{"id":"59707cdb-9390-4508-b0ec-2d2d48f2bc38","arxiv_id":"2505.18395","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":3.0,"correctness_risk":"high","formal_verification":"none","parameter_count":1,"one_line_summary":"A new underwater dark-room facility for testing neutrino detector components is described, with a preliminary underwater camera calibration that shows large reprojection errors in parts of the image.","lead":"Researchers at the University of Winnipeg built a dark, water-filled test tank with a motorized 3D gantry for testing cameras and detectors destined for neutrino experiments like Hyper-Kamiokande. They report a first attempt at calibrating an underwater camera's optics, but their own data show large calibration errors in part of the image.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Paper's own Fig. 16/17 show 10–25 px reprojection residuals in the intermediate field, contradicting the conclusion of 'minimal error'; with no RMS/uncertainties or independent ground truth, the central calibration claim is unsupported.","rationale":"The reader's weakest assumption—unverified gantry truth and target rigidity—is real and would matter even if residuals were small; in the present paper, however, the more immediate load-bearing problem is that the reported residuals themselves contradict the headline success claim. I therefore partially agree with the reader's weakest_assumption: their emphasis is on an underlying assumption, while the decisive evidence is the paper's own Fig. 16/17 and §5 text. I credit the facility construction and the gantry calibration plots; those parts are useful. But the central claim—that underwater calibration was successfully accomplished with minimal error—is not supported. The calibration result could potentially be salvaged if the authors quantify residuals, provide uncertainties, justify the 10–25 px residuals, and validate target positions independently; as written, I would keep the reader's REJECT verdict.","tokens_in":6924,"tokens_out":5728,"duration_ms":49422,"concrete_test":"Obtain the raw calibration images and the logged gantry/pan-tilt coordinates (or request a data release), rerun the same OpenCV fisheye calibration independently, and compute the aggregate reprojection error: mean, RMS, median, 95th percentile, and maximum over all detected checkerboard corners, plus the median in each radial bin corresponding to Fig. 17. If the aggregate RMS exceeds 5 px, or if any intermediate radial bin retains a median of 10 px or more, the paper's 'minimal error introduction' conclusion is contradicted.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing failure is internal to the paper. In §5 the authors state that corner reprojection deviations are 'mostly less than 5 pixels at the central and extreme regions... peaking at around 2.5 pixels,' but then state 'many points in the intermediate region tend to have a higher deviation in the range ∼10-25 pixels' (Figs. 16 and 17). The Conclusion nevertheless calls the calibration 'successfully accomplished with minimal error introduction.' These statements cannot both stand without a quantitative error budget. The paper gives no RMS, median, or 95th-percentile reprojection error, no uncertainty on fx, fy, or the principal point, and no number of images/corners used. 'Minimal' is therefore undefined; 10–25 px on a 9504×6336 image is about 0.1–0.26% of the width, and may or may not be acceptable for the intended photogrammetry, but the paper does not argue that it is. If those intermediate residuals are outliers or corrupted frames, the paper must exclude them explicitly and re-report the fit; as written, they are part of the claimed successful calibration. This is a correctness/consistency problem, not a disagreement with consensus. The lack of independent verification of gantry/pan-tilt ground truth makes the claim even harder to assess, because reprojection residuals alone cannot separate camera-model error from target-position error.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper describes the design, construction, and commissioning of the Under-Water Dark-Room Test Facility (UWDTF) at the University of Winnipeg: a 1000-gallon HDPE water tank inside an optically isolated room, a Cartesian gantry with a pan-tilt head providing five axes of motion, a water-circulation and purification system, and auxiliary infrastructure for underwater detector R&D. It also reports a preliminary underwater camera calibration performed with a checkerboard target mounted on the gantry, analyzing images with the OpenCV fisheye-lens model. The authors report focal lengths of fx = 3.292e3 and fy = 3.290e3 pixels, a principal point at (4.662e3, 3.092e3) pixels, and conclude that \"first of its kind\" underwater calibration was \"successfully accomplished with minimal error introduction.