{"id":"2f578d29-8f71-4127-8fb8-65128d9b9908","arxiv_id":"2504.16086","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A single-panorama kitchen remodeling pipeline with low-cost photometric calibration that recovers absolute radiance, validated at 3.988 cd/m² mean error on 141 scenes.","lead":"The authors built a virtual staging tool that edits a single 360-degree photo of a kitchen, replaces cabinets and appliances, and relights the scene with real outdoor light. They also propose a $30 light meter calibration shortcut to replace a $5,000 luminance meter, with a dataset of 141 paired indoor and outdoor HDR panoramas.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Calibration is verified only on one indoor target; the outdoor HDR scaling and full-panorama radiometric uniformity remain unverified, yet the relighting claim depends on both.","rationale":"The reader's verdict is CONDITIONAL, and I agree it should remain CONDITIONAL. The reader's weakest_assumption identifies a real limitation but attributes it to a uniform-radiance assumption; that assumption is not actually needed for a global scale factor, since the orthographic average used in the calibration has exactly the cosine weighting of the illuminance integral in Eq. 1. The more defensible concern is that the reported 3.988 cd/m^2 error comes from a single whiteboard target per scene, in the front hemisphere, and cannot validate spatial uniformity across the stitched 360-degree panorama or the scaling of the paired outdoor HDR that serves as the environment map. I do not see an internal contradiction in the calibration itself; with a multi-target spatial test and an explicit outdoor exposure-validation step, the method could be supported. I therefore recommend no change to the reader's conditional verdict: the scientific core is plausible but the strong full-panorama/outdoor calibration claim needs additional evidence before acceptance.","tokens_in":10157,"tokens_out":12107,"duration_ms":127511,"concrete_test":"Select at least 10 scenes; place 5-10 matte Lambertian patches in different directions (near window, dark corner, back hemisphere, seam regions) and measure each with the LS-160. Calibrate each panorama using only the TS-710 reading, then report per-patch mean absolute and maximum luminance errors. On a sunny day, also validate the paired outdoor HDR by comparing its calibrated radiance for known sky and wall patches against an outdoor light-meter or luminance-meter measurement, under both locked and auto exposure. If per-patch errors systematically exceed the reported 3.988 cd/m^2 or outdoor errors are materially larger, the global calibration claim fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The uniform-radiance worry in the reader's weakest_assumption is not actually the main issue: for a linear HDR response, a global scale k from E/pi divided by the orthographic-image mean is valid even under non-uniform radiance, because the orthographic mean has the same cos-weighting as the integral in Eq. 1. The load-bearing gap is that the paper validates with one indoor whiteboard comparison per scene and then claims absolute scene radiance for the whole panorama and for the paired outdoor panorama. This requires three conditions that are not established: (1) the camera response after vignetting correction and stitching is spatially constant across both fisheye lenses and the seam; (2) the indoor and outdoor captures share exactly the same exposure and gain; and (3) the outdoor HDR is not clipped in bright sky or window regions. The text says the indoor and outdoor panoramas are 'captured simultaneously' with a single camera, which is physically impossible; if this means sequentially, the exposure-lock requirement is unstated. A one-point, front-hemisphere whiteboard check cannot detect mis-scaling in the back lens, at stitching seams, or in the outdoor environment map, so the 3.988 cd/m^2 error does not by itself support the full calibration claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a virtual staging system for kitchen remodeling from 360° HDR panoramas. It contributes a capture protocol using a single Ricoh Theta Z1 to obtain paired indoor and outdoor panoramas, a low-cost photometric calibration method based on a TS-710 illuminance meter that scales each HDR panorama to absolute radiance, a dataset of 141 paired panoramas with photometric labels, and a rendering pipeline that estimates 3D layout, inserts new kitchen components, and relights the edited scene under global illumination. The calibration method is validated against a Konica Minolta LS-160 luminance meter on one whiteboard target per scene, reporting a mean absolute luminance error of 