{"id":"add48c20-8289-4c90-9623-182af1f39b29","arxiv_id":"2604.02785","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"CANDLE uses DINOv3 semantic features as illumination-invariant priors plus custom decoder refinements to normalize images under colored ambient lighting, reporting a 1.22 dB PSNR gain and strong challenge rankings.","lead":"The paper presents CANDLE, a neural network method that injects multi-layer features from the pre-trained DINOv3 model to correct color shifts caused by multi-colored lighting and recover object-intrinsic colors. A smart generalist might read it to understand how self-supervised vision models can serve as robust priors for practical image restoration tasks like photography and surveillance under uncontrolled illumination.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"No direct quantification or ablation isolates whether DINOv3 consistency drives gains versus the B FACG + SFFB refinement modules.","rationale":"The reader's weakest assumption correctly flags the transfer risk into the specific encoder-decoder; the above concern sharpens it to a missing isolation experiment that would directly test whether the motivating observation actually carries the performance.","tokens_in":1711,"tokens_out":321,"duration_ms":20325,"concrete_test":"Retrain the full CANDLE model on CL3AN with D.O.G. disabled (zeroing the multi-layer DINO injections) while retaining B FACG + SFFB; compare PSNR/FID to the reported baseline. If the drop is <0.4 dB, the DINO consistency claim is not load-bearing for the headline result.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim rests on the observation that DINOv3 features are highly consistent across colored vs. ambient inputs, which then motivates their injection via D.O.G. into the encoder. However, the reported +1.22 dB PSNR and challenge rankings are end-to-end results; without (a) explicit metrics (cosine similarity, CKA) comparing DINO features on colored inputs to ambient ground truth, or (b) an ablation that removes D.O.G. while keeping the color-frequency refinement blocks, it remains possible that the architectural refinements alone account for the improvement and that the DINO priors contribute little or introduce the very chromatic artifacts the decoder is then forced to suppress.","agreement_with_reader":"partial"},"referee_report":{"model":"grok-4.3","summary":"The paper proposes CANDLE for color ambient lighting normalization under multi-colored illumination. It observes that DINOv3 self-supervised features are consistent across colored-light inputs and ambient ground truth, then introduces DINO Omni-layer Guidance (D.O.G.) to inject multi-layer DINOv3 features into encoder stages, plus color-frequency refinement blocks (BFACG + SFFB) in the decoder. On the CL3AN dataset it reports a +1.22 dB PSNR gain over prior methods, 3rd place on the NTIRE 2026 ALN Color Lighting Challenge, and 2nd place in fidelity on the White Lighting track with lowest FID.","tokens_in":1870,"tokens_out":583,"duration_ms":28332,"significance":"If the central claim holds, the work shows that off-the-shelf self-supervised ViT features can serve as illumination-robust semantic priors for low-level normalization tasks without task-specific fine-tuning. The challenge rankings provide external validation of generalization across chromatic and luminance-dominant conditions, and the parameter-free nature of the DINO prior is a clear strength.","major_comments":[{"comment":"Experiments section: the abstract reports a +1.22 dB PSNR gain and challenge rankings, yet no error bars, multiple-run statistics, or ablation tables are referenced; without these the robustness of the improvement cannot be assessed and the central claim remains under-supported.","section":"Experiments"},{"comment":"Motivation and Section 3.1: the claim that DINOv3 features remain highly consistent between colored-light inputs and ambient ground truth is stated qualitatively but never quantified (no cosine similarity, CKA, or layer-wise distance metrics are provided), so the motivation for D.O.G. injection rests on an unverified assumption.","section":"Section 3.1"},{"comment":"Ablation studies: no experiment removes D.O.G. while retaining B FACG + SFFB (or vice versa); the end-to-end PSNR gain therefore does not isolate whether the DINO priors contribute meaningfully or whether the color-frequency refinement blocks alone suffice.","section":"Ablation studies"}],"minor_comments":[{"comment":"Abstract: acronyms D.O.G., B FACG and SFFB appear without parenthetical expansions on first use.","section":"Abstract"},{"comment":"Figure 1: the architecture diagram would benefit from explicit arrows or labels indicating the precise encoder stages where D.O.G. features are injected.","section":"Figure 1"}],"recommendation":"major_revision","confidential_remarks":"The manuscript's dependence on challenge leaderboard results is acceptable for significance but the absence of controlled ablations may limit reproducibility assessment for this journal."