{"id":"5ee82a94-c195-42b9-949a-151d7441c5cd","arxiv_id":"2411.16946","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":4,"one_line_summary":"The paper specifies a two-texture STMap encoding (View Map and Footage Map) that maps footage through a common spherical coordinate space to animate and interchange lens distortion profiles.","lead":"This paper introduces LDES, a proposed file format for storing lens distortion as STMap textures so footage shot on one lens can be warped to look like it was shot on another. It is aimed at film and VFX post-production, with the promise of animatable transitions between lens looks, similar to how ACES standardizes color.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Calibration accuracy is the load-bearing unvalidated link: the paper's distortion-accurate interchange claim requires real lenses to be encoded into accurate View/Footage Maps, but Section 2.3's methods are never evaluated and Section 6 defers them.","rationale":"The reader's weakest-assumption analysis identifies the same load-bearing risk: the accuracy of real-lens calibration is asserted, not demonstrated, and the paper's own Future Work section confirms this. I found no independent internal inconsistency in the map-composition mathematics; Eq. (4) is a standard composition of two coordinate textures, and the FOV scaling in Eqs. (3) and (5) is coherent if both maps live in the same equidistant angular space. The proposal is plausible as a design document, but its headline promise of distortion accuracy depends entirely on empirical constants that are currently unmeasured. Therefore the CONDITIONAL verdict is appropriate: acceptance should be tied to a concrete calibration evaluation. My read does not move the reader's verdict, so I recommend no change.","tokens_in":9859,"tokens_out":5681,"duration_ms":58110,"concrete_test":"Acquire a lens with a high-accuracy reference calibration (e.g., a precision test chart or factory distortion data), run both Section 2.3 calibration methods to produce View and Footage Maps, then use Eq. (4) to re-project a known target (e.g., a checkerboard grid) to rectilinear and compare against known geometry. Report per-pixel residual error and map-to-map disagreement; if the maximum residual at 4K exceeds about 0.1 px or the two methods disagree by more than that, the distortion-accurate interchange claim is not supported. For a purely computational variant, generate synthetic footage from a known projection model, recover View/Footage Maps from that model, and verify round-trip error is at interpolation precision—this isolates the encoding, but a real-lens test is still required to validate calibration.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is the 'distortion-accurate workflow' and 'seamless interchange' summarized in the Abstract and implemented as Footage-Map-sampled-through-View-Map (Eq. 4). The pipeline's accuracy is bounded by the accuracy of the initial View/Footage Maps for real lenses, and the paper provides no measurement of that accuracy. Section 2.3 sketches two calibration routes (checkerboard fitting; motorized gimbal with a collimated infinity target), and Section 6 explicitly defers 'an evaluation of the calibration methods' to future work. Without such an evaluation, the claim that arbitrary footage can be converted between lens distortions at production quality is unsupported: errors in the maps propagate directly into the final STMap and then into every resampled frame. The mathematical construction itself is plausible, and the STMap encoding has external compatibility support, but the empirical load-bearing assumption—that a real lens can be captured to sub-pixel accuracy in a single spherical equidistant space by either method—is untested. This is a gap in verification rather than an internal inconsistency; however it is the exact point on which the abstract's 'distortion-accurate' promise rises or falls.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces the Lens Distortion Encoding System (LDES) as a specification for representing lens distortion in a common spherical equidistant projection space using two 32-bit STMap textures: a View Map (mapping output image coordinates to equidistant-space coordinates) and a Footage Map (mapping equidistant-space coordinates to original footage coordinates). The core operation is a two-stage sample (Eq. 4) where the View Map is used to sample the Footage Map, producing an animatable and combineable distortion mapping texture. The paper also defines synthetic projection models (parameterized fisheye through an aximorphic extension), anamorphic and Brown-Conrady extensions, two calibration methods for real lenses (checkerboard fitting and a gimbal-based infinity-target rig), and a workflow example for compositing and ray-tracing. No experimental results, calibration outputs, or error metrics are reported; Section 6 explicitly defers calibration-method