3.988 cd/m². The relighting results are demonstrated qualitatively in several figures.","tokens_in":10463,"tokens_out":7776,"duration_ms":70266,"significance":"If the calibration result holds for full panoramas and outdoor environment maps, the work has practical value: it lowers the cost of photometric calibration by two orders of magnitude, provides a new paired indoor-outdoor HDR dataset, and demonstrates a complete editing/relighting pipeline for a complex indoor scene type. The calibration factor is not circular: it is computed from an independent illuminance measurement and checked against a separate luminance meter. The main strengths are the dataset scale, the low-cost measurement procedure, and the integration of capture, layout generation, and physically based rendering. However, the quantitative support is concentrated on a single target per scene, and the relighting quality is not compared with prior methods, so the broader claims currently outrun the evidence.","major_comments":[{"comment":"The reported 3.988 cd/m² error validates the calibration factor only at one whiteboard location in the front lens hemisphere, yet the same factor is applied to the entire 360° panorama and to the paired outdoor HDR used as an environment map. This transfer requires (i) spatially uniform radiometric response after vignetting correction and stitching across both lenses and the seam, (ii) identical camera exposure and gain for the indoor and outdoor captures, and (iii) an outdoor HDR that is not clipped in bright sky or window regions. None of these conditions is demonstrated. I ask for multi-position luminance measurements (front hemisphere, back hemisphere, seam) and an outdoor radiance check, or an explicit argument that the Theta Z1's internal processing guarantees these properties.","section":"Section 3, Algorithm 1"},{"comment":"The manuscript says the indoor and outdoor panoramas are \"captured simultaneously\" with a single camera and later says the outdoor photograph is taken \"immediately after\" the indoor photograph; these are inconsistent, and simultaneous capture by one camera is physically impossible. Because the calibration factor derived indoors is applied to the outdoor panorama, the exposure-locking procedure and the time interval between the two captures must be stated precisely. Without this, the outdoor environment map may be mis-scaled even if the indoor calibration is accurate.","section":"Section 3, paragraph following Eq. (2)"},{"comment":"The abstract and contribution list describe \"automatic kitchen layout generation,\" but Algorithm 2 takes a user-defined \"Sequence Order\" as input and the text states that objects are placed \"in a sequence order defined by user inputs.\" The amount of user interaction should be clarified, and the automatic scaling of the last object to cover the wall corner should be evaluated or at least discussed as a geometric approximation that can distort object proportions.","section":"Section 4.1, Algorithm 2"},{"comment":"The central claim of \"high-quality scene relighting\" is supported only by qualitative images. There is no quantitative comparison with prior relighting methods [6,9,22,32], nor a quantitative comparison with the real captured reference shown in Figure 10(d). Please add an evaluation protocol (e.g., rendered-versus-captured error on held-out regions, or a perceptual study) or restrict the claims accordingly.","section":"Section 4.2, Figures 9 and 10"}],"minor_comments":[{"comment":"The uniform-radiance assumption is stronger than needed. For an orthographic projection, the average HDR value over the disk is E/π for any radiance distribution, so the method does not require a constant L(θ,φ); this should be stated to avoid a misleading derivation.","section":"Section 3, Eq. (2)"},{"comment":"The input label \"Illuminace\" should be spelled \"Illuminance.\"","section":"Section 3, Algorithm 1"},{"comment":"The error-percentage plot would be easier to interpret with error bars or a density overlay; as printed, the color coding by absolute luminance makes it difficult to read the distribution of the 141 data points.","section":"Section 3, Fig. 6"},{"comment":"The phrase \"from a Single Panorama\" is potentially misleading because the pipeline uses paired indoor and outdoor panoramas; please clarify that the indoor panorama is the edited scene input while the outdoor panorama supplies the environment map.","section":"Title and Abstract"},{"comment":"The dataset contribution mentions labels for outdoor illuminance and room orientation, but Section 3 describes only the indoor illuminance and whiteboard luminance protocols; please document how these additional labels were measured.","section":"Section 3, Dataset description"},{"comment":"Reference [13] contains a typo: \"Evalution\" should be \"Evaluation.