{"id":"a1f53e04-fb31-471c-abbe-a95074372655","arxiv_id":"2507.10006","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":2.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"This review paper compiles 17 public anti-UAV datasets and categorizes recent vision-based detection and tracking methods, then proposes seven future research directions.","lead":"This paper surveys the field of vision-based anti-drone technology, listing 17 public datasets with links and grouping recent detection and tracking algorithms into families. It is a reference for researchers entering anti-UAV computer vision, though its value depends on the accuracy and accessibility of the compiled information.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Dataset statistics in Table I conflict with the body text, and several URLs in Table II are broken; this undercuts the paper's main value as a reliable dataset index.","rationale":"The reader's weakest assumption and my own analysis converge: a survey whose value is compilation accuracy is only as good as its statistics, attributions, and links, and this paper shows verifiable failures in all three. I do not see a deeper problem: the qualitative taxonomy is coherent, the SA and acc metric formulas are presented as parameter-free definitions, and the seven future directions are reasonable. The SA metric is described in Section IV.C as proposed in this paper while also citing [17]; this attribution should be clarified but is secondary to the dataset-index problem. Since the reader's CONDITIONAL verdict already captures this concern and no new issue changes the required action, I recommend the verdict remain UNCHANGED. The errors are fixable, but until they are corrected the paper cannot safely serve as a reliable index, so conditional acceptance is the right call.","tokens_in":31070,"tokens_out":4603,"duration_ms":42974,"concrete_test":"Fetch each URL in Table II as printed (with the embedded spaces removed to form valid URIs) and record which return a live dataset repository page. Independently fetch the official DUT Anti-UAV repository (github.com/wangdongdut/DUT-Anti-UAV) and compare the stated image count and split to Section III and Table I. If the official page reports 10,000 images with the 5,200/2,600/2,200 split, Table I is wrong and must be corrected; if it reports 1,000, Section III is wrong. Finally, verify the first author of reference [96] and the dataset associated with the Table III row labeled ISD-UNet against the cited paper.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central contribution is an index of Anti-UAV datasets and methods; its usefulness depends on accurate statistics and accessible links. In Section III, the DUT Anti-UAV entry states the detection subset has 10,000 images with a 5,200/2,600/2,200 train/val/test split, but Table I reports Image 1,000 and omits the split. One of these is wrong, and a researcher relying on either entry will be misled. Table II also contains several mangled URLs, e.g., github.com/gdpinntit/-anti-interference-and-anti-UA V-dataset and github.com/UA V DetectionThesis/UA V -detection-dataset, which contain embedded spaces and likely do not resolve to the intended repositories. The same table is the backbone of the paper's promise to provide effective links. Additionally, Table III's mapping to the described methods is not fully consistent: the RF-vision fusion method of Xie et al. [80] appears as ISD-UNet, and ISTD-DETR is attributed in the text to Yuan et al. while reference [96] lists Yang as first author. These errors do not destroy the survey's taxonomy, but they undermine the paper's usefulness as a trustworthy starting point for benchmarking, which is the central claim. The weakest link is therefore the internal consistency and accessibility of the dataset index.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript is a survey of vision-based Anti-UAV detection and tracking. It reviews threat categories and sensing modalities, compiles nineteen datasets with descriptions in Table I and public links in Table II, categorizes recent methods into UAV-based detection (mainly YOLO variants) and three tracking families (Siamese, self-attention, vision fusion), reports representative results in Table III, defines evaluation metrics, discusses dataset and method limitations, and proposes seven future research directions. The paper's central value is as a compiled index of datasets, methods, and benchmark numbers.","tokens_in":31219,"tokens_out":4297,"duration_ms":48407,"significance":"If the compilation were reliable, the paper would serve as a convenient entry point for researchers: it gathers recent 2020--2025 datasets, links them, summarizes representative detection and tracking methods, and names concrete future directions such as multimodal fusion and deformation-aware tracking. The paper contains no new experiments or derivations, so its contribution rests on the accuracy and completeness of its tables and citations. That load-bearing premise is currently weakened by several internal inconsistencies in the dataset statistics, URLs, and method attributions. The inconsistencies are correctable, and the underlying taxonomy and future-direction discussion are reasonable; I therefore evaluate this as a fixable manuscript rather than one with a fundamentally flawed central idea.","major_comments":[{"comment":"The DUT Anti-UAV entry is inconsistent: the text states the detection subset consists of 