},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive feedback. We agree that the experimental section requires stronger statistical support and more complete ablations to substantiate the claims. We will revise the manuscript to address all three points by adding the requested analyses and metrics.","responses":[{"response":"We agree that reporting error bars and multiple-run statistics is necessary to demonstrate robustness. In the revised manuscript we will add mean PSNR and standard deviation computed over five independent training runs for the CL3AN results, include error bars on the main comparison table, and clarify that the challenge rankings reflect a single fixed submission.","revision_made":"yes","referee_comment":"[Experiments] Experiments section: the abstract reports a +1.22 dB PSNR gain and challenge rankings, yet no error bars, multiple-run statistics, or ablation tables are referenced; without these the robustness of the improvement cannot be assessed and the central claim remains under-supported."},{"response":"We acknowledge that the consistency claim was presented only qualitatively. We will add a new quantitative analysis subsection in the revised Section 3.1 that reports layer-wise cosine similarity and CKA scores between DINOv3 features of colored-light inputs and their ambient ground-truth counterparts, thereby providing empirical grounding for the D.O.G. design.","revision_made":"yes","referee_comment":"[Section 3.1] Motivation and Section 3.1: the claim that DINOv3 features remain highly consistent between colored-light inputs and ambient ground truth is stated qualitatively but never quantified (no cosine similarity, CKA, or layer-wise distance metrics are provided), so the motivation for D.O.G. injection rests on an unverified assumption."},{"response":"We agree that the current ablations do not fully isolate the contribution of each component. We will add a new ablation table in the revised manuscript that includes a variant with D.O.G. removed while retaining B FACG and SFFB, reporting the resulting PSNR to quantify the incremental benefit of the DINO priors.","revision_made":"yes","referee_comment":"[Ablation studies] Ablation studies: no experiment removes D.O.G. while retaining B FACG + SFFB (or vice versa); the end-to-end PSNR gain therefore does not isolate whether the DINO priors contribute meaningfully or whether the color-frequency refinement blocks alone suffice."}],"tokens_in":1442,"tokens_out":518,"duration_ms":30687,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main point is that the authors notice DINOv3 features stay fairly stable when the input has colored lighting versus normal ambient light. They build CANDLE around that by injecting multi-layer DINO features into the encoder stages with their D.O.G. schedule and then add BFACG and SFFB blocks in the decoder to handle chromatic issues and detail loss. On the CL3AN set this gives a 1.22 dB PSNR gain over prior work, and the method placed third overall in the NTIRE 2026 color lighting challenge with strong fidelity numbers on the white-light track as well. Code is released, which is helpful for anyone who wants to try it.","headline":"CANDLE shows a workable way to use DINOv3 features for colored lighting normalization plus some decoder fixes, but the results do not separate what the priors actually contribute.","tokens_in":2404,"tokens_out":216,"would_cite":false,"duration_ms":34100,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":{"model":"grok-4.3","evidence":[{"relation":"unclear","rs_module":"IndisputableMonolith/Cost/FunctionalEquation.lean","rs_theorem":"washburn_uniqueness_aczel","paper_passage":"We observe that DINOv3's self-supervised features remain highly consistent between colored-light inputs and ambient-lit ground truth, motivating their use as illumination-robust semantic priors."},{"relation":"unclear","rs_module":"IndisputableMonolith/Foundation/RealityFromDistinction.lean","rs_theorem":"reality_from_one_distinction","paper_passage":"DINO Omni-layer Guidance (D.O.G.) ... color-frequency refinement design (BFACG + SFFB)"}],"headline":"CANDLE's DINOv3 feature-consistency priors and encoder-decoder refinements are orthogonal to RS recognition-cost forcing","alignment":"orthogonal","rationale":"Paper centers on empirical observation that DINOv3 patch features remain consistent across colored vs. ambient illumination (Figs. 1-2, cosine similarity), motivating D.O.G. multi-layer injection plus BFACG/SFFB decoder blocks. No J-cost, cosh(ρ ln φ), ratio-symmetric functional equations, φ-ladder spacings, 8-tick periodicity, or parameter-free derivation of constants appears. Domain (cs.CV image restoration) lies outside the RS forcing chain from distinction to spacetime/constants.","tokens_in":49912,"confidence":"high","tokens_out":320,"duration_ms":22810,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"DINOv3 self-supervised features stay consistent across colored