{"id":"ad2676ef-dff1-4ebc-bd3b-683c8dc6b022","arxiv_id":"2509.22685","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":3,"one_line_summary":"VIRTUS-FPP implements a complete virtual fringe projection profilometry pipeline in NVIDIA Isaac Sim and claims sub-millimeter reconstruction plus digital-twin fidelity to a physical scanner.","lead":"This paper builds a simulated version of a fringe projection 3D scanner entirely inside NVIDIA's Isaac Sim robotics simulator, including virtual calibration and reconstruction. A real scanner was then mirrored in the simulation to compare virtual and physical measurements, with the authors reporting sub-millimeter accuracy.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The digital-twin validation in Section V.D is not an independent prediction: Section VI.B discloses a post-hoc 1.37 scaling factor applied to virtual reconstructions, so the sub-millimeter C2M histograms may reflect a one-parameter fit to the real data rather than simulator fidelity.","rationale":"The reader's weakest assumption identifies the empirically tuned scaling factor and the Z=1 m projector setting as the load-bearing assumptions; my analysis agrees and pinpoints the 1.37 scaling factor as the more serious of the two because it directly affects the interpretation of the digital-twin C2M comparison. The paper is transparent about the scaling factor's existence, and I credit the authors for disclosing it, but the disclosure also reveals that the central sim-to-real fidelity claim is not an independent prediction. The sphere experiment (Section V.A) is a legitimate demonstration of virtual reconstruction accuracy under perfect ground truth, but it does not involve the cross-realm transfer, so it cannot rescue the digital-twin conclusion. The C2M histograms in Figure 13 lack numeric statistics, making it impossible to know whether the 0-1 mm peak reflects typical points or merely a visual mode. If the 1.37 scaling was applied before computing C2M, then the comparison is circular: a global scale fitted to real-world correspondence will trivially reduce C2M distances to the STL, and the residual depends on shape fidelity, not on whether the simulator predicted the real system. The concrete test would settle this by removing the scaling and checking whether the virtual reconstruction still achieves sub-millimeter agreement with the STL. If it does not, the paper should be revised to either derive the scaling factor from a physical model or weaken the digital-twin claim. Since the framework itself remains valuable as a virtual FPP development platform, and the disclosed limitations already led the reader to a CONDITIONAL verdict, I do not recommend changing that verdict; the condition should explicitly include demonstrating the digital-twin correspondence without the fitted 1.37 scaling.","tokens_in":19231,"tokens_out":5134,"duration_ms":44299,"concrete_test":"Reproduce the Section V.D astronaut experiment without applying the 1.37 scaling factor: compute C2M distance between the raw virtual point cloud and the original STL mesh, and plot the histogram alongside the real-system result. Then fit a free isotropic scale s that minimizes C2M error to the STL and report the fitted value and the residual mean/95th percentile. If the unscaled virtual C2M no longer peaks in 0-1 mm and the fitted s is close to 1.37, the digital-twin correspondence is a one-parameter fit, not a prediction. Also recompute both real and virtual C2M histograms with numeric statistics (mean, median, 95th percentile) and state explicitly whether the 1.37 correction was applied in Figure 13.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that VIRTUS-FPP achieves \"digital twin fidelity through sub-millimeter accuracy\" (Section V.D) rests on a global scaling correction that is empirically fitted, not derived from the inverse camera model. Section VI.B states: \"The virtual FPP reconstruction exhibits a consistent empirically calculated scaling factor of 1.37 for accurate real-world correspondence... integrated into the processing workflow.\" The paper does not state whether this 1.37 scale was applied to the virtual astronaut reconstruction before computing the C2M histograms in Figure 13, nor does it report any numeric C2M mean/percentiles. If the factor was applied, the 0-1 mm peak demonstrates that a globally rescaled point cloud can be brought into coarse alignment with a real scan, which is expected for any scale-calibrated reconstruction and is not evidence of physical correspondence. The 0.512 mm sphere result (Section V.A) is a virtual-only self-consistency check and does not involve the 1.37 correction, so it cannot validate sim-to-real transfer. A scale error of 1.37 implies tens of percent dimensional error in the raw virtual reconstruction, far outside the claimed sub-millimeter regime unless corrected. Together with the empirically discovered Z=1 m projector attribute setting (Section V.C), the digital-twin validation appears to depend on two implementation-specific tunings, weakening the claim that the virtual sensor models the physical system from first principles.