10,000 images with a 5,200/2,600/2,200 train/val/test split, while Table I reports 'Image 1,000' and omits the split. These cannot both be correct. Since Table I is the paper's dataset index, the authors must verify the count against the original source and correct one entry, or explain the discrepancy.","section":"Section III vs. Table I"},{"comment":"The text attributes ISTD-DETR to 'Yuan et al. [96]', but reference [96] is authored by Yang, Wang, Bo, and Wang (Neurocomputing, 2025). The in-text attribution or the reference entry must be corrected. This matters because the paper's contribution is a reliable summary of recent methods, and an author-name mismatch undermines that reliability.","section":"Section IV.A and reference [96]"},{"comment":"The RF-vision fusion method of Xie et al. [80] is described in the text without a model name, but Table III lists an 'ISD-UNet' row under the tracking section with TCR and AP results attributed to [80]. Please clarify whether ISD-UNet is the proposed method name, whether it belongs in the detection or tracking category, and align the text and table so the mapping from method to result is unambiguous.","section":"Section IV.A and Table III"},{"comment":"The text before Eq. (1) says 'a comprehensive evaluation index State Accuracy (SA) is proposed in this paper,' but the equation is explicitly cited to [17] and the surrounding text also cites [17] for the same metric. This is an internal attribution conflict: either remove the local-proposal claim or provide a genuinely new definition. As written, the paper claims novelty for a borrowed metric.","section":"Section IV.C, Eq. (1)"},{"comment":"The paragraph describing SiamDT states 'Author proposed a Anti-UA V tracking algorithm named SiamDT' without naming the authors. In a survey whose contribution is method attribution, an unresolved placeholder is not acceptable; replace it with the actual authors or a proper citation (Table III currently attributes the method to [17]).","section":"Section IV.B.1"}],"minor_comments":[{"comment":"The table header 'Access Data' should be 'Access Date', and several URLs contain embedded spaces that likely break the links, e.g., github.com/gdpinntit/-anti-interference-and-anti-UA V-dataset and github.com/UA V DetectionThesis/UA V -detection-dataset. The authors should re-check the exact URLs and verify that they resolve.","section":"Table II"},{"comment":"There are several typos in the dataset entries: 'he detection subset' should be 'The detection subset', 'ICG-Dron' should be 'ICG-Drone', and 'DJl' appears multiple times where 'DJI' is intended.","section":"Section III and Table I"},{"comment":"The dataset label 'Real ward [43]' in the YOLOv8-E row should be 'Real World [43]' to match Table I and the text.","section":"Table III"},{"comment":"Eq. (4) defines Accuracy, but the sentence introducing Eq. (5) refers to 'The accuracy measure [127]' while Eq. (5) is the standard Precision formula. Please re-label the equations and their introductory sentences so Accuracy and Precision are not conflated.","section":"Section IV.C, Eqs. (4)--(5)"},{"comment":"The categories 'UA V-based Detection methods' and 'UA V-based Tracking methods' are confusing because the methods are vision-based, not UAV-based. Consider renaming them 'Vision-based detection' and 'Vision-based tracking' to match the paper's own framing.","section":"Section II.B and Section IV"}],"recommendation":"major_revision","confidential_remarks":"The paper is a survey with no new experimental results, so its acceptance should hinge on the accuracy and reliability of its compilation. The internal inconsistencies in Tables I--III, the unresolved 'Author' placeholder, and the borrowed-metric novelty claim are all correctable but currently compromise the central stated contribution. I see no grounds for rejection beyond these fixable issues. I recommend major revision rather than minor because the dataset-statistics conflict and broken URLs affect the paper's primary utility as a trustworthy index."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nQuick take: this is a genuinely useful survey of Anti-UAV vision, but its main selling point—a reliable index of datasets and links—is undercut by a set of small, verifiable errors. The authors compile 17 datasets, give a sensible two-family taxonomy (detection vs tracking, with Siamese/self-attention/fusion under tracking), and include metric formulas and seven future directions. For a new researcher, that is a real entry point.\n\nWhere it is solid: the dataset list is broad and recent, the method summaries track the source papers closely enough that a reader can find the originals, and the metric section, including SA and acc formulas, is a useful reference. No new algorithm or dataset, but that is not a flaw for a review.\n\nThe soft spots are exactly where the reader's report points. The DUT Anti-UAV entry gives 1,000 images in Table I but 10,000 in Section III with a 5,200/2,600/2,200 split; one is wrong. Table II has several mangled URLs with embedded spaces (e.g., github.com/gdpinntit/-anti-interference-and-anti-UA V-dataset) that will not resolve. ISTD-DETR is attributed to Yuan in the text but the reference list gives Yang as first author. Table III lists 'ISD-UNet' for Xie et al.'s RF-vision fusion work, which does not match the described method. And the SA metric is first called 'proposed in this paper' then cited to [17] two lines later. None of these are fatal to the taxonomy, but they are not cosmetic either—this paper's value is as a reference, and a reference with internal contradictions and broken links misleads precisely the audience it is meant to help.