{"id":"efe5c113-6000-4d1b-a689-072f0d3ebe2e","arxiv_id":"2412.02393","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":5,"one_line_summary":"A CNN that regresses neighbor density over distance is proposed for UAV swarm relative localization, with evaluation on synthetic and a small set of real images.","lead":"This paper proposes a camera-based method for UAV swarms that estimates how many neighbors are at each distance, instead of detecting each neighbor individually. The authors argue this scales to large swarms and is more robust to the target's pose, though the reported comparisons contain internal contradictions.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Table I contradicts the key accuracy claim: SotA detector shows lower close-range error E' and total integral error T than the proposed method, and the promised swarm control algorithm is absent.","rationale":"I read the manuscript in good faith. The core idea, regressing a distance-density histogram, is interesting and the synthetic high-density experiment suggests a possible scalability advantage. However, the load-bearing weakness is internal inconsistency: the paper's primary claim of superior distance accuracy is directly contradicted by its own Table I, regardless of sim-to-real transfer concerns. The reader's rationale also noticed this contradiction and the missing control algorithm, but the reader's stated weakest assumption focused on the sim-to-real gap. I focus on the internal contradiction because it is more fundamental: if Table I is accurate, the main claim fails even in simulation, making external validation moot. The concrete test is a simple recomputation from the released dataset, which should settle the matter. If the table were somehow misreported, the sim-to-real concern would become the most important, but as printed the internal inconsistency is decisive. Therefore I agree with the REJECT verdict, but for a different primary reason than the reader's stated weakest assumption.","tokens_in":10854,"tokens_out":2818,"duration_ms":31000,"concrete_test":"Recompute the E' and T columns of Table I from the released dataset using the metric definitions in Eqs. (8)-(10) for the rows 'SotA detector' and 'Ours (1x1)', including the detector outputs and nearest-neighbor distance assignment. If the detector's E' remains below 1.25 and its T below 0.142, the paper's central accuracy claim is refuted by its own results.","verdict_should_be":"REJECT","load_bearing_attack":"The central claim is that regressing neighbor density over distance gives 'more accurate distance estimation' and 'better scalability' than per-agent detection. The direct evidence for this is the comparison with detector baselines in Table I. Under the metric definitions in Eqs. (8)-(10), lower error is better. Table I reports E' (close-range per-bin error) as 0.93 for the SotA detector versus 1.25 for Ours (1x1) and 1.27 for Ours (3x3); T (total integral error) is 0.02 versus 0.142 and 0.174. The text asserts 'Our method outperforms the state-of-the-art detector using all metrics,' but the table does not support this. This is not a matter of preference: if the table and equations are correct, the proposed method is less accurate at close distances and has higher total misdetection error, directly undermining the abstract's 'more accurate distance estimation' claim. The high-density experiment (Section IV-D) only uses synthetic data, so it cannot establish real-world scalability. Additionally, the abstract promises 'a novel swarm control algorithm,' yet no control law is presented in Sections II-V; the conclusion explicitly defers it to future work. Thus the paper's headline contributions are contradicted or missing on its own evidence.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes replacing per-UAV detection with a convolutional neural network that regresses a histogram of neighbor density over distance from a focal UAV, inspired by animal swarms. The authors claim that this yields more accurate distance estimation than detector-based localization, better scalability with swarm size, and compatibility with a novel swarm control algorithm. Evaluation is performed on a synthetic dataset of 16,000 images (Sim2Air-based), an additional synthetic high-density dataset with up to 150 UAVs, and a set of 40 real-world photos. Baselines include a YOLOv4 Tiny/EfficientDet-style state-of-the-art detector, an ideal detector, and several ablations of the proposed architecture and label smoothing.","tokens_in":11156,"tokens_out":4103,"duration_ms":48675,"significance":"The density-over-distance formulation is an interesting and potentially useful idea for large heterogeneous swarms, and the paper has concrete strengths: the dataset is released, the synthetic pipeline is described in some detail, the comparison includes a hypothetical ideal detector, and the high-density scaling experiment targets a real gap in the literature. If the claims were supported, the paper would provide a lightweight vision pipeline for swarm relative localization. However, the paper's own Table I contradicts the headline accuracy claim, the promised swarm control algorithm is absent, and the real-world validation is qualitative and tiny. As a result, the central claims are not established by the evidence presented.","major_comments":[{"comment":"The statement that 'Our method