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript presents VIRTUS-FPP, an end-to-end virtual fringe projection profilometry framework built in NVIDIA Isaac Sim. The framework uses a pinhole camera asset and a rectangular light source with texture-based fringe projection, procedurally generates asymmetric circular calibration boards, performs virtual camera-projector calibration with 18-step phase shifting, reconstructs a sphere and a figurine, tests twelve material/lighting configurations, and constructs a digital twin of a physical FPP system by transferring calibrated camera intrinsics and computing projector dimensions from an inverse pinhole model. The central claimed results are sub-millimeter reconstruction accuracy (0.512 mm radial error on a 50 mm sphere) and digital-twin fidelity shown by cloud-to-mesh (C2M) distance histograms peaking between 0 and 1 mm for an astronaut figurine. The paper also discloses an empirically determined scaling factor of 1.37 for virtual-to-real correspondence and a Z = 1 m reference distance for the Isaac Sim projector light dimensions.","tokens_in":19689,"tokens_out":7323,"duration_ms":64408,"significance":"FPP simulation in Isaac Sim with end-to-end calibration and reconstruction is a timely contribution for synthetic data generation, rapid prototyping, and sim-to-real transfer in robotics and optical metrology. The standard FPP equations in Section II-A are mathematically correct, the virtual calibration reprojection errors are reasonable, and the procedural calibration board generation and high-throughput data acquisition are practical strengths. The paper is also unusually candid in its limitations discussion, explicitly disclosing rendering noise and the 1.37 scaling correction. However, the headline numerical claims rest on a single sphere fit with no uncertainty quantification and on C2M histograms that are not reported numerically, and the digital-twin comparison depends on two empirically tuned settings (the Z = 1 m reference and the 1.37 scale). If the required statistics, raw-versus-scaled analyses, and multi-distance validation are added, the results could justify the claimed sub-millimeter fidelity; as currently presented, the paper overstates the strength of its validation.","major_comments":[{"comment":"The digital-twin fidelity claim is not an independent validation. Section VI.B discloses that virtual FPP reconstructions \"exhibit a consistent empirically calculated scaling factor of 1.37 for accurate real-world correspondence,\" but Section V.D does not state whether this scale was applied before computing the C2M histograms in Figure 13, and no numeric C2M mean, RMSE, or percentiles are given. A global rescale is a one-parameter fit that can bring any similarly shaped reconstruction into coarse alignment, so histograms peaking at 0-1 mm do not by themselves demonstrate simulator fidelity. Please report C2M statistics for both scaled and unscaled virtual reconstructions, and specify the exact role of the 1.37 factor in the reported comparison.","section":"V.D and VI.B"},{"comment":"The projector model is validated with an implementation-specific convention rather than a derived physical property. The paper states that correct projected dimensions at 0.4 m are obtained by setting Isaac Sim's RectLight width and height to the inverse-camera-model values computed at Z = 1 m, and attributes this to a presumed internal reference distance in RectLight. This, together with the 1.37 scale correction, means the claimed sim-to-real correspondence depends on two empirically tuned parameters. Please validate the scaling relation over multiple distances, interrogate the RectLight implementation for the reference-distance behavior, or explicitly reclassify both settings as calibration parameters of the simulator; as written, the claim that projector intrinsics are established through theoretical formulation rather than empirical transfer is not fully supported.","section":"V.C, Eq. (22)"},{"comment":"The sub-millimeter accuracy claim in Eq. (14) is based on one MSAC radius fit (50.512 mm, 133,802/134,512 inliers) with no repeated trials, no standard deviation, and no variation of pose, lighting, or reconstruction parameters. Because the sphere experiment is also entirely virtual and does not involve the 1.37 scale correction, it cannot validate sim-to-real transfer. Please provide multiple trials with error bars and, if possible, a separate validation on a non-spherical target.","section":"V.A, Eq. (14)"},{"comment":"The calibration-board feature dimensions are reported to match intended values within 7% error (1-2 mm difference), which is large relative to the claimed sub-millimeter reconstruction accuracy. This apparent inconsistency should