and ambient lighting, enabling accurate recovery of intrinsic object colors.","keywords":["illumination normalization","color ambient lighting","DINOv3 features","semantic priors","image restoration","self-supervised learning","chromatic shift correction"],"falsifier":"A controlled test set of scenes with previously unseen material types and extreme colored highlights where the CANDLE output shows larger chromatic deviation from ground-truth ambient images than a strong baseline that does not use DINO features.","tokens_in":2612,"feed_emoji":"💡","tokens_out":708,"duration_ms":18744,"temperature":0.7,"pith_summary":"The paper establishes that DINOv3 features extracted from images under multi-colored lights remain highly similar to those from the same scenes under neutral ambient light. This consistency supplies semantic guidance that is largely free of illumination-induced color bias, which existing geometric or low-level priors cannot provide when chromatic shifts dominate. The authors build CANDLE around this observation by injecting the features at multiple encoder layers and adding decoder-side filters to prevent color collapse and detail loss. Experiments report a 1.22 dB PSNR improvement on the CL3AN benchmark and competitive ranking in the NTIRE 2026 lighting normalization challenge.","feed_headline":"DINOv3 features correct color tints from mixed lighting","feed_subtitle":"Self-supervised features stay consistent under colored illumination and guide recovery of true object colors in an encoder-decoder network.","key_machinery":"DINO Omni-layer Guidance (D.O.G.) that adaptively injects multi-layer DINOv3 features into successive encoder stages, paired with a color-frequency refinement design (BFACG + SFFB) in the decoder.","core_discovery":"DINOv3 features remain highly consistent between colored-light inputs and ambient-lit ground truth; this property is exploited as illumination-robust semantic priors inside an encoder-decoder network that uses DINO Omni-layer Guidance to inject multi-layer features adaptively and color-frequency refinement modules (BFACG + SFFB) to suppress chromatic artifacts.","pith_inferences":["The same DINO consistency property could be tested on video frames to enforce temporally stable color normalization.","If the priors prove robust to other degradations, the approach might extend to joint lighting and shadow correction tasks.","Replacing DINOv3 with later self-supervised models could be measured to check whether newer feature sets further reduce residual color bias.","The method's reliance on a frozen backbone suggests it may run faster on edge devices than methods that retrain large illumination estimators."],"forward_implications":["The network achieves a 1.22 dB PSNR gain over the previous best method on the CL3AN dataset.","CANDLE places third overall on the NTIRE 2026 ALN Color Lighting Challenge and second in fidelity on the White Lighting track with lowest FID.","The same design generalizes across both strongly chromatic and luminance-dominant illumination conditions.","Multi-layer DINO injection plus frequency-aware refinement reduces both highlight saturation and material-dependent reflectance errors."],"fun_headline_variants":["DINOv3 priors recover intrinsic object colors from colored lights","Adaptive DINO layer injection normalizes ambient lighting colors","CANDLE leverages consistent DINOv3 features for chromatic normalization","DINO Omni-layer Guidance suppresses chromatic collapse in decoding"],"cache_read_input_tokens":64,"weakest_assumption_plain":"The consistency seen in DINOv3 features between colored and ambient conditions will carry over to the chosen network architecture and refinement modules without creating new color errors or losing fine detail.","fun_headline_variants_meta":{"raw":{"variants":["DINOv3 priors recover intrinsic object colors from colored lights","Adaptive DINO layer injection normalizes ambient lighting colors","CANDLE leverages consistent DINOv3 features for chromatic normalization","DINO Omni-layer Guidance suppresses chromatic collapse in decoding"]},"model":"grok-4.3","cost_usd":0.010259,"raw_usage":{"total_tokens":4453,"prompt_tokens":645,"num_sources_used":0,"completion_tokens":65,"cost_in_usd_ticks":102590500,"prompt_tokens_details":{"text_tokens":645,"audio_tokens":0,"image_tokens":0,"cached_tokens":64},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":3743,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":645,"tokens_out":65,"duration_ms":46970,"temperature":1.0,"reasoning_tokens":3743,"cache_read_input_tokens":64,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-05-13T19:52:36.374200+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A controlled test set of scenes with previously unseen material types and extreme colored highlights where the CANDLE output shows larger chromatic deviation from ground-truth ambient images than a strong baseline that does not use DINO features.","supporting_citations":[],"review_version":1}