\"","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The paper is closely related to the authors' prior ISVC 2023 and MVA 2024 papers on virtual home staging; the novelty here is the single-camera capture, low-cost illuminance calibration, larger dataset, and kitchen-specific layout generation. The most important revision is to substantiate the full-panorama and outdoor calibration claims. I would not insist on a full relighting benchmark if the scope is an application paper, but at least one quantitative sanity check for the environment map would strengthen the paper."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper’s real nugget is the low-cost photometric calibration: a $30 light meter, combined with the orthographic projection trick, recovers absolute radiance for the front hemisphere with a 3.988 cd/m² mean absolute error across 141 scenes. That is a genuinely useful adaptation of Inanici’s fisheye method to a dual-lens equirectangular camera, and the paired indoor/outdoor HDR dataset is a potentially valuable resource. The kitchen remodeling pipeline is competent integration of existing layout estimation, segmentation, and rendering, but it is not scientifically new; the relighting results are shown only qualitatively, with no comparison to prior methods.\n\nWhere the paper gets wobbly is the extrapolation from that one whiteboard measurement to the full panorama and to the paired outdoor image. The validation is a single indoor target per scene, measured at the center of the front lens. The calibration factor is then applied to the entire equirectangular image and to the outdoor environment map, which is what actually drives the relighting. None of that is verified. The text even says the indoor and outdoor panoramas are captured simultaneously with a single camera, which is impossible; if they mean sequentially, the exposure-lock requirement is unstated. That is a load-bearing gap, not a nitpick.\n\nOne thing the stress-test note got right: the reader’s \"weakest assumption\" about uniform hemispherical radiance is a dud. For a linear camera response, the mean of the orthographically projected fisheye image has exactly the same cos-weighting as the illuminance integral, so L = E/π holds for the front hemisphere regardless of angular distribution. The math is fine. The problem is purely the unvalidated transfer to other directions and to the outdoor capture.\n\nThe calibration derivation itself is not circular, and the validation against an independent luminance meter is good evidence for the front hemisphere. The dataset, if actually released, would support future relighting work, but the manuscript as submitted gives no accessible link. The \"automatic\" layout generation also overstates things: the user specifies sequence order and layout type, so it is interactive rather than fully automatic.\n\nBottom line: worth a serious referee. The calibration contribution is novel enough and the dataset is valuable enough to justify review, but the paper needs revision before acceptance: clarify the capture protocol and exposure lock, validate calibration on more than one target (at least the back lens and a bright outdoor region), and either release the dataset or explicitly state where it lives. I would send it to review, with those changes required.","headline":"A credible low-cost HDR calibration trick and a useful dataset, but the validation only covers one whiteboard while the claims cover the whole panorama and the paired outdoor image.","tokens_in":10899,"tokens_out":4581,"would_cite":true,"duration_ms":41793,"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 $30 light meter replaces a $5,000 one for believable relighting.","keywords":["virtual staging","panoramic HDR","photometric calibration","relighting","kitchen remodeling","indoor scene editing","global illumination","Pano-Pano HDR dataset"],"falsifier":"Take a room lit mainly through a side window, place a luminance meter on white targets facing several different directions, and compare each reading with the same pixel in a light-meter-calibrated HDR panorama; if errors grow systematically for targets facing the window while the single whiteboard target reports near 3.988 cd/m², the uniform-radiance assumption is the failing link.","tokens_in":9930,"feed_emoji":"🏠","tokens_out":5350,"duration_ms":47517,"temperature":0.7,"pith_summary":"This paper tries to show that a complete kitchen remodeling preview can be generated from a single 360-degree photograph, with lighting that is physically believable rather than painted on. The key move is