\n\nThe stress-test note is right; I checked the full text and these errors are present and verifiable. The authors should fix them, and also clarify whether SA is their contribution or a citation.\n\nVerdict: send to peer review, but with the clear expectation that the dataset table and links be corrected before acceptance. The compilation is useful enough to deserve referee time, and the errors are fixable.\n\nBest.","headline":"Useful survey scaffolding, shaky dataset index: fix the stats and links before relying on it.","tokens_in":31854,"tokens_out":2404,"would_cite":false,"duration_ms":25218,"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":"This review compiles vision-based anti-UAV detection and tracking work into a dataset index, method families, and seven open research directions.","keywords":["anti-UAV","UAV detection","UAV tracking","infrared small target","dataset survey","YOLO detectors","Siamese trackers","multimodal fusion"],"falsifier":"Visit the listed repositories and compare each row of Table I and Table II with the primary source: the DUT Anti-UAV row is a concrete test, since Section III says the detection subset contains 10,000 images while Table I lists 1,000, and the printed URL for the Anti2 dataset in Table II contains an embedded space that would break navigation. If several such mismatches or dead links appear, the paper's central claim of providing effective dataset links and reliable statistics fails.","tokens_in":1610,"feed_emoji":"🛸","tokens_out":2046,"duration_ms":79304,"temperature":0.7,"pith_summary":"The paper is a review of computer-vision methods for detecting and tracking unauthorized drones (UAVs). It assembles publicly available RGB, infrared, and multimodal datasets used to train and test such systems, gives their access links, and organizes recent detection and tracking algorithms into families with reported performance figures. The intended service to the field is an index: a researcher facing the anti-UAV problem can use this paper to pick a dataset, choose a method baseline, and see where the open difficulties lie. The paper also defines the evaluation metrics used by these methods and closes by proposing seven future research directions. Its value therefore rests on whether the compiled dataset statistics, links, and method summaries are accurate and representative.","feed_headline":"New survey links 20 anti-drone vision datasets","feed_subtitle":"RGB, infrared, and fused-sensor detection and tracking at a glance, with dataset links and reported scores.","key_machinery":"The load-bearing organizational device is a two-part dataset index: Table I lists each dataset's scene, type and total size, modality, complexity, UAV type, and whether it is multimodal, while Table II lists its source, public link, and access date. On the method side, the organizing device is a four-family taxonomy of detection and tracking approaches, with Table III linking each method to the datasets it was tested on and the results it reports. The review also turns evaluation practice into a shared vocabulary by giving formulas for state accuracy, tracking accuracy, MOTA, AP, mAP, precision, recall, F1, FPS, and target coverage rate. These devices carry the argument because the paper's contribution is precisely the structured assembly of scattered datasets and methods.","core_discovery":"The paper's claim is organizational: the current state of vision-based anti-UAV detection and tracking can be captured by a structured dataset index and a method taxonomy. Table I and Table II list twenty public datasets with their scene types, modalities, sizes, UAV types, and access links, while the accompanying text notes the strengths and limits of each dataset. The method section groups recent work into UAV-based detection, mostly YOLO-family detectors modified with attention mechanisms, lightweight heads, and small-target losses, and UAV-based tracking, divided into Siamese, self-attention or Transformer, and vision-fusion approaches. For each method the paper reports the performance claimed on one or more of these datasets, and it defines the evaluation metrics, including a state accuracy formula for trackers that must handle target disappearance. It then names seven future directions, from balancing real-time operation with accuracy to legal and privacy constraints on anti-UAV systems.","pith_inferences":["As an editorial extension, the same index could evolve into a live benchmark resource that tracks dataset versions, license terms, and link status, since several listed datasets live on institutional pages that may move or change over time.","As an editorial inference, the reported method results come from different training splits and evaluation protocols, so direct cross-method comparison from the tables is unsafe without recalibration on shared subsets; a standardized evaluation protocol would make the survey more actionable.","As an editorial extension, the paper's list of failure modes suggests a convergent design, in which future anti-UAV trackers combine a lightweight detector, a Siamese or attention-based tracker, and a global re-detection module, with each component addressing a different listed weakness.","As an editorial inference, the legal and ethical research