{"id":"6572eae2-ab53-4ea0-9943-ba6e87a10d16","arxiv_id":"2511.17126","paper_version":5,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":9,"one_line_summary":"A blind aberration-correction model trained on a larger, uniformly sampled lens library with a latent PSF codebook achieves state-of-the-art zero-shot correction across diverse lenses.","lead":"OmniLens++ trains a blind lens-aberration-correction network on a larger, more uniformly sampled synthetic lens library and guides it with a learned latent PSF codebook. On simulated and real-captured lenses it reports state-of-the-art zero-shot correction, plus a new benchmark for evaluating CAC models.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"AODLibpro's 'uniform' coverage is measured with the same OIQ/OD-Class proxy used to construct it; independent PSF-space diversity is never shown, leaving the data-scalability claim potentially circular.","rationale":"The reader's weakest assumption — that OIQ and OD-Class are faithful proxies for the full space of optical degradations — is exactly the load-bearing point I would stress-test. The paper's own uniformity evidence is defined in the same metric space used for sampling, so Figure 5 cannot independently certify that AODLibpro is more diverse in PSF space. If the proxy is uninformative, then the claimed scalability of AODLibpro and the validity of AODLibproTest as a benchmark are weakened, though the LPR mechanism might still contribute. This is a correctness risk rather than an internal contradiction: the method could still work, but the paper's central data contribution would lack support until independent PSF-space coverage is shown. I am not raising this as a rejection. The paper has real independent support: detailed loss formulations, open-source simulator usage, extensive ablations, and a clear blind-evaluation protocol. Secondary issues — no error bars, unreleased code/data, the abstract's few-shot claim absent from the body — reinforce conditionality but are less central to the scientific claim. The proposed PSF-space clustering test is feasible with the artifacts the paper promises to release and would settle whether the OIQ/OD-Class sampling basis actually improves coverage or merely reshuffles a proxy histogram. Until then, the CONDITIONAL verdict stands, so no verdict change is needed.","tokens_in":26549,"tokens_out":10704,"duration_ms":106607,"concrete_test":"Re-derive the sampling comparison without OIQ/OD-Class: for the same EAOD lens source, compute the full per-FoV PSF maps via DeepLens (as the paper already does), reduce them to a descriptor (e.g., PCA or a learned embedding over concatenated per-FoV PSF kernels), cluster with k-means, and measure the number of occupied clusters and nearest-neighbor distances of RealLens-Sim/RealLens-Snap test PSFs to AODLibpro vs AODLib-EAOD. If AODLibpro does not occupy substantially more PSF-space clusters and cover the test lenses at least as well as AODLib-EAOD, the hybrid sampling basis is not the source of the reported gains, and the uniformity claim in §3.2/Figure 5 is an artifact of the proxy.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that AODLibpro scales because it uniformly covers optical degradation (OD) hinges on the hybrid sampling basis in §3.2: OIQ (Eq. 11, hand-set weights 0.4/0.3/0.3) and OD-Class with empirically set α=0.85. AODLibpro is then sampled to be uniform over the 18 sub-classes created from these two proxies. The paper's evidence of uniformity is partly tautological: Figure 5(a) is a histogram of OD-Class — the same categories used for sampling — and Figure 5(b) visualizes coverage using OIQ per FoV/wavelength. If OIQ/OD-Class miss PSF structure that matters for correction (e.g., asymmetric coma/astigmatism, PSF shape changes that leave average OIQ trends unchanged, chromatic patterns beyond the Appendix D analysis, or simulator-specific artifacts), AODLibpro may be no more diverse in the true degradation space than AODLib-EAOD. The scalability conclusion (Figure 5c, Table 6) and the meaningfulness of AODLibproTest as a benchmark then rest on an unvalidated proxy. This is load-bearing for the data contribution specifically; the LPR guidance claim is less affected, but the paper's headline contribution includes AODLibpro.