be resolved: either the 7% error is a misstatement or the calibration is insensitive to these dimension errors, and that insensitivity should be demonstrated. The exact source of the error (for example, texture-mapping scaling in USD) should also be quantified, since calibration-board geometry underpins all downstream accuracy claims.","section":"IV.A, Eqs. (12)-(13)"},{"comment":"The adverse-conditions testing section is qualitative only: no RMSE, phase error, or reconstruction error is reported for the twelve configurations in Figure 8. The conclusion that the simulator enables robust FPP development under difficult conditions is not supported by numerical evidence. Please add quantitative metrics for each material/lighting combination.","section":"V.B"}],"minor_comments":[{"comment":"The dimensions 202.7 and 323.3 in Eq. (25) are labeled \"expressed in meters,\" which is inconsistent with the physical measurements shown in Figure 12 in millimeters; if these are millimeters, correct the units.","section":"V.C, Eq. (25)"},{"comment":"The C2M distance histograms lack axis labels and numeric bin counts; add these so the \"peaking between 0-1 mm\" claim can be verified by the reader.","section":"Figure 13"},{"comment":"There are typographical errors such as \"syntheitc\" and \"dimesnions\"; a careful proofreading pass is needed, although these do not affect the technical content.","section":"VI.B"},{"comment":"The \"first\" claims in the abstract and Section III should be positioned against the existing Isaac Sim structured-light work in [13] and the ray-tracing FPP simulators in [48]; state explicitly which capabilities are new relative to these references.","section":"III"},{"comment":"Table I reports the camera focal length as 50 cm while Eqs. (15)-(17) convert between millimeters and pixels; specify the units consistently and report the focal length as either 500 mm or in pixels.","section":"Table I"},{"comment":"The throughput claim of 3 FPS should define whether FPS refers to fringe-image capture rate or completed 3D reconstruction rate, since the text reports approximately 10,530 fringe captures per hour, or about 2.9 captures per second, not full 3D scans per second.","section":"IV.B"}],"recommendation":"major_revision","confidential_remarks":"Please verify the novelty claim relative to reference [13], which is described as physically based structured light synthetic data simulation in Isaac Sim at ICRA 2024; the manuscript should make the difference unambiguous. In addition, the abstract's statement that the framework operates \"without dependence on pre-calibrated physical systems\" is in tension with the digital-twin experiment, which transfers a physical calibration and adds an empirical scale factor. The authors should be asked to align the abstract and conclusion with the actual validation protocol."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: VIRTUS-FPP is a genuinely useful engineering contribution — a complete virtual FPP pipeline in Isaac Sim with virtual calibration, projector intrinsics from the inverse camera model, and a digital-twin comparison against a real system. The math is standard and correctly laid out. The paper is also unusually candid about its own limitations, which counts for something. But the strongest accuracy claims are softer than the abstract suggests.\n\nWhat's new: the combination of end-to-end virtual calibration with the inverse-camera-model projector representation in Isaac Sim, plus the digital twin validation. Blender/Unity/MATLAB simulators exist [46,47,49], and [13] already did physically-based structured light synthetic data, so 'first' should be qualified. Still, the specific framework is a real implementation with a working pipeline. The 0.512 mm sphere error is a reasonable self-consistency result for the virtual system.\n\nSoft spots: First, the sphere fit is one MSAC run on one sphere — no repeated trials, no spread. Second, the C2M histograms in Figure 13 are shown without numeric RMSE/percentiles, so 'peaking between 0-1 mm' is under-quantified. Third, and most importantly, the digital-twin validation is not an independent test. Section VI.B discloses a 1.37 empirically fitted scaling factor applied to virtual reconstructions, and Section V.C discloses that the projector width/height had to be set to the inverse-camera-model values at Z=1 m rather than the actual screen distance to get correct dimensions. If the 1.37 factor was applied before the C2M comparison, then the '0-1 mm' peak is a one-parameter fit to the real data — good for alignment, not evidence of physical fidelity. The paper doesn't say whether it was applied. That needs to be stated explicitly, and ideally an ablation without the scale factor.\n\nI don't think this is fatal. For an engineering paper, empirical tuning of a simulator isn't disqualifying, especially when disclosed. But the claims should be reworded: 'digital twin fidelity through sub-millimeter accuracy' overstates what is demonstrated. The framework is still valuable as a testbed for calibration studies, synthetic data generation, and sim-to-real transfer experiments.