\" The facility description is detailed, but the calibration result is presented with only a qualitative and internally inconsistent error statement.","tokens_in":7293,"tokens_out":3206,"duration_ms":25377,"significance":"If the facility performs as described, it provides a useful and relatively low-cost testbed for underwater optical and detector R&D relevant to WCTE, IWCD, and Hyper-Kamiokande. The paper's strengths are the concrete engineering details: tank selection, gantry calibration plots, water-purification loop, dark-room construction, and payload positioning. These are valuable for other groups building similar facilities. The underwater camera calibration, however, is not yet established. The reported reprojection residuals contradict the summary statement, no uncertainty estimates are given for the fitted parameters, and the validation is purely internal to the calibration fit. The central claim of a successful minimal-error underwater calibration therefore needs substantive revision or re-analysis, though the facility description itself could support a publishable paper after major corrections.","major_comments":[{"comment":"The paper states that corner reprojection deviations are \"mostly less than 5 pixels at the central and extreme regions... peaking at around 2.5 pixels,\" then immediately states that \"many points in the intermediate region tend to have a higher deviation in the range ∼10-25 pixels.\" The Conclusion nevertheless calls the calibration \"successfully accomplished with minimal error introduction.\" These statements are contradictory. The manuscript must provide a quantitative error budget: RMS, median, and 95th-percentile reprojection error, the number of images and corners used, and a clear statement of whether the 10–25 pixel deviations are outliers, systematic model mismatch, or a separate population of frames. Without such a budget, \"minimal error\" is undefined, and the central claim is unsupported.","section":"§5, Figs. 16–17, and Conclusion"},{"comment":"The reported intrinsic parameters fx = 3.292e3, fy = 3.290e3, principal point (4.662e3, 3.092e3) are presented without any uncertainties. Since these are fitted values, the claim that the principal point \"matches close to the mid-point\" of the 9504×6336 array cannot be evaluated without standard errors or confidence intervals. The paper should report at least the covariance of the intrinsic parameters or a bootstrap/leave-one-out estimate.","section":"§5, fitted parameters paragraph"},{"comment":"The reprojection error is computed from the same fit that determined the intrinsic and extrinsic parameters, so it measures internal consistency, not absolute calibration accuracy. To support the claim of minimal error introduction, the authors should provide an independent check: for example, holding out a subset of images from the fit and computing reprojection error on those held-out views, or comparing gantry-commanded target positions with positions reconstructed from the calibrated camera. The gantry and pan-tilt calibrations shown in Figs. 3–8 give linear relationships between counts and commanded coordinates, but the paper does not report positioning uncertainties, and the underwater rigidity and flatness of the checkerboard target are not demonstrated. Without such ground truth, the residuals cannot be attributed to the camera model alone.","section":"§5, validation method"},{"comment":"The claim that underwater camera calibration is \"first of its kind\" and \"has never been presented before\" is not supported by a literature survey. Underwater camera calibration for refraction-aware and housing-based systems exists in the computer vision and marine robotics literature. The manuscript should either cite and differentiate from prior underwater calibration work or moderate the novelty claim to \"a calibration of this specific housing and gantry system.\" This is a load-bearing point for the paper's stated significance, though it is secondary to the internal error inconsistency.","section":"Introduction and Conclusion"}],"minor_comments":[{"comment":"The keyword list is empty; the authors should provide standard keywords for indexing.","section":"Abstract/Title"},{"comment":"There are numerous typographical errors, e.g., \"CP volation\" in §1, \"Fortuantely\" in §2, \"di fferent\" in §2, \"purspose\" in §2, \"instrinsic\" in §5, and \"realisized\" in the Acknowledgements. The manuscript needs a careful proofreading pass.","section":"Throughout"},{"comment":"The axis labels in the gantry-coordinate plots are difficult to read and the units are not specified. The captions should state what the coordinate values represent (millimeters? encoder counts?) and the relationship to the plots in Figs. 3–8.","section":"Figures 13 and 15"},{"comment":"The abbreviations RO, UV, and mPMT are used without full expansion at first use (RO is explained only in passing, mPMT is expanded in a footnote, UV is not expanded). Please define all abbreviations at first occurrence.","section":"§4"},{"comment":"The paper mentions that the camera was rotated by 90 degrees after each set of scans and the process repeated four times, but does not state how many images were acquired in total or how many were used in the calibration. This information is needed to assess the reliability of the fit.","section":"§5"}],"recommendation":"major_revision","confidential_remarks":"The paper's facility description is useful and likely within the scope of physics.ins-det, but the calibration section and its conclusion need substantial rework. The internal contradiction in §5 is the main blocker; even if resolved, the missing uncertainty quantification and lack of independent validation will need to be addressed. I recommend the editor ask for a major revision rather than reject, because the facility description is solid and the calibration issue is fixable in principle by re-analysis or by substantially tempering the claims. The \"first of its kind\" novelty claim should be checked carefully by the editor against the broader computer-vision literature on underwater camera calibration."