evaluation to future work.","tokens_in":10137,"tokens_out":6853,"duration_ms":65443,"significance":"The proposed encoding concept is novel and potentially valuable: if accurate per-lens calibration maps can be produced, LDES could serve as an ACES-like interchange standard for lens distortion, enabling seamless emulation of expensive lens looks, animatable transitions (e.g., anamorphic to spherical), and direct integration with existing STMap workflows in compositing software. The mathematical construction of the two-stage sampling and the use of a common spherical space are internally coherent as a specification, and the paper honestly discloses that the calibration methods are not yet validated. However, the central abstract claims of a 'distortion-accurate workflow' and 'seamless interchange ... regardless of the lens source' rest entirely on the unverified accuracy of the View/Footage map calibration. The paper provides no data showing that real lenses can be encoded to production quality by either proposed method, nor any comparison with established STMap or calibration pipelines. As a specification, the paper is a useful contribution; as a demonstration of the promised capability, it is incomplete.","major_comments":[{"comment":"The anamorphic extension defines the radius as r = sqrt(v_x^2 + v_y^2 / s) without specifying the meaning, range, or reference of the squeeze factor s. For anamorphic lenses, the common convention is a horizontal squeeze (the image is compressed horizontally by a factor typically 1.33x or 2x); the presented formula instead scales the vertical contribution to the radius by 1/s, which is not a horizontal squeeze in the usual sense. This ambiguity directly affects synthetic View Map generation for anamorphic profiles, a key use case claimed in the abstract. The definition of s (e.g., 's is the anamorphic squeeze ratio such that the effective horizontal focal length is multiplied by s') should be stated explicitly, and the resulting equations checked against a known anamorphic projection model.","section":"Section 4.1, Eq. (9a)"}],"minor_comments":[{"comment":"The synthetic projection model is plausible, but the definitions of f and r assume a specific relationship between image-space coordinates and pixel units; it would help to state explicitly that v is in aspect-corrected image-space units normalized to the horizontal FOV Ω.","section":"Section 4.1, Eq. (8a)-(8c)"}],"recommendation":"major_revision","confidential_remarks":"This is a specification paper with no experimental validation. The core encoding idea is sound as a formal construct, but the abstract overstates readiness by claiming 'distortion-accurate' and 'seamless interchange' without any calibration data. For a journal publication, the authors should either add a validation section (even a small-scale proof-of-concept) or substantially soften the claims to describe the system as a proposed standard with an evaluation roadmap. The Brown-Conrady and anamorphic equations also need careful revision before the spec can be relied upon. I lean toward major_revision rather than rejection because the concept is promising and the missing pieces are identifiable and fixable within the manuscript's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a clear, well-put-together specification for a lens distortion interchange format, and the two-map idea is genuinely nice. But the central claim—that it enables a 'distortion-accurate workflow' with seamless interchange—is not tested anywhere in the paper. The calibration methods that would produce the maps are sketched and explicitly deferred to future work.\n\nWhat's actually new: the View Map / Footage Map split, with the footage sampled through the view map to produce a final STMap, is a simple and clever way to separate lens characterization from per-shot distortion. Encoding coordinates in absolute spherical (equidistant) space and labeling maps by FOV avoids the rectilinear-only limitation of standard STMaps and makes transitions between profiles well-defined. The compatibility argument is reasonable: because the output is a standard 32-bit STMap, it should drop into After Effects or Resolve with existing plug-ins.\n\nThe paper does a good job of assembling standard material: the fisheye factor table, Brown-Conrady extensions, anamorphic and aximorphic variants are all sourced, and the author cites his own earlier work where relevant. The workflow example is concrete enough to show the intended usage. This is not a sloppy paper; the math is internally consistent as a specification.\n\nThe soft spots are real. The load-bearing assumption is that real lenses can be calibrated into accurate View and Footage Maps in a single spherical space using either the checkerboard fit or the motorized-gimbal rig. The paper provides no calibration output, no error metrics, no comparison to existing tools. Section 6 says so plainly. That means the abstract's promise of 'distortion accuracy' is not yet supported—it's a design goal. Also, Eqs. 9 and 12 have a few ambiguities (missing parentheses, unclear operator precedence) that would need cleanup before implementation. Table 2 is labeled as an empirical self study, which is honest, but it's not a result.