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a blind lens aberration correction framework (called OmniLens++ in the main text and FoundCAC in the arXiv metadata abstract) with two main contributions. On the data side, it constructs AODLibpro by extending an EAOD-based lens-source generator with aspheric surfaces and image-distance perturbation, then samples lenses uniformly over a hybrid basis formed by an OIQ severity score (Eq. 11) and an OD-Class taxonomy of spatial-variation patterns, yielding 18 subclasses with 200 train and 3 test lenses per subclass. On the model side, it introduces LPR, a VQVAE-based discrete representation of PSF maps regularized by an ODN that simulates the optical degradation process; the learned codebook is frozen and used to guide a UNet/Swin restoration backbone via predicted latent PSF features. Experiments include a new AODLibproTest benchmark, a RealLens-Sim suite covering minimalist spherical/aspheric lenses, a metalens, misaligned smartphone lenses, and high-end lenses, plus real-snapped qualitative and non-reference quantitative evaluations. The paper reports state-of-the-art zero-shot results, with AODLibpro and LPR each contributing average PSNR gains over the OmniLens baseline on RealLens-Sim.","tokens_in":26905,"tokens_out":8190,"duration_ms":87124,"significance":"If the reported results hold, the paper makes a meaningful advance in a relatively underexplored area: it demonstrates that a large automatically designed lens library can be made more scalable and more uniform, and that PSF-derived priors can be injected in a fully blind manner through a learned discrete latent representation. The evaluation is broader than in prior work, spanning simulated, synthetic-benchmark, and real-captured data, and the appendix gives substantial implementation detail. The paper is also careful to ablate data-specification choices, sampling bases, representation components, and codebook size, and promises open release of code, lens design files, PSF arrays, and simulated/real images. However, the central scalability claim relies on a proxy that is partly used both to construct and to validate the library; several quantitative claims in the paper have internal inconsistencies; and the arXiv abstract claims a few-shot adaptation capability that is not evaluated in the body. These issues do not invalidate the overall framework, but they do require correction and additional validation before the paper can be accepted.","major_comments":[{"comment":"The arXiv metadata abstract states that the framework 'unlock[s] highly efficient few-shot adaptation for unseen lenses.' The body contains no few-shot experiments: Section 4 reports only zero-shot evaluations, and Section 5 explicitly lists 'a flexible fine-tuning pipeline is urged' as future work. This is a load-bearing capability claim with no supporting evidence. Either add a few-shot adaptation experiment (e.g., fine-tuning on a small number of real-lens pairs) or remove the claim from the abstract.","section":"Abstract / Sections 4-5"},{"comment":"The percentage improvements in Table 8 are internally inconsistent. For LPR at 1% scale, PSNR decreases from 23.82 to 23.39 (about -1.8%), yet the table reports '↓10.41%'; at 10% scale PSNR increases from 25.23 to 25.72 (about +1.9%), not '↑10.67%'; at 100% scale PSNR increases from 25.92 to 26.66 (about +2.9%), not '↑15.67%'. The reported percentages appear to have been copied from the LPIPS column or computed by an unstated formula. Because this table is the primary evidence for the claim that LPR leverages AODLibpro scalability, the numbers must be recomputed and corrected.","section":"Section 4.4, Table 8"},{"comment":"The claim that AODLibpro 'uniformly covers' optical-degradation space is validated only through the same OIQ/OD-Class proxy used to construct the library. Figure 5(a) is a histogram over OD-Class after stratified sampling, so near-uniformity is expected by construction, and Figure 5(b) visualizes coverage with per-FoV/wavelength OIQ, i.e., the same metric. This is partly tautological. The authors should provide independent evidence of diversity in PSF/degradation space (e.g., PSF shape statistics, Zernike coefficients, or generalization across a larger set of held-out real lens designs) or explicitly qualify the coverage claim. RealLens-Sim is a useful independent check but contains only 10 lenses and cannot by itself establish the scalability conclusion.","section":"Section 3.2, Section 4.4, Figure 5"},{"comment":"The OD severity partition into Strong/Medium/Mild classes is shown only as a schematic figure with no numerical thresholds, and the OD-Class uniformity threshold is stated as 'α = 0.85 empirically.' Since the entire hybrid sampling basis depends on these values, the exact intervals