\n\nWho it's for: people working in FPP calibration, structured-light sim2real, and robotics perception who want a configurable virtual sensor. They'll get a useful starting point even with the caveats.\n\nRecommendation: send it to peer review, but require (1) code/data release, (2) numeric C2M statistics from multiple trials, (3) explicit statement of whether and how the 1.37 scale was applied, and (4) a sensitivity analysis on the Z=1 m projector setting. With those, the accuracy claims become testable. Without them, the headline claims should be softened.","headline":"Useful, honest Isaac Sim FPP testbed, but the digital-twin 'sub-millimeter' claims hang on two empirically tuned knobs (Z=1 m projector dimensions and a 1.37 scale factor) that need to be reported and ablated before the accuracy headline holds.","tokens_in":20067,"tokens_out":3155,"would_cite":true,"duration_ms":24976,"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":"VIRTUS-FPP claims to be the first end-to-end virtual fringe projection profilometry framework in NVIDIA Isaac Sim, achieving sub-millimeter reconstruction accuracy and a digital twin of a physical FPP system.","keywords":["fringe projection profilometry","virtual sensor modeling","NVIDIA Isaac Sim","structured light","digital twin","inverse camera model","synthetic data generation","3D reconstruction"],"falsifier":"Measure the projected fringe dimensions in Isaac Sim at several screen distances for a projector whose intrinsic matrix is known and entered at a reference distance other than 1 m (e.g., set width/height to the Eq. (22) values at Z = 0.4 m); if the on-screen size at 0.4 m no longer matches the physical measurements (about 202.7 mm × 323.3 mm), then the 1 m reference-distance rule is an artifact of that renderer rather than a geometric property of the inverse camera model. As a complementary check, recompute virtual reconstructions without the 1.37 factor and measure the cloud-to-mesh error against the real scan: if the error is unaffected, the factor is cosmetic; if the error degrades sharply, the claimed correspondence depends on a fitted constant rather than on the modeling alone.","tokens_in":19014,"feed_emoji":"📐","tokens_out":10258,"duration_ms":80896,"temperature":0.7,"pith_summary":"VIRTUS-FPP tries to establish that fringe projection profilometry—a structured-light 3D scanning method that projects sinusoidal fringe patterns and turns their deformations into depth maps—can be run end to end inside a physics-based simulator with sub-millimeter accuracy, eliminating the need for a pre-calibrated physical system during algorithm or sensor development. The paper's central technical move is to model the projector with an inverse pinhole camera model, so the metric dimensions of the projected pattern follow from the system's calibration matrices instead of being transferred empirically from real hardware. If this is right, researchers could prototype scanner configurations, generate perfectly labeled synthetic fringe data, and build digital twins of real FPP scanners in hours rather than recalibrating hardware for every change. The authors support the claim with virtual calibration errors of 0.0555 (stereo reprojection) and 0.0486 (projector), a 0.512 mm radial error on a 50 mm sphere, and cloud-to-mesh distance histograms peaking in the 0–1 mm range when virtual and physical reconstructions are compared.","feed_headline":"Virtual fringe projection hits sub-millimeter accuracy in simulation","feed_subtitle":"A full simulator-to-reality digital twin reproduces a real scanner's measurements with sub-millimeter error.","key_machinery":"The load-bearing object is the inverse camera model of the projector, written as $$[x_w, y_w, 1]^T = Z\\big((M_{\\mathrm{ext}})^+ (M_{\\mathrm{int}})^{-1}[u, v, 1]^T\\big)$$, where $(M_{\\mathrm{ext}})^+$ is the Moore–Penrose pseudo-inverse of the projector's 3x4 extrinsic matrix and $Z$ is the depth factored out by the similar-triangle constraint of pinhole projection. This identity converts a projector pixel coordinate $(u,v)$ into a metric world point along the projection ray, so the width and height of the projected fringe pattern at any working distance can be read off the corner points. It is what lets a real system's calibration matrices set the rectangular light source's physical dimensions in simulation, and the paper validates it both inside the simulator (measured vs. theoretical pattern sizes at 400–1000 mm) and against the physical scanner. The rest of the pipeline—18-step phase shifting, temporal phase unwrapping with Gray coding, and an 18-pose calibration on procedurally generated asymmetric circular boards—is the standard FPP processing chain adapted to the virtual scene.","core_discovery":"The central