a low-cost photometric calibration: a roughly $30 illuminance meter, the TS-710, is held next to the camera while the panorama is shot, and the measured illuminance is converted into an absolute scene radiance value that rescales every HDR pixel. Across 141 captured scenes the rescaled panoramas agree with a reference luminance meter to a mean absolute error of 3.988 cd/m². If that calibration holds, the expensive $5,000 luminance meter that prior virtual-staging workflows required can be replaced by a consumer light meter, and relighting edited interiors becomes a practical service rather than a laboratory procedure. The same pipeline then automatically detects the kitchen area, places new cabinets and appliances along the wall, and renders the result under the real outdoor illumination captured in a paired outdoor panorama.","feed_headline":"A $30 light meter replaces a $5,000 one for relighting","feed_subtitle":"Cheap calibration makes physically accurate virtual kitchen remodeling from a single panorama practical.","key_machinery":"The load-bearing identity is the relation between illuminance and radiance over a hemisphere: $E = \\int_0^{2\\pi} \\int_0^{\\pi/2} L(\\theta,\\phi)\\,\\sin\\theta\\cos\\theta\\,d\\theta\\,d\\phi$, which under the assumption of uniform radiance reduces to $L = E/\\pi$. To use it, the panorama's front-lens region, spanning the central half of the equirectangular image, is geometrically warped into a 180-degree orthographic fisheye image whose average pixel luminance can be compared with $L$. That comparison yields one scalar calibration factor per capture, applied to both the indoor and outdoor panoramas. The rest of the application pipeline assembles around this: layout estimation supplies 3D walls, semantic segmentation locates the kitchen, a $4\\times 4$ rigid transformation $M = R_z(\\theta)\\cdot T(t_x, t_y)$ positions each new component along the wall, and a global-illumination renderer consumes the calibrated outdoor panorama as its environment map.","core_discovery":"The central claim is that absolute scene radiance can be recovered from a single HDR panorama using only a scalar illuminance measurement. The measured horizontal illuminance $E$ is converted to a uniform hemispherical luminance $L = E/\\pi$, and after cropping the camera's front fisheye image into a 180-degree orthographic projection, the average displayed luminance of that image is rescaled by the factor $k = L / L_{\\text{avg}}$ to match physical units. The same factor is applied to the paired outdoor panorama, which then acts as a calibrated 360-degree environment map for rendering. Validated against a professional luminance meter on a whiteboard target, the calibration reaches a mean absolute error of 3.988 cd/m² over the 141 scenes in the new Pano-Pano HDR dataset. This is what makes the kitchen staging results relightings in actual light, with reflections and spatially varying window illumination, rather than composited edits.","pith_inferences":["The uniform-radiance simplification suggests a cheap upgrade the paper does not explore: weight each orthographic fisheye pixel by $\\cos\\theta$ before averaging, which would reduce the influence of near-horizon bright windows and likely lower the error where directional light dominates.","The calibration idea is not kitchen-specific; the same $L = E/\\pi$ scaling could be applied to any indoor-outdoor panorama pair, so bathrooms, offices, and whole-home tours are natural extensions of the stated pipeline.","The reported error is measured on one whiteboard per scene; the strongest follow-up test would be to measure luminance targets at several depths and orientations and check whether the single scalar factor still holds away from the meter's position.","The two-lens geometry of the Ricoh Theta Z1 means the 180-degree orthographic crop is only an approximation of what the light meter sees; using the camera's full calibrated projection function instead of a crop-and-warp could tighten the alignment."],"forward_implications":["Photometric calibration of HDR panoramas drops from a $5,000 instrument to a $30 light meter, so any real-estate photographer with a 360-degree camera can produce physically scaled radiance images.","Because the outdoor panorama is calibrated with the same factor and used as an environment map, virtual cabinets and countertops receive the room's actual window light, including directional shadows and reflections.","The automatic layout arrangement handles I, L, and U kitchens by splitting walls into linear sections, so remodeling previews can be generated without manual 3D modeling per room.","The 141-pair Pano-Pano HDR dataset, labeled with indoor and outdoor