{"id":"362ee0d1-3236-4565-b099-77aace1a5525","arxiv_id":"2508.11932","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":4.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"A survey of deep learning point cloud denoising, proposing a taxonomy of outlier removal and surface restoration.","lead":"This paper reviews deep learning methods for removing noise from 3D point clouds. It is presented as the first comprehensive survey, organizing the field into outlier removal and surface restoration.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Unsubstantiated claim of being the first comprehensive survey; no search methodology, and the provided full text is corrupted and mismatched, so comprehensiveness cannot be verified.","rationale":"The reader identified the weakest assumption as the completeness of the survey, specifically the negative existential claim that no comprehensive survey exists. I agree with this. The abstract's 'to our best knowledge' is a clear admission of uncertainty, and no search methodology is provided to support the claim. The corrupted full text exacerbates the problem because it prevents any substantive check of the survey's content, but it is not a separate concern; it is a barrier to resolving the completeness question. My proposed concrete test—clean-text recovery plus a systematic literature search for prior surveys—directly targets the negative existential claim and the coverage completeness. If the test fails, the central contribution collapses. Since the reader's verdict was UNVERDICTED due to the inability to assess the paper, and my concern aligns with that, the verdict should remain UNCHANGED. I am not proposing a different verdict because the current evidence does not support accepting or rejecting; it supports remaining 'unverdictable' until the paper is examined in a readable form and the completeness claim is checked against the literature.","tokens_in":9331,"tokens_out":2886,"duration_ms":31759,"concrete_test":"Obtain a clean full text from the arXiv LaTeX source (e.g., via the e-print). Then perform a systematic literature search across Google Scholar, Scopus, DBLP, and arXiv for surveys of point cloud denoising published before August 16, 2025 (e.g., query 'point cloud denoising survey'). If any comparable comprehensive survey is found, the first-survey claim is false. Additionally, compare the recovered paper's reference list against a manually curated list of known DL-based point cloud denoising methods (e.g., PointCleanNet, ScoreDenoise, etc.) to test coverage completeness.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central contribution is the claim to be the first comprehensive survey of deep-learning-based point cloud denoising, with a two-step taxonomy (outlier removal and surface noise restoration). For this to hold, two conditions must be met: (a) no prior comprehensive survey exists, and (b) the survey actually covers all significant existing methods. Neither is established. The abstract only states 'to our best knowledge,' which is a hedge, not a demonstration. No search strategy, inclusion criteria, or database list is provided, so the negative existential claim is unsupported. Furthermore, the full text as provided is unreadable due to a character-encoding error and contains a header for a different arXiv paper (arXiv:2508.11933v1 [cs.CL]), so the proposed taxonomy, method comparisons, and discussion cannot be inspected. Even if the full text were clean, the completeness claim would require external verification, but the current document makes any internal check impossible. This is load-bearing because if a prior comprehensive survey exists, or if the survey omits a substantial class of methods, the paper's contribution fails.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper claims to be the first comprehensive survey of deep-learning-based point cloud denoising (PCD). It proposes a two-step taxonomy that separates PCD into outlier removal and surface noise restoration, and states that it compares existing methods, discusses limitations, and outlines future directions. The abstract is readable and makes these claims, but the supplied full text is severely corrupted by an encoding error and includes the header of a different arXiv paper (arXiv:2508.11933 [cs.CL]). As a result, the substantive content of the survey—taxonomy details, method descriptions, comparisons, and discussion—cannot be inspected or verified.","tokens_in":9562,"tokens_out":2608,"duration_ms":31986,"significance":"If the claimed comprehensive survey were fully readable, it would provide a useful structured reference for a rapidly growing area. The proposed two-step formulation is a plausible organizing principle, and a careful mapping of existing methods onto it could be a genuine contribution. However, the current submission does not allow this contribution to be assessed. The paper contains no derivations, predictions, or reproducible artifacts; its value rests entirely on literature coverage and organization, both of which are unverifiable in the provided text. The strength of the abstract is its clear statement of scope and intended taxonomy, but that strength is undermined