

    J. Li, M. Kaess, R. M. Eustice, and M. Johnson-Roberson, ”Pose- Graph SLAM Using Forward-Looking Sonar,”IEEE Robotics and Automation Letters, vol. 3, no. 3, pp. 2330–2337, 2018

    2018

  9. [9]

    C. Lei, H. Rajani, N. Gracias, R. Garcia, and H. Wang, ”A Geometri- cally Consistent Matching Framework for Side-Scan Sonar Mapping,” arXiv:2509.11255, 2025

    work page arXiv 2025

  10. [10]

    Doersch et al., ”TAP-Vid: A Benchmark for Tracking Any Point in a Video,”Advances in Neural Information Processing Systems, vol

    C. Doersch et al., ”TAP-Vid: A Benchmark for Tracking Any Point in a Video,”Advances in Neural Information Processing Systems, vol. 35, pp. 13610–13626, 2022

    2022

  11. [11]

    A. W. Harley, Z. Fang, and K. Fragkiadaki, ”Particle Video Revisited: Tracking Through Occlusions Using Point Trajectories,” inProc. ECCV, 2022

    2022

  12. [12]

    Bian et al., ”Context-PIPs: Persistent Independent Particles De- mands Context Features,” inProc

    W. Bian et al., ”Context-PIPs: Persistent Independent Particles De- mands Context Features,” inProc. NeurIPS, 2023

    2023

  13. [13]

    Doersch et al., ”TAPIR: Tracking any point with per-frame initial- ization and temporal refinement,” inProc

    C. Doersch et al., ”TAPIR: Tracking any point with per-frame initial- ization and temporal refinement,” inProc. IEEE/CVF Intl. Conf. on Computer Vision (ICCV), 2023, pp. 10061–10072

    2023

  14. [14]

    Karaev, I

    N. Karaev, I. Rocco, B. Graham, N. Neverova, A. Vedaldi, and C. Rupprecht, ”CoTracker: It is Better to Track Together,” inProc. ECCV, 2024

    2024

  15. [15]

    Alcantarilla, J

    P. Alcantarilla, J. Nuevo, and A. Bartoli, ”Fast Explicit Diffusion for Accelerated Features in Nonlinear Scale Spaces,” inProc. British Machine Vision Conference (BMVC), 2013

    2013

  16. [16]

    DeTone, T

    D. DeTone, T. Malisiewicz, and A. Rabinovich, ”SuperPoint: Self- Supervised Interest Point Detection and Description,” in2018 IEEE/CVF CVPR Workshops (CVPRW), 2018, pp. 337–33712

    2018

  17. [17]

    K. He, X. Zhang, S. Ren, and J. Sun, ”Deep Residual Learning for Image Recognition,” in2016 IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770–778

    2016

  18. [18]

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, ”Im- ageNet: A Large-Scale Hierarchical Image Database,” inProc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2009, pp. 248–255

    2009

  19. [19]

    Zhang and D

    Z. Zhang and D. Scaramuzza, ”A Tutorial on Quantitative Trajectory Evaluation for Visual(-Inertial) Odometry,” in2018 IEEE/RSJ Intl. Conf. on Intelligent Robots and Systems (IROS), 2018, pp. 7244–7251

    2018

  20. [20]

    Aydemir, W

    G. Aydemir, W. Xie, and F. G ¨uney, ”Track-On2: Enhancing On- line Point Tracking with Memory,”IEEE Transactions on Pat- tern Analysis and Machine Intelligence, pp. 1–17, 2026, doi: 10.1109/TPAMI.2026.3675257

    work page doi:10.1109/tpami.2026.3675257 2026