\n\nWho should read this: engineers and artists working on lens mapping standards, or anyone building on STMap pipelines. It gives you a concrete proposal to react against. Don't read it expecting validation of the accuracy claim.\n\nRecommendation: send it to peer review. A specialist referee can assess whether the encoding design is sound and what it would take to validate it, but the review must insist on either an actual calibration evaluation or a clear reframing of the paper as a specification without the accuracy claim. I'd treat it as a conditional accept at best.","headline":"A clean, well-structured specification for lens distortion interchange whose core accuracy claim is untested because the calibration methods are deferred to future work.","tokens_in":10676,"tokens_out":2685,"would_cite":false,"duration_ms":26074,"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 two-texture encoding lets any footage be re-lensed in post, from fisheye to anamorphic and back.","keywords":["Lens distortion encoding","STMap workflow","Camera calibration","Anamorphic lenses","Aximorphic distortion","Fisheye lenses","Visual sphere mapping","Lens distortion correction"],"falsifier":"Calibrate one lens with both proposed methods, checkerboard fitting and the motorized gimbal with a target at infinity, and compare the derived View Maps; if corresponding pixels differ by more than one image pixel at the lens's native resolution, the claimed distortion-accurate interchange fails. Alternatively, shoot the same static scene through lens A and lens B, remap A's footage with an A-to-B map, and measure the pixel difference against B's actual footage.","tokens_in":9639,"feed_emoji":"🎥","tokens_out":6398,"duration_ms":56084,"temperature":0.7,"pith_summary":"The paper proposes the Lens Distortion Encoding System (LDES): a way to store a lens's distortion as a pair of texture maps so that any footage shot with one lens can be rewarped into the look of another lens, or into a synthetic projection, without reshooting. The two textures are the View Map, which records where each output pixel points in a shared spherical image space, and the Footage Map, which records where each point of that sphere lands in the footage. Direct conversion from one lens to another is a single sampling of the Footage Map through the View Map, and because View Maps live in a common spherical space they can be blended and animated. The stated benefit is that an expensive lens's distortion look could be emulated with cheaper glass, and that transitions such as anamorphic-to-spherical, previously impractical, become ordinary opacity ramps. The system intentionally stays inside the existing 32-bit STMap format so mainstream compositing tools can use it unchanged.","feed_headline":"Two texture maps turn any footage into any lens look","feed_subtitle":"One lens profile can morph into fisheye, anamorphic, or rectilinear looks, using tools editors already have.","key_machinery":"The central object is the LDES profile pair: a View Map and a Footage Map, both in 32-bit STMap form (a texture whose channels store sample coordinates rather than colors). The Footage Map encodes absolute pixel positions in a spherical equidistant projection, with its alpha channel marking the footage boundary; the View Map encodes, per output pixel, the corresponding point of that spherical space, with its blue channel optionally storing linear-space vignetting. The load-bearing operation is Equation 4, sampling the Footage Map through the View Map, which synthesizes the final distortion STMap, and Equation 3's tile scaling that reconciles differing FOV labels. The mechanism is completed by the synthetic projection model parameterized by the fisheye factor k (rectilinear, stereographic, equidistant, equisolid, orthographic), the anamorphic and aximorphic extensions for axis-dependent distortion, and Equation 6's conversion of View Map coordinates into ray-tracing incidence vectors.","core_discovery":"The central claim is that lens distortion interchange can be reduced to composing two maps in a spherical equidistant coordinate system. A Footage Map assigns each footage pixel to its absolute position on the visual sphere, labeled by a field-of-view angle Ω; a View Map assigns each output-image pixel to a point on that same sphere, also labeled by Ω. The final distortion texture is obtained by sampling the Footage Map through the View Map (Equation 4), with a simple tile scaling correcting any FOV mismatch between the two maps (Equation 3). Because the View Map is defined in a shared spherical space rather than in the rectilinear space of a