direction implies that deployed systems will be evaluated by privacy compliance and authorization rules as much as by detection accuracy, a consequence the paper names but does not develop."],"forward_implications":["A researcher can use the dataset table as a launch point for benchmarking, with each dataset tied to its scene type, modality, and stated difficulty.","New detection papers can be positioned against the YOLO-family improvements the paper summarizes, such as attention modules, small-object detection heads, and lightweight backbones, using the reported mAP and AP values as baselines.","Trackers are separated into Siamese, self-attention, and vision-fusion families, which makes it easier to see which architectural choices address which failure modes, such as occlusion, target disappearance, and small-target drift.","The evaluation metrics section gives a common language for comparing trackers that must report not only localization but also whether the target is absent, a central requirement for realistic anti-UAV scenarios.","The seven proposed directions, including real-time and accuracy balance, multimodal fusion, multi-UAV confusion, deformation modeling, and legal constraints, form a concrete agenda for the next wave of work."],"supporting_citations":[{"why":"Supplies the Anti-UAV410 thermal infrared benchmark and the SiamDT tracker, central to the dataset index and the Siamese tracking category.","marker":"[17]"},{"why":"Provides the Anti-UAV large-scale RGB-IR benchmark that grounds the paper's multimodal dataset entries and fusion discussions.","marker":"[48]"},{"why":"Defines the 4th Anti-UAV competition tracks and evaluation style used by several tracking methods in Table III.","marker":"[55]"},{"why":"Introduces the Anti2 dataset and DotD-YOLOv9-C detector, supporting the detection method table and the anti-interference discussion.","marker":"[56]"},{"why":"Contributes the MMFW-UAV multimodal fixed-wing UAV dataset entry in the dataset index.","marker":"[57]"},{"why":"Supplies the DUT Anti-UAV detection and tracking dataset used as a benchmark by multiple methods in the survey.","marker":"[65]"},{"why":"Represents the lightweight YOLOv11-based detection method whose results the paper reports in Table III.","marker":"[92]"},{"why":"Represents the self-attention and DETR-based detection line with reported SIRST and Anti-UAV410 numbers.","marker":"[96]"},{"why":"Supplies StrongSiamTracker, a representative Siamese tracker with 4th Anti-UAV challenge results in the tracking table.","marker":"[113]"},{"why":"Supplies UTTracker, a representative Transformer-based tracker with Anti-UAV challenge results in the tracking table.","marker":"[127]"}],"fun_headline_variants":["Survey maps 20 anti-drone vision datasets","Anti-UAV vision: 20 datasets and methods","Vision-based anti-drone: datasets and challenges","20 anti-drone datasets for vision tracking","Anti-drone vision survey lists 20 public datasets"],"cache_read_input_tokens":33920,"weakest_assumption_plain":"The paper's usefulness as an index depends on the compiled dataset statistics, method summaries, and public links being accurate and accessible.","fun_headline_variants_meta":{"raw":{"variants":["Survey maps 20 anti-drone vision datasets","Anti-UAV vision: 20 datasets and methods","Vision-based anti-drone: datasets and challenges","20 anti-drone datasets for vision tracking","Anti-drone vision survey lists 20 public datasets"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000227,"raw_usage":{"total_tokens":1448,"prompt_tokens":895,"completion_tokens":553,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":511,"completion_tokens_details":{"reasoning_tokens":481}},"tokens_in":511,"tokens_out":553,"duration_ms":6770,"temperature":1.0,"reasoning_tokens":481,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T17:41:48.623784+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Visit the listed repositories and compare each row of Table I and Table II with the primary source: the DUT Anti-UAV row is a concrete test, since Section III says the detection subset contains 10,000 images while Table I lists 1,000, and the printed URL for the Anti2 dataset in Table II contains an embedded space that would break navigation. If several such mismatches or dead links appear, the paper's central claim of providing effective dataset links and reliable statistics fails.","supporting_citations":[{"cited_title":"A lightweight anti-unmanned aerial vehicle detection method based on improved yolov11,","cited_arxiv_id":null,"evidence_quote":"Represents the lightweight YOLOv11-based detection method whose results the paper reports in Table III."},{"cited_title":"Istd-detr: A deep learning algorithm based on detr and super-resolution for infrared small target detection,","cited_arxiv_id":null,"evidence_quote":"Represents the self-attention and DETR-based detection line with reported SIRST and Anti-UAV410 numbers."},{"cited_title":"Strongsiamtracker: A siamese tracker with dynamic global detection for robust anti-uav tracking,","cited_arxiv_id":null,"evidence_quote":"Supplies StrongSiamTracker, a representative Siamese tracker with 4th Anti-UAV challenge results in the tracking table."},{"cited_title":"A unified transformer based tracker for anti-uav tracking,","cited_arxiv_id":null,"evidence_quote":"Supplies UTTracker, a representative Transformer-based tracker with Anti-UAV challenge results in the tracking table."}],"review_version":1}