outperforms the state-of-the-art detector using all metrics' is contradicted by the paper's own Table I. Under Eqs. (8)-(10), lower values of T, E, and E' are better. Table I reports for the SotA detector T = 0.02, E = 1.31, E' = 0.93, while Ours (1x1) reports T = 0.142, E = 1.31, E' = 1.25, and Ours (3x3) reports T = 0.174, E = 1.30, E' = 1.27. The proposed method therefore has substantially higher total integral error and higher close-range error than the detector baseline. This directly undermines the abstract's claim of 'more accurate distance estimation.'","section":"§IV-C, Table I"},{"comment":"The abstract and contributions list promise 'a novel swarm control algorithm' that is compatible with the density regression, but no control law, closed-loop simulation, or stability analysis appears in Sections II-V. The conclusion instead states: 'In the future, we aim to design a swarming rule intended to utilize the proposed relative localization method.' The paper therefore advertises a central contribution that is missing, and the claim that the method is 'suitable to be used as the main source of relative localization for swarm stabilization' is not supported by any closed-loop experiment.","section":"Abstract, §I-B, §V"},{"comment":"The real-world generalization evaluation is not strong enough to carry the load assigned to it. The dataset consists of 40 manually annotated photos; 30 are used for fine-tuning and only 10 for testing. The text itself describes the analysis as 'qualitative,' and no quantitative distance-error metric is reported on these 10 images. Since the central scalability and robustness claims are based on synthetic data, the sim-to-real transfer question is effectively open. A quantitative real-world comparison with the detector baseline is needed before the method can be claimed suitable for real swarm stabilization.","section":"§IV-E, §III-B"},{"comment":"There is a partial circularity in the evaluation setup: the loss function is weighted to emphasize close distances, the dataset cropping is biased toward images with close UAVs, and the summary metric E' is defined specifically as close-distance per-bin error. Any method trained with these choices is expected to perform better on close-range metrics than an untuned detector baseline. Table II's high-density experiment also trains and tests only on synthetic data generated by the same pipeline, so it does not demonstrate scalability to real dense swarms. The paper should separate 'performance by construction' from an independent measure of task accuracy, for example by reporting metrics over the full distance range with the training distribution ablated, or by evaluating on real high-density footage.","section":"§II-B, §III, §IV-B, Eq. (10)"},{"comment":"No error bars, confidence intervals, or repeated-run statistics are reported for any of the trained models. Several of the claimed differences are small (e.g., E = 1.31 for both SotA detector and Ours 1x1), and the central accuracy comparison relies on numbers that could plausibly change with random seeds, initialization, or hyperparameter choices. The authors should provide multiple training runs or statistical measures before claiming superiority on any metric.","section":"Tables I and II"}],"minor_comments":[{"comment":"The training setup paragraph says 'Stochastic Gradient Descent (SGD) back-propagation' but then states 'We used the ADAM optimizer.' This is inconsistent; please specify the optimizer correctly.","section":"§II-C"},{"comment":"The ablation study references 'partially smoothed labels lgt[d] (eq. (2))', but Eq. (2) defines the last-bin truncation, not the smoothed label. The correct reference is Eqs. (3)-(4).","section":"§IV-B"},{"comment":"The identity of the 'SotA detector' is unclear: §IV-C introduces a YOLOv4 Tiny detector, while the surrounding text mentions EfficientDet, and the table caption does not specify which detector is reported. This needs to be resolved for reproducibility.","section":"§IV-C, Table I"},{"comment":"The real-world photos are described as containing '1 to 8 UAVs,' but the paper does not report how the density ground truth was obtained for these images; the nearest-neighbor distance estimation method is mentioned, but the annotation procedure should be described more concretely.","section":"§IV-E"},{"comment":"Several reference entries are marked as 'submitted to T-RO' rather than as published or accepted works (e.g., [5], [7]); these should be updated or marked as preprints to avoid implying peer-reviewed status.","section":"General"}],"recommendation":"reject","confidential_remarks":"The paper's own results contradict its central accuracy claim, and a promised control contribution is missing. A revision would require not only editing claims but rerunning experiments and adding a substantive control component plus quantitative real-world evaluation, which goes beyond the scope of a normal revision. The density-regression idea itself is worth further study, but this manuscript does not establish its two headline claims."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Karel,\n\nYou should know this paper is a mixed bag. The core idea—regress a neighbor density over distance instead of detecting individual drones—is genuinely new as far as I can tell, and it's an interesting direction for large swarms. The authors build on crowd counting but output a 1D histogram over distance bins, which is a sensible representation for collision avoidance and is more scalable than per-agent detection. They also release the dataset and labels, and the network is small and fast.