illuminance, indoor luminance, and room orientation, gives other relighting and staging methods a common calibration benchmark."],"supporting_citations":[{"why":"Supplies the radiometric self-calibration used to merge bracketed exposures into an HDR radiance image.","marker":"[24]"},{"why":"Establishes the illuminance-based calibration approach for fisheye HDR images that this method adapts to panoramas.","marker":"[13]"},{"why":"Provides the standard luminance-meter calibration workflow whose measurements are used as the reference for comparison.","marker":"[18]"},{"why":"Prior paired indoor-outdoor virtual staging approach that this work simplifies to a single camera and cheap meter.","marker":"[15, 16]"},{"why":"Supplies the single-panorama 3D layout estimation used to build the geometry for kitchen remodeling.","marker":"[38]"},{"why":"The rendering system used to compute global illumination with the calibrated outdoor panorama as environment map.","marker":"[14]"},{"why":"Semantic segmentation that identifies the kitchen area in the panorama before component insertion.","marker":"[39]"},{"why":"Provides measured light-source spectra used to add realistic electrical lighting to the edited kitchen.","marker":"[25]"}],"fun_headline_variants":["One panorama, $30 meter: physical relighting for kitchen remodels","Cheap light meter unlocks realistic relighting from a single shot","Single HDR panorama, budget calibration, true-to-life kitchen relighting","Cut relighting costs 100x with one panorama and a $30 sensor"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The calibration assumes the light meter's reading came from a constant brightness across the whole 180-degree hemisphere, so one average value $L = E/\\pi$ rescales every pixel; a room with a bright window or direct sunlight violates that, and the reported validation only checks one whiteboard target per scene.","fun_headline_variants_meta":{"raw":{"variants":["One panorama, $30 meter: physical relighting for kitchen remodels","Cheap light meter unlocks realistic relighting from a single shot","Single HDR panorama, budget calibration, true-to-life kitchen relighting","Cut relighting costs 100x with one panorama and a $30 sensor"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000192,"raw_usage":{"total_tokens":1306,"prompt_tokens":862,"completion_tokens":444,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":478,"completion_tokens_details":{"reasoning_tokens":366}},"tokens_in":478,"tokens_out":444,"duration_ms":5005,"temperature":1.0,"reasoning_tokens":366,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T10:55:58.010550+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a room lit mainly through a side window, place a luminance meter on white targets facing several different directions, and compare each reading with the same pixel in a light-meter-calibrated HDR panorama; if errors grow systematically for targets facing the window while the single whiteboard target reports near 3.988 cd/m², the uniform-radiance assumption is the failing link.","supporting_citations":[{"cited_title":"Ra- diometric self calibration","cited_arxiv_id":null,"evidence_quote":"Supplies the radiometric self-calibration used to merge bracketed exposures into an HDR radiance image."},{"cited_title":"Evalution of high dynamic range image-based sky models in lighting sim- ulation","cited_arxiv_id":null,"evidence_quote":"Establishes the illuminance-based calibration approach for fisheye HDR images that this method adapts to panoramas."},{"cited_title":"Hdrscope: high dynamic range im- age processing toolkit for lighting simulations and analysis","cited_arxiv_id":null,"evidence_quote":"Provides the standard luminance-meter calibration workflow whose measurements are used as the reference for comparison."},{"cited_title":"Semantically supervised appear- ance decomposition for virtual staging from a single panorama","cited_arxiv_id":null,"evidence_quote":"Supplies the single-panorama 3D layout estimation used to build the geometry for kitchen remodeling."},{"cited_title":"Mitsuba 3 renderer,","cited_arxiv_id":null,"evidence_quote":"The rendering system used to compute global illumination with the calibrated outdoor panorama as environment map."},{"cited_title":"Semantic understanding of scenes through the ade20k dataset","cited_arxiv_id":null,"evidence_quote":"Semantic segmentation that identifies the kitchen area in the panorama before component insertion."},{"cited_title":"Light Sources — SPD Curves — National Gallery, London — Informa- tion","cited_arxiv_id":null,"evidence_quote":"Provides measured light-source spectra used to add realistic electrical lighting to the edited kitchen."}],"review_version":1}