and a sensitivity analysis (or at least a justification) for α are needed for reproducibility and for assessing how robust the uniformity claim is to the cutoff choices.","section":"Appendix D.2, Figure 7"}],"minor_comments":[{"comment":"The arXiv title ('...Discrete Degradation Priors') differs from the full-text title ('OmniLens++: Blind Lens Aberration Correction via Large LensLib Pre-Training and Latent PSF Representation'), and the arXiv abstract introduces the model as FoundCAC while the body consistently uses OmniLens++. This version mismatch should be resolved before publication.","section":"Title/Abstract vs. main text"},{"comment":"All quantitative results are single-run averages without standard deviations, confidence intervals, or significance tests. Several claimed improvements over the next best method are below 0.3 dB, so reporting multiple seeds or per-lens variability would materially strengthen the SOTA claims.","section":"Tables 2-3 and 12"},{"comment":"The text states that OmniLens++ 'performs better overall' on RealLens-Snap, but Table 12 is mixed: e.g., on Single-Lens-I the Universal IR model has higher CLIPIQA and MANIQA, and on several rows NIQE favors another method. An aggregate or paired comparison would be more convincing than per-lens three-metric lists.","section":"Appendix G.5, Table 12"},{"comment":"The OD-Class table has rendering artifacts ('US>?', '???_???(???)') that obscure the classification criteria, and Table 8 contains a typo ('Sclae'). Please clean up the figure and text.","section":"Appendix D.2, Figure 8"}],"recommendation":"major_revision","confidential_remarks":"The paper appears to be a merged/mismatched version: the arXiv metadata abstract, the full-text title, and the body use different model names, and the metadata abstract contains a few-shot claim absent from the experiments. The central difficulty is that the data-scalability claim is validated with the very OIQ/OD-Class proxy used to build the data; however, the independent RealLens-Sim results and the LPR ablations suggest the framework is promising. With corrected statistics, an explicit removal or evaluation of the few-shot claim, and an independent diversity analysis, the paper could become publishable. The editor may wish to check whether the manuscript version under review matches the arXiv record."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a legit extension of the OmniLens line, with real contributions on the data side (AODLibpro) and model side (LPR), but the central 'uniform coverage' claim is partly self-confirming, and the paper has enough—missing error bars, unreleased data, an abstract that overclaims few-shot—to keep it from being a clean pass.\n\nWhat's actually new: they add aspheric surfaces and image-distance perturbation to the EAOD lens generation, define OIQ (hand-weighted PSNR/SSIM/SFR composite) and an OD-Class taxonomy (six spatial-variation patterns), then use those to sample 3600 training lenses in 18 subclasses. That's a real construction, and it yields measurable gains on the RealLens-Sim suite—0.71 dB over OmniLens baseline from the data, 0.81 dB from LPR, and the full model beats prior LensLib-PT methods on severe aberration cases. The LPR idea—train a VQVAE on PSF maps with an ODN forward-model constraint, freeze the codebook, then predict latent PSF features to guide the restoration network—is well-motivated and the ablations support it. The benchmark AODLibproTest is a useful addition. I believe the empirical story: bigger balanced library plus PSF-latent guidance helps.\n\nSoft spots, in order of importance:\n\n1. The uniformity of AODLibpro is measured with the same OIQ/OD-Class proxy used to build it. Figure 5(a) plots the histogram of OD-Class—the very categories you sampled to be uniform. Figure 5(b) uses OIQ per FoV/wavelength. So the evidence that you've covered 'all possible OD patterns' is partly tautological. The stress-tester's point lands: independent PSF-space diversity is never shown. It's not fatal, because the data still gives empirical gains over a wider range of real lenses, and the appendix's discussion of chromatic aberration is honest, but the scalability claim is weaker than the text implies.\n\n2. No error bars on any metric. Single-run metrics on two benchmarks; differences of 0.3–0.8 dB can flip with seeds. For a foundation-model claim, this needs fixing.