\n\nThe major problem is that the paper's headline claim does not hold up against its own numbers. In Table I, the SotA detector has T=0.02 and E'=0.93, while the proposed method (Ours 1x1) has T=0.142 and E'=1.25. Lower is better for both metrics. So the detector is better on total integral error and close-range per-bin error. The text says 'Our method outperforms the state-of-the-art detector using all metrics,' which is not true based on the table. The only metric where they tie is the overall per-bin error E (1.31 vs 1.31). This isn't a side issue; it's the central comparative claim of the paper.\n\nSecond, the abstract promises a novel swarm control algorithm, but the conclusion explicitly defers it to future work. There is no control algorithm in the paper. That's an overpromise and inflates the contribution.\n\nThird, the real-world validation is thin: 40 photos, 30 used for fine-tuning, 10 for qualitative testing, with no quantitative metric reported. The high-density experiment (up to 150 UAVs) is promising but only synthetic, so it can't establish real-world scalability.\n\nThere's also a mild circularity concern: the loss weighting and dataset cropping are tuned to favor close-range performance, and the evaluation metrics E' focus on close ranges. That said, the proposed method doesn't actually win on E' in Table I, so this is secondary.\n\nIf the authors can fix the claims—correct the Table I interpretation, add real quantitative real-world results, and either present the control algorithm or drop it from the abstract—the underlying idea is worth keeping. As it stands, the paper is not reliable in its current form, but it shouldn't be tossed aside either.\n\nI'd send it to peer review, but with the expectation of major revision. The problem isn't that the idea is bad; it's that the paper oversells it. For a reader working on swarm perception, this is worth a look, but cite it cautiously.\n\nBest,\n[Your name]","headline":"Fresh idea, but the paper's own Table I contradicts its central accuracy claim and the promised control law is missing.","tokens_in":11681,"tokens_out":5584,"would_cite":false,"duration_ms":54381,"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 CNN that regresses neighbor density over distance outperforms per-drone detection for UAV swarm relative localization.","keywords":["UAV swarms","relative localization","density over distance","monocular vision","convolutional neural network","swarm stabilization","sim-to-real transfer","collision avoidance"],"falsifier":"Fly a real multi-drone experiment with ground-truth positions from RTK GPS or motion capture, run the trained density regressor on the camera feed, compute the per-bin error $\\bar{e}[d]$ and total integral error $\\bar{T}$, and compare them with the synthetic test results in Table I; a large degradation at close ranges after fine-tuning on 30 real photos would refute the sim-to-real transfer claim.","tokens_in":10725,"feed_emoji":"🚁","tokens_out":14707,"duration_ms":141544,"temperature":0.7,"pith_summary":"The paper claims that a UAV in a large swarm can localize its neighbors more accurately by regressing how many neighbors are at each distance in its camera image than by detecting and estimating the pose of each neighbor individually. This density-over-distance output is a small histogram with one bin per distance range, and it is insensitive to the number of visible neighbors and to each neighbor's orientation, allowing the perception network to be smaller and faster. The authors show on synthetic imagery with up to 150 drones that their regressor has lower per-bin distance error than a conventional bounding-box detector and even than a hypothetical perfect detector, and they report qualitative transfer to real photos after fine-tuning on 30 images. If the claim holds, dense collision-avoiding swarms become feasible with lightweight onboard monocular cameras.","feed_headline":"Density beats detection for drone-swarm localization","feed_subtitle":"One small CNN estimates neighbor distance distribution, scales to 150 drones, and handles tilted views.","key_machinery":"The central object is the distance-density histogram $l_{\\mathrm{raw}}[d]$ from eq. (1): the count of UAVs in each distance bin $[d, d+\\Delta d)$, capped at $d_{\\max}$ by eq. (2). The network is trained against the partially smoothed label $l_{\\mathrm{gt}}[d] = (l_{\\mathrm{raw}} * g)[d]$, where $g[d]$ is the identity kernel for the closest $k$ bins and a Gaussian kernel $G_\\sigma[d]$ otherwise; smoothing preserves total count while forgiving small distance misassignments. A convolutional feature extractor feeds a $1\\times1$ convolutional layer that scores feature maps per distance bin, followed by average pooling and a fully connected layer, and the weighted Euclidean loss in eq. (7) gives larger weights to close bins so collision-relevant distances are learned despite their underrepresentation in the dataset. This machinery converts an image into a sparse distribution that scales with swarm size and avoids per-target pose estimation.","core_discovery":"The central discovery is that relative