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is a useful facility paper with a calibration section that fails on its own terms. The construction details—tank, gantry, pan-tilt, water purification, dark-room setup—are concrete and will save time for anyone building similar infrastructure for water-Cherenkov R&D. The gantry calibration plots (X/Y/Z/theta/phi) are a nice touch. If you work on WCTE/IWCD/Hyper-K calibration, this is worth a look.\n\nThe problem is the camera calibration. The paper claims 'first of its kind' underwater calibration and says it was 'successfully accomplished with minimal error introduction.' The data in Figs. 16 and 17 contradict that: deviations are mostly under 5 px only in central and extreme regions, while the intermediate region shows 10–25 px residuals. There is no RMS, no median, no 95th percentile, no uncertainty on fx, fy, or the principal point, and no statement of how many images/corners were used. So 'minimal' is undefined, and the internal inconsistency between the text and the figures is not a matter of taste. The conclusion as written does not follow.\n\nThe 'first of its kind' claim is also overreach. Underwater camera calibration has been done before in computer vision and oceanography; the paper cites only aerial calibration literature. The standard OpenCV fisheye routine is an internal consistency check: it reports residuals of the same fit that produced the parameters, not an independent validation. If the gantry positions are not independently verified, the residuals can't separate model error from target positioning error.\n\nThat said, the soft spot is one section, not the whole paper. The facility description is solid and publishable. The calibration section could be fixed by removing the overclaim, reporting a proper error budget, and either excluding or explicitly explaining the intermediate-region outliers. If the authors reanalyze, the facility paper remains valuable.\n\nI'd send this to a referee, not desk-reject: the infrastructure is real and the calibration claim is checkable. But the referee should insist on the revision. I would not cite the calibration result as is.","headline":"A genuinely useful facility paper whose camera-calibration section overclaims and is internally inconsistent; the facility description is the real contribution.","tokens_in":7762,"tokens_out":2602,"would_cite":false,"duration_ms":21244,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A newly built underwater dark-room facility demonstrates that camera intrinsic parameters can be calibrated in water, reporting focal lengths near 3,290 pixels and reprojection errors mostly below five pixels.","keywords":["underwater camera calibration","dark-room test facility","five-axis gantry","camera intrinsic parameters","fisheye lens model","water Cherenkov detector","photogrammetry calibration"],"falsifier":"Independently measure the checkerboard positions underwater with a laser tracker or an external camera system; if that measurement finds target-position errors comparable to or larger than the reported 2.5–25 pixel reprojection deviations, the calibration claim is not separately established.","tokens_in":6714,"feed_emoji":"📷","tokens_out":9805,"duration_ms":84090,"temperature":0.7,"pith_summary":"This paper reports the construction of an optically isolated, water-filled test facility consisting of a 1000-gallon tank, a five-axis gantry, a pan-tilt head, and water purification and temperature-control loops. It then uses the facility to perform what it describes as the first underwater camera calibration, placing a checkerboard target on the gantry arm and scanning it through spherical arcs around a sealed camera at the tank bottom. The calibration returns focal lengths of about $3.292 \\times 10^3$ and $3.290 \\times 10^3$ pixels and a principal point close to the center of the $9504 \\times 6336$ pixel array, with reprojection errors mostly below five pixels. The sympathetic reading is that underwater calibration through a water–dome–air–lens optical path is feasible with minimal introduced error, and that the facility can support photogrammetry development for large water Cherenkov neutrino detectors.","feed_headline":"Underwater camera calibration yields focal lengths near 3,290 pixels","feed_subtitle":"Gantry-guided checkerboard scans in a dark water tank keep reprojection errors mostly under five pixels.","key_machinery":"The load-bearing object is the five-axis motion system: a Cartesian gantry with three stepper-driven axes and a two-motor pan-tilt head on the vertical arm. The gantry moves a flat printed checkerboard through spherical and cylindrical scan patterns around a fixed underwater camera, while the pan-tilt adds polar and azimuthal rotation. Corner positions extracted by a standard computer-vision routine are fitted with a generic fish-eye lens model, and the intrinsic and extrinsic parameters are refined by minimizing reprojection error over many non-repetitive