\n\n3. Code and data are promised but not released, and the GitHub link is inconsistent with the full-text header. The abstract also claims 'highly efficient few-shot adaptation' but the body doesn't show a single few-shot experiment—it's listed as future work. And the arXiv metadata title/author list doesn't match the full-text title/authors (OmniLens++/Jiang et al. vs the entry title/Qian et al., with FoundCAC appearing in the abstract). That's sloppy and confusing.\n\nWho this is for: people working on blind aberration correction or lens design-aware restoration. They'll get a solid template for data construction and a good model paradigm. It deserves a serious peer review; the ideas are coherent, the experiments are broad, and the methods are described well enough to reproduce once artifacts drop.\n\nRecommendation: send it to referees, but ask for (a) code/data release or clear justification for delay, (b) variance or significance estimates on main tables, (c) alignment of abstract/title/metadata and either few-shot results or removing the claim.","headline":"Solid incremental advance in blind aberration correction, with a useful data library and PSF-latent guidance, but the uniformity claim is partly self-confirming and reproducibility/framing issues need attention.","tokens_in":27494,"tokens_out":4296,"would_cite":true,"duration_ms":33596,"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 single blind network can correct severe, spatially varying lens aberrations across minimalist optics, metalenses, misaligned lenses, and high-end DSLRs without per-lens PSF calibration.","keywords":["blind lens aberration correction","computational aberration correction","lens library pretraining","point spread function","vector quantization","spatially varying degradation","zero-shot generalization","image restoration"],"falsifier":"Cluster AODLibpro's 3,600 training lenses by their full PSF maps rather than by OIQ summaries: if some of the eighteen OD subclasses contain near-duplicate PSF kernels while sizable PSF-shape clusters fall outside them, the hybrid sampling basis has not balanced the true degradation space and the reported scalability gain should not transfer to the missing patterns. A single counterexample lens—classified as 'spatial-uniform' but with PSFs that alternate between sagittal- and tangential-dominated lobes across the field—would directly expose the taxonomy's insufficiency.","tokens_in":26393,"feed_emoji":"📷","tokens_out":9565,"duration_ms":85297,"temperature":0.7,"pith_summary":"This paper asks whether blind lens aberration correction—restoring images from unknown lenses without per-lens PSF calibration—can generalize across the full range of real optics. The paper answers yes, provided the training lens library is balanced in both degradation severity and spatial-variation pattern, and provided the correction network is guided by a latent discrete prior learned from PSFs. On the data side it builds AODLibpro, a 3,600-lens library sampled uniformly over eighteen severity-by-pattern classes, adding aspheric surfaces and image-plane perturbation to widen coverage. On the model side it proposes LPR, a vector-quantized codebook of PSF features regularized by an optical-degradation network, so a blind encoder can pull degradation priors out of the image itself. If right, one pretrained model can replace per-lens calibration for cheap minimalist optics and high-end lenses alike.","feed_headline":"Blind model fixes unseen lens blur, from singlets to DSLRs","feed_subtitle":"A lens library sampled for uniformity plus a discrete PSF prior gives zero-shot correction across real lenses.","key_machinery":"Two coupled objects carry the argument. AODLibpro is the data engine: a lens library generated with enriched optical design specifications and sampled with a hybrid basis crossing three severity classes (OIQ) with six spatial-variation classes (OD-Class), giving eighteen subclasses and a deliberately uniform aberration distribution. LPR is the guidance engine: a vector-quantized autoencoder turns PSF maps into a discrete codebook of latent PSF features, while an Optical Degradation Network—conditioning a clear-image encoder on quantized PSF codes to reproduce the degraded image—regularizes the codes to carry optical meaning. At inference a separate encoder predicts latent features from the d","core_discovery":"The paper establishes that the bottleneck in blind aberration correction is not model capacity alone but two design choices: the distribution of the lens library and the form of degradation guidance. It constructs AODLibpro by enriching the lens-source specifications and sampling uniformly over severity (a weighted