particular camera, view maps from different lenses can be opacity-blended, animated, or fed directly into a ray-tracing engine as incidence directions. The paper thereby claims that distortion becomes a fully exchangeable, animatable creative parameter rather than a fixed property of the lens that captured the image.","pith_inferences":["Inference: if the calibration step were validated, the same two-map scheme could be extended beyond lenses to camera motion, using the View Map's rotation mechanism to simulate pan, tilt, and roll in post, a step the paper only sketches.","Inference: the encoding's resolution independence suggests it could serve real-time engines for games and virtual production as a distortion lookup that is cheaper than ray tracing, though the paper lists real-time support only as future work.","Inference: the perceptual claims of Table 2 invite a user study; if the aximorphic region-splitting in the paper's Exercise 1 measurably changes perceived speed, distance, and spaciousness, LDES would supply a direct control surface for cinematic directing.","Inference: a standard library of View Maps would make the system's value depend on calibration quality; comparing the two calibration methods on the same lens would quantify the interchange error that the paper leaves open."],"forward_implications":["A well-calibrated LDES profile of a single lens becomes a reusable asset: any footage shot with that lens can be transformed to any other profile with a matching FOV, or to a synthetic projection, by a single STMap sample.","View Maps can be opacity-blended and animated, so a transition like anamorphic-to-spherical is a continuous interpolation between two textures rather than a lens change.","Because View Maps convert to ray directions, CGI can be rendered with the distortion baked in, letting three-dimensional scenes match live-action footage without a separate warp pass.","Existing STMap pipeline tools in compositing software can consume LDES maps directly, so adoption does not require new hardware or proprietary formats.","The FOV label and tile-scaling rule allow profiles shot at different field widths to be combined without reshooting, provided the FOV normalization of Equation 5 is applied."],"supporting_citations":[{"why":"Supplies the conceptual model of a color interchange space that LDES adapts to distortion.","marker":"[Academy 2015]"},{"why":"Provides the cartographic fisheye projection family parameterized by the factor k, used as the synthetic base model.","marker":"[Krause 2019]"},{"why":"Supplies the Brown-Conrady distortion model that LDES extends for real-lens aberration matching.","marker":"[Wang et al. 2008]"},{"why":"Introduces the aximorphic projection model used for axis-dependent distortion control.","marker":"[Fober 2021]"},{"why":"Documents STMapping as the existing production workflow that LDES builds upon.","marker":"[Houdini help 2024]"},{"why":"Provides the ray-traced Ray Portal BSDF that consumes LDES view maps in the workflow example.","marker":"[Blender Manual 2024]"}],"fun_headline_variants":["Swap lens distortion with two spherical maps","Two maps link any lens to any footage","Lens effects become animatable with two maps","Spherical maps turn any lens into any look","Distortion interchange: two maps do it all"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole workflow assumes that real lenses can be calibrated into accurate View and Footage Maps in a single spherical equidistant coordinate space; the paper describes two calibration methods but defers their evaluation to future work, so if calibration is inaccurate the interchange and animation cannot deliver the claimed distortion accuracy.","fun_headline_variants_meta":{"raw":{"variants":["Swap lens distortion with two spherical maps","Two maps link any lens to any footage","Lens effects become animatable with two maps","Spherical maps turn any lens into any look","Distortion interchange: two maps do it all"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000215,"raw_usage":{"total_tokens":1470,"prompt_tokens":1031,"completion_tokens":439,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":647,"completion_tokens_details":{"reasoning_tokens":385}},"tokens_in":647,"tokens_out":439,"duration_ms":4269,"temperature":1.0,"reasoning_tokens":385,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T12:43:33.044956+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Calibrate one lens with both proposed methods, checkerboard fitting and the motorized gimbal with a target at infinity, and compare the derived View Maps; if corresponding pixels differ by more than one image pixel at the lens's native resolution, the claimed distortion-accurate interchange fails. Alternatively, shoot the same static scene through lens A and lens B, remap A's footage with an A-to-B map, and measure the pixel difference against B's actual footage.","supporting_citations":[],"review_version":1}