by the absence of any readable supporting content.","major_comments":[{"comment":"The entire full text is encoding-corrupted (mojibake) and contains the header 'arXiv:2508.11933v1 [cs.CL]', which is a different paper. Consequently, the proposed taxonomy, the method summaries, the comparative tables, and the discussion of limitations and future directions cannot be inspected. Since the paper's central claim is comprehensiveness of a survey, this corruption makes the manuscript unverifiable in its current form and prevents any substantive evaluation.","section":"Full text (all pages)"},{"comment":"The claim to be the first comprehensive survey is a negative existential claim about all prior literature. The abstract provides no search methodology, no inclusion/exclusion criteria, no list of databases, and no time window. The phrase 'to our best knowledge' is an appropriate hedge but does not substitute for a documented literature-search procedure. Unless the unreadable full text contains such a methodology, this load-bearing claim is unsupported.","section":"Abstract"},{"comment":"The two-step formulation—outlier removal and surface noise restoration—is asserted to 'encompass most scenarios and requirements of PCD.' No definitions, illustrations, or evidence are visible in the readable material. Because the taxonomy is the paper's main organizational contribution, its coverage claim needs to be demonstrated with concrete examples and mapping of representative methods; this is not possible with the current corrupted text.","section":"Abstract / Proposed taxonomy"}],"minor_comments":[{"comment":"The phrase 'compare methods in terms of similarities, differences, and respective advantages' is vague. A survey abstract should indicate the comparison axes, e.g., architecture, loss function, noise type, or computational cost.","section":"Abstract"},{"comment":"The text includes the arXiv header of a different submission (cs.CL). The authors should verify that the correct PDF/source was uploaded.","section":"Metadata"},{"comment":"Even where numbers and table-like fragments are visible, the surrounding labels and captions are unreadable. After re-encoding, the authors should ensure that all tables and figures render correctly.","section":"Full text"}],"recommendation":"reject","confidential_remarks":"As submitted, this manuscript is not reviewable because the full text is corrupted and mismatched with the claimed topic. A clean resubmission with a proper search methodology and readable full text could be reconsidered, but the current version cannot support acceptance or even a meaningful revision cycle."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"I couldn't verify this paper as submitted. The full text is not just garbled; it includes a header for a different arXiv paper (cs.CL). So any assessment has to rest on the abstract, which is coherent enough. The two-step framing of point cloud denoising as outlier removal plus surface noise restoration is a sensible way to organize the literature, and if the survey actually carries that through, it would be a useful reference for people entering the subfield. That is the paper's real contribution, and it deserves some credit for it.\n\nThe soft spot is load-bearing and the stress-test note is right about it. The abstract claims this is the first comprehensive survey of deep-learning-based point cloud denoising, but that is a negative existential claim with no supporting search methodology, database list, or inclusion criteria. The phrase 'to our best knowledge' is a hedge, not evidence. Because the body is unreadable, there is no way to check whether the coverage is actually comprehensive or whether the taxonomy omits a significant class of methods. That would be the central thing a referee would need to verify.\n\nI can't comment on the comparisons, the citation pattern, or the limitations discussion because none of that is legible. The reader's low confidence is justified.\n\nWho is this for? Someone looking for a structured overview of DL-based point cloud denoising, if they can get a clean copy. As presented, I would not cite it or trust its completeness claims.\n\nRecommendation: the right move is to ask the authors for a clean PDF or a corrected source, then send it for peer review. A survey with a plausible taxonomy and a claimed first-mover status deserves referee time to check the coverage. But this version is not auditable, so my verdict is a conditional one: review it after the authors fix the text.","headline":"A plausible survey framing with an unsupported 'first comprehensive survey' claim, and the provided text is too corrupted to check anything else.","tokens_in":9909,"tokens_out":1937,"would_cite":false,"duration_ms":26564,"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":"The first systematic survey of deep-learning point cloud denoising organizes the field into two subproblems.","keywords":["point cloud denoising","deep learning","survey","taxonomy","outlier removal","surface noise restoration","3D vision"],"falsifier":"Finding a peer-reviewed systematic survey of deep-learning point cloud denoising published before this one, or locating a published