claim, as the authors state it, is that a rectangular light source in NVIDIA Isaac Sim can serve as a faithful digital light projector once its width and height are computed from the physical projector's calibration matrices through the inverse pinhole camera model. Concretely, they take the projector's intrinsic and extrinsic matrices, invert the intrinsic, take the Moore–Penrose pseudo-inverse of the 3x4 extrinsic, and factor out the depth $Z$ using the similar-triangle scaling that is inherent to pinhole geometry; the resulting world coordinates of the pattern corners give the metric dimensions to enter into the light source. Setting those dimensions at a reference distance of $Z = 1$ m produces projected fringe sizes that track the measured values with mean absolute errors of 1.70 mm in width and 1.42 mm in height over distances from 400 to 1000 mm. The same procedure applied to a real FPP system yields a digital twin whose reconstructions of the same object have cloud-to-mesh distance distributions peaking between 0 and 1 mm, matching the physical system; the authors note that a consistent empirically determined factor of 1.37 must be applied to virtual reconstructions to achieve this match, which they attribute to the simulation environment's optical modeling.","pith_inferences":["Beyond the paper's claims, the 1-meter reference-distance behavior of the rectangular light source looks like a renderer convention rather than a consequence of pinhole optics; if so, the inverse-camera-model step is portable to any simulator that normalizes projected texture size at a fixed reference plane, and the width/height computation becomes a documented platform setting rather than a physi","A natural next test the paper motivates but does not run is to train a phase-unwrapping or reconstruction network exclusively on VIRTUS-FPP data and measure how well it transfers to real fringe images; the simulator's ground-truth annotations are exactly the resource such training needs.","The fixed 1.37 scaling factor should be probed across multiple baselines, working distances, and object sizes: if it remains constant it is a global unit-convention correction, but if it drifts the digital twin needs a distance-dependent or geometry-dependent model rather than one post-hoc multiplier.","Since the digital-twin validation covers a single screen distance and a single target object, the paper's own virtual-versus-theoretical dimension test across 400–1000 mm provides a ready template for a stronger multi-distance, multi-geometry validation of the twin."],"forward_implications":["FPP system configurations can be prototyped and fully calibrated in simulation before any hardware is built, since the virtual calibration achieves stereo reprojection error 0.0555 and projector error 0.0486.","Synthetic fringe datasets with exact ground truth can be produced at scale: 936 captured images for an 18-pose calibration in about five minutes, roughly 10,530 captures per hour at 3 FPS acquisition.","A calibrated real-world FPP system can be replicated as a digital twin from its calibration matrices alone, so measurement campaigns can continue in simulation when the physical scanner is unavailable or being reconfigured.","Domain randomization over ambient lighting and material properties becomes a routine experiment: the paper shows 12 material-lighting combinations degrading fringe visibility in the expected way, which is impractical to test systematically on real hardware.","The 0.512 mm radial error on a 50 mm sphere places the virtual system in the same accuracy class as physical structured-light scanners, making simulation a credible testbed for precision metrology development."],"supporting_citations":[{"why":"Supplies the phase-shifting FPP foundation and the sub-millimeter accuracy claim that motivates treating simulation as a metrology tool.","marker":"[1]"},{"why":"Provides the gray-code plus phase-shift temporal unwrapping method used to convert wrapped phase maps to unwrapped ones in the virtual pipeline.","marker":"[7]"},{"why":"The prior digital-twin FPP simulator that VIRTUS-FPP extends and claims to outperform in acquisition speed; its 7200-image/1.5-hour baseline is the comparison point for the 3 FPS claim.","marker":"[46]"},{"why":"The platform reference that documents NVIDIA Isaac Sim's rendering, PhysX, and USD capabilities the framework relies on.","marker":"[53]"},{"why":"MLESAC is the robust sphere-fitting routine used to measure the 50.512 mm fitted radius and hence the 0.512 mm radial error.","marker":"[66]"},{"why":"Provides the formulas converting calibrated camera intrinsics into Isaac Sim focal length and aperture settings for the digital twin.","marker":"[67]"},{"why":"The inverse camera calibration method from which the projector's metric dimensions are derived.","marker":"[68]"},{"why":"The pinhole-camera similar-triangle constraint used to resolve the pseudo-inverse's depth ambiguity by