views.","core_discovery":"The central discovery, stated on the paper's own terms, is that a camera's intrinsic parameters—the focal length and the principal point, the image location where the optical axis lands—can be determined while the camera is underwater. The effective optical system includes water, an acrylic dome, a thin air layer, and the camera's own lenses, and the paper reports that this compound system was calibrated by moving a checkerboard with the gantry and fitting a fish-eye lens model. The resulting focal lengths were $f_x = 3.292 \\times 10^3$ and $f_y = 3.290 \\times 10^3$ pixels, with the principal point at $(x = 4.662 \\times 10^3, y = 3.092 \\times 10^3)$ pixels, close to the center of the $9504 \\times 6336$ array. The authors take this near-centered principal point as evidence of a spherically symmetric effective lens system, with reprojection errors mostly under five pixels across the field of view.","pith_inferences":["An inference beyond the paper: because no independent check of the gantry's underwater positioning accuracy is reported, the stated reprojection errors should be read as an upper bound on calibration quality, not separated from positioning error.","A further inference: the observed radial error pattern could be tested by fitting the same images with a model that explicitly includes refraction at the water–dome–air interfaces; a systematic improvement there would point to compound optics rather than target motion as the source of the intermediate-ring deviations.","Another testable extension: the five-axis motion could also be used to calibrate the relative pose of multiple cameras by moving a single target through their overlapping field of view, a next step the paper does not discuss."],"forward_implications":["Underwater camera calibration can be carried out in situ, so the effective lens system that will actually operate in water is calibrated rather than inferred from air measurements.","The near-centered principal point implies that the dome and housing introduce no large decentering distortion, supporting the use of a single rotationally symmetric model.","The same gantry-driven scan-and-fit procedure can be repeated with more views or different target paths, since the calibrated coordinate space already provides the needed geometry.","The elevated 10–25 pixel deviations in an intermediate radial band identify where the current fish-eye model fits least well and where future calibration data should be concentrated."],"supporting_citations":[{"why":"Defines the large water Cherenkov detector whose calibration requirements motivate the underwater facility and the camera-calibration tests.","marker":"[1]"},{"why":"Supplies the generic fish-eye lens model used in the calibration fit; this is the central model whose parameters are being determined.","marker":"[14]"},{"why":"Provides the computer-vision implementation, including corner detection and the calibration routine, that turns the checkerboard images into intrinsic parameters.","marker":"[18]"}],"fun_headline_variants":["Underwater camera calibration pins focal length near 3,290 pixels","Dark tank test yields camera focal lengths around 3,290 pixels","Underwater dark room calibrates cameras with sub-5-pixel error","Camera calibration in dark water tank: focal length ~3,290 pixels"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The gantry and pan-tilt position readouts are treated as true coordinates for the checkerboard, and the board is assumed rigid and flat underwater; if these assumptions fail, the reprojection errors include positioning and shape error and are not purely camera calibration error.","fun_headline_variants_meta":{"raw":{"variants":["Underwater camera calibration pins focal length near 3,290 pixels","Dark tank test yields camera focal lengths around 3,290 pixels","Underwater dark room calibrates cameras with sub-5-pixel error","Camera calibration in dark water tank: focal length ~3,290 pixels"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000854,"raw_usage":{"total_tokens":3742,"prompt_tokens":1006,"completion_tokens":2736,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":622,"completion_tokens_details":{"reasoning_tokens":2659}},"tokens_in":622,"tokens_out":2736,"duration_ms":17260,"temperature":1.0,"reasoning_tokens":2659,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T14:30:56.602234+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Independently measure the checkerboard positions underwater with a laser tracker or an external camera system; if that measurement finds target-position errors comparable to or larger than the reported 2.5–25 pixel reprojection deviations, the calibration claim is not separately established.","supporting_citations":[{"cited_title":"A generic camera model and calibration method for conventional, wide-angle, and fish- eye lenses","cited_arxiv_id":null,"evidence_quote":"Supplies the generic fish-eye lens model used in the calibration fit; this is the central model whose parameters are being determined."},{"cited_title":"The opencv library","cited_arxiv_id":null,"evidence_quote":"Provides the computer-vision implementation, including corner detection and the calibration routine, that turns the checkerboard images into intrinsic parameters."}],"review_version":1}