denoising method with reported results that cannot be placed in either the outlier-removal or the surface-restoration category, would undercut the paper's central claim.","tokens_in":9299,"feed_emoji":"🧹","tokens_out":7139,"duration_ms":72541,"temperature":0.7,"pith_summary":"Real-world point clouds carry noise, and denoising is the preprocessing step that makes downstream tasks such as reconstruction, segmentation, and recognition reliable. This paper argues that deep-learning methods now surpass traditional denoising and that the field has grown enough to need a structured overview. It proposes to view denoising as two linked subproblems: removing outlier points and restoring the surface from noisy inlier positions. On that basis, the paper classifies existing methods, compares their similarities and trade-offs, and identifies open challenges and future directions. The contribution is a map of the field, not a new denoising algorithm.","feed_headline":"See deep-learning point cloud denoising as two subproblems","feed_subtitle":"A first systematic survey maps methods and open problems onto that two-part view.","key_machinery":"The organizing device is a task decomposition of point cloud denoising into outlier removal and surface noise restoration. This two-step scheme acts as the taxonomy backbone: every surveyed method is positioned according to which of the two subproblems it targets and how it combines them. It is what lets the survey compare methods that otherwise look architecturally different.","core_discovery":"The central claim is that no prior survey systematically covers deep-learning-based point cloud denoising, and that the field can be organized by a two-step formulation: first detect and remove outliers, then restore surface noise on the remaining points. Working from that definition, the paper builds a taxonomy of deep-learning denoising methods, summarizes the main technical contributions within each category, and compares methods by similarity, difference, and advantage. It also reviews the key challenges that remain and outlines directions for future work. If the coverage holds, researchers gain a single reference that defines the task, orders the methods, and shows where progress is sti","pith_inferences":["The paper does not argue this, but the two-step formulation implies a modular design in which outlier removal and surface restoration are trained separately; methods treating both jointly may be needed when outlier and surface noise are entangled.","The paper does not argue this, but if its coverage is indeed the first, its inclusion criteria become as important as its taxonomy, because later surveys will use it as a baseline.","A testable extension beyond the paper would be to turn the taxonomy into a leaderboard: assign each method to its cell and see whether performance gaps correlate with which subproblem a method emphasizes."],"forward_implications":["Practitioners can use the two-step decomposition to diagnose why a denoising pipeline fails: either outliers survive or the surface estimate is still rough.","New methods can be positioned within the taxonomy by stating which subproblem they address, making the literature easier to compare.","Benchmarking and evaluation can be organized around the two subproblems separately rather than only reporting end-to-end metrics.","The limitations the survey identifies define a checklist of unsolved issues for researchers working on point cloud denoising."],"supporting_citations":[],"fun_headline_variants":["First systematic survey maps deep learning point cloud denoising","Deep learning point cloud denoising: a two-step taxonomy","Survey organizes point cloud denoising into outlier removal and restoration","New taxonomy for deep learning point cloud denoising methods"],"cache_read_input_tokens":2816,"weakest_assumption_plain":"The load-bearing premise is that the survey's literature coverage is complete, so that no earlier systematic survey of deep-learning point cloud denoising exists and every significant method fits the proposed taxonomy.","fun_headline_variants_meta":{"raw":{"variants":["First systematic survey maps deep learning point cloud denoising","Deep learning point cloud denoising: a two-step taxonomy","Survey organizes point cloud denoising into outlier removal and restoration","New taxonomy for deep learning point cloud denoising methods"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000484,"raw_usage":{"total_tokens":2188,"prompt_tokens":670,"completion_tokens":1518,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":414,"completion_tokens_details":{"reasoning_tokens":1449}},"tokens_in":414,"tokens_out":1518,"duration_ms":11247,"temperature":1.0,"reasoning_tokens":1449,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T19:40:54.381013+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Finding a peer-reviewed systematic survey of deep-learning point cloud denoising published before this one, or locating a published denoising method with reported results that cannot be placed in either the outlier-removal or the surface-restoration category, would undercut the paper's central claim.","supporting_citations":[],"review_version":1}