PSNR/SSIM/SFR composite) and spatial-variation class (six patterns defined from per-FoV quality trends). It then learns LPR, a discrete codebook of latent PSF features via vector-quantized autoencoding, supervised by an Optical Degradation Network that must use the code to degrade a clear image into the observed one. During correction, a UNet-style model predicts and retrieves cod","pith_inferences":["The LPR codebook could be reused as a blind descriptor of optical behavior, not just a correction guide: since its entries track per-field degradation patterns, clustering unknown lenses by their retrieved codes would partition optics by effect rather than by design type.","If the approach transfers, optical designers could ship cheaper, aberration-heavy optics and let a universal correction model absorb the quality loss—a different cost/quality trade-off for minimalist and metalens systems than the paper spells out.","A direct stress test would apply the same frozen codebook to PSF structures outside the generated lens-source space (e.g., metasurfaces or diffractive elements), which the paper names as future data extensions but does not claim to cover.","The OIQ weights are hand-set; recomputing AODLibpro with different weights or a learned perceptual metric would probe how much of the scalability result rests on the specific proxy."],"forward_implications":["A single blind model can correct severe aberrations from minimalist spherical/aspheric lenses and metalenses, handle stochastic misalignment blur, and improve high-end DSLR images without per-lens PSF calibration at inference.","Scaling a lens library pays off only when samples are uniform in degradation severity and spatial-variation pattern; the paper finds larger improvements from 1% to 100% data with the new library than with the previous RMS-sampled one.","LPR guidance beats direct PSF prediction, PSF-feature prediction, and variants with only the vector-quantized codebook or only the degradation network, and its benefit grows as the library scales.","The frozen discrete prior makes few-shot adaptation to a new lens efficient, since the codebook transfers instead of being retrained per lens.","Predicted latent PSF features are discriminative across degradation patterns—the attention maps align with per-field degradation—supporting the claim that the guidance is optical rather than categorical."],"fun_headline_variants":["Lens library pre-training enables zero-shot blind aberration correction","Discrete PSF priors: key to blind lens correction across real lenses","From library to sharp: discrete priors for unknown lens blur","Few-shot lens adaptation via codebook-freezing and a sampled library"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that the hand-weighted PSNR/SSIM/SFR composite (OIQ) and the six-category spatial-pattern taxonomy (OD-Class), with its empirically chosen threshold, faithfully capture the optical-degradation structure that matters—if real PSF variations such as chromatic structure or non-monotonic field patterns slip through both, the library is not truly uniform and the scalability conclusions weaken.","fun_headline_variants_meta":{"raw":{"variants":["Lens library pre-training enables zero-shot blind aberration correction","Discrete PSF priors: key to blind lens correction across real lenses","From library to sharp: discrete priors for unknown lens blur","Few-shot lens adaptation via codebook-freezing and a sampled library"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001227,"raw_usage":{"total_tokens":4909,"prompt_tokens":802,"completion_tokens":4107,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":546,"completion_tokens_details":{"reasoning_tokens":4033}},"tokens_in":546,"tokens_out":4107,"duration_ms":24212,"temperature":1.0,"reasoning_tokens":4033,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T20:58:09.798939+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Cluster AODLibpro's 3,600 training lenses by their full PSF maps rather than by OIQ summaries: if some of the eighteen OD subclasses contain near-duplicate PSF kernels while sizable PSF-shape clusters fall outside them, the hybrid sampling basis has not balanced the true degradation space and the reported scalability gain should not transfer to the missing patterns. A single counterexample lens—classified as 'spatial-uniform' but with PSFs that alternate between sagittal- and tangential-dominated lobes across the field—would directly expose the taxonomy's insufficiency.","supporting_citations":[],"review_version":1}