parameterizing world coordinates as Z times a fixed ray.","marker":"[69]"},{"why":"The C2M distance methodology used to compare both real and virtual reconstructions to the reference mesh and to claim sub-millimeter digital-twin fidelity.","marker":"[70]"}],"fun_headline_variants":["Simulated fringe projection matches real to sub-mm","Digital twin for fringe projection hits sub-mm in Isaac Sim","First virtual FPP sensor in Isaac Sim equals real scans","Sub-mm reconstruction accuracy from simulated FPP","Isaac Sim enables high-fidelity fringe projection twin"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the rectangular light source in the simulator obeys the inverse-camera-model geometry set at Z = 1 m and that the resulting reconstructions only need one fixed 1.37 scale factor to match physical measurements; if either is an implementation artifact, the claimed digital-twin fidelity is a fitted coincidence rather than a geometric prediction.","fun_headline_variants_meta":{"raw":{"variants":["Simulated fringe projection matches real to sub-mm","Digital twin for fringe projection hits sub-mm in Isaac Sim","First virtual FPP sensor in Isaac Sim equals real scans","Sub-mm reconstruction accuracy from simulated FPP","Isaac Sim enables high-fidelity fringe projection twin"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000378,"raw_usage":{"total_tokens":2061,"prompt_tokens":1047,"completion_tokens":1014,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":663,"completion_tokens_details":{"reasoning_tokens":939}},"tokens_in":663,"tokens_out":1014,"duration_ms":8053,"temperature":1.0,"reasoning_tokens":939,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T15:50:11.773424+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the projected fringe dimensions in Isaac Sim at several screen distances for a projector whose intrinsic matrix is known and entered at a reference distance other than 1 m (e.g., set width/height to the Eq. (22) values at Z = 0.4 m); if the on-screen size at 0.4 m no longer matches the physical measurements (about 202.7 mm × 323.3 mm), then the 1 m reference-distance rule is an artifact of that renderer rather than a geometric property of the inverse camera model. As a complementary check, recompute virtual reconstructions without the 1.37 factor and measure the cloud-to-mesh error against the real scan: if the error is unaffected, the factor is cosmetic; if the error degrades sharply, the claimed correspondence depends on a fitted constant rather than on the modeling alone.","supporting_citations":[{"cited_title":"Zhang,High-Speed 3D Imaging with Digital Fringe Projection Techniques","cited_arxiv_id":null,"evidence_quote":"Supplies the phase-shifting FPP foundation and the sub-millimeter accuracy claim that motivates treating simulation as a metrology tool."},{"cited_title":"Three-dimensional vision based on a combination of gray-code and phase-shift light projection: analysis and compensation of the systematic errors,","cited_arxiv_id":null,"evidence_quote":"Provides the gray-code plus phase-shift temporal unwrapping method used to convert wrapped phase maps to unwrapped ones in the virtual pipeline."},{"cited_title":"Fringe projection profilometry by conducting deep learning from its digital twin,","cited_arxiv_id":null,"evidence_quote":"The prior digital-twin FPP simulator that VIRTUS-FPP extends and claims to outperform in acquisition speed; its 7200-image/1.5-hour baseline is the comparison point for the 3 FPS claim."},{"cited_title":"What is isaac sim?","cited_arxiv_id":null,"evidence_quote":"The platform reference that documents NVIDIA Isaac Sim's rendering, PhysX, and USD capabilities the framework relies on."},{"cited_title":"Mlesac: A new robust estimator with application to estimating image geometry,","cited_arxiv_id":null,"evidence_quote":"MLESAC is the robust sphere-fitting routine used to measure the 50.512 mm fitted radius and hence the 0.512 mm radial error."},{"cited_title":"Camera sensors — isaac sim documentation","cited_arxiv_id":null,"evidence_quote":"Provides the formulas converting calibrated camera intrinsics into Isaac Sim focal length and aperture settings for the digital twin."},{"cited_title":"Projector calibration by","cited_arxiv_id":null,"evidence_quote":"The inverse camera calibration method from which the projector's metric dimensions are derived."},{"cited_title":"Camera models,","cited_arxiv_id":null,"evidence_quote":"The pinhole-camera similar-triangle constraint used to resolve the pseudo-inverse's depth ambiguity by parameterizing world coordinates as Z times a fixed ray."},{"cited_title":"Characterizing the 3-dimensional printability of alginate–gelatin and nanocellulose gels via fringe projection,","cited_arxiv_id":null,"evidence_quote":"The C2M distance methodology used to compare both real and virtual reconstructions to the reference mesh and to claim sub-millimeter digital-twin fidelity."}],"review_version":2}