localization for swarm stabilization can be solved as a one-dimensional regression of neighbor counts over distance instead of a multi-object detection problem. For each output grid cell the network predicts $l_o[d]$, an estimate of $l_{\\mathrm{raw}}[d]$, the number of UAVs whose position in the camera frame satisfies $\\lVert p \\rVert \\in [d, d+\\Delta d)$. Training uses partially smoothed labels $l_{\\mathrm{gt}}[d] = (l_{\\mathrm{raw}} * g)[d]$ with a Gaussian kernel for distant bins and an identity kernel for the closest bins, plus a weighted Euclidean loss that emphasizes close distances. The paper reports that this representation removes the inherent bounding-box bias caused by target pose, keeps error low at high neighbor counts, and is suitable as the primary input to swarm stabilization.","pith_inferences":["Editorial inference: the density-over-distance output could be accumulated over time and across multiple cameras, giving a lightweight sensor-fusion front end for swarm state estimation without requiring data association.","Editorial inference: the approach should be naturally tolerant of partial occlusion and motion blur, since an occluded or blurred neighbor shifts a count between nearby distance bins instead of vanishing; augmenting the synthetic dataset with such degradations would test this directly.","Editorial inference: the distance-bin weight vector in the loss is a direct control lever: if a collision-avoidance potential is specified, the weights could be optimized to minimize expected collision risk rather than uniform per-bin error, coupling perception and swarm control more tightly than the paper does."],"forward_implications":["The same trained network handles images containing from 1 to 150 drones, with per-bin error on the high-density test set lower than at lower densities, so swarm size is no longer a bottleneck for perception.","A drone observing a tilted or rotated neighbor no longer suffers the systematic distance bias that bounding-box width and height estimates produce, because the regressor learns distance from visual features directly.","The network's small parameter count and 18 ms per-frame inference on a laptop GPU make it deployable onboard SWaP-constrained UAVs as the main relative-localization source.","Because the output is a distribution rather than discrete detections, the method can be fine-tuned to real-world imagery with only tens of labeled photographs, and the paper shows qualitative improvement from 30 fine-tuning images."],"supporting_citations":[{"why":"It supplies the synthetic aerial image-generation pipeline that the paper extends from single drones to groups of up to 150.","marker":"[22]"},{"why":"It is the crowd-counting work that motivates training on Gaussian-smoothed density labels and a relative-count loss.","marker":"[15]"},{"why":"It provides the density-map CNN design that motivates using a one-by-one convolution to weight feature maps per distance bin.","marker":"[19]"},{"why":"It defines the deep convolutional feature extractor used as the network's perception head.","marker":"[18]"},{"why":"It is cited as a representative bounding-box detection CNN whose per-object output the density approach is compared against.","marker":"[4]"},{"why":"It is mentioned as another state-of-the-art detection architecture of the kind that the proposed method outperforms.","marker":"[2]"}],"fun_headline_variants":["Neighbor density, not detection, stabilizes UAV swarms","Density regression outperforms detection for swarm localization","One CNN maps neighbor density for drone-swarm control","Swarm stability via distance-density regression, not detection"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the synthetic training images are realistic enough that a network fine-tuned on a few dozen real photos will estimate distance distributions on actual drone imagery about as accurately as it does in simulation.","fun_headline_variants_meta":{"raw":{"variants":["Neighbor density, not detection, stabilizes UAV swarms","Density regression outperforms detection for swarm localization","One CNN maps neighbor density for drone-swarm control","Swarm stability via distance-density regression, not detection"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000596,"raw_usage":{"total_tokens":2745,"prompt_tokens":856,"completion_tokens":1889,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":472,"completion_tokens_details":{"reasoning_tokens":1825}},"tokens_in":472,"tokens_out":1889,"duration_ms":15879,"temperature":1.0,"reasoning_tokens":1825,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T23:30:40.236194+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Fly a real multi-drone experiment with ground-truth positions from RTK GPS or motion capture, run the trained density regressor on the camera feed, compute the per-bin error $\\bar{e}[d]$ and total integral error $\\bar{T}$, and compare them with the synthetic test results in Table I; a large degradation at close ranges after fine-tuning on 30 real photos would refute the sim-to-real transfer claim.","supporting_citations":[{"cited_title":"Sim2Air-synthetic aerial dataset for UA V monitoring,","cited_arxiv_id":null,"evidence_quote":"It supplies the synthetic aerial image-generation pipeline that the paper extends from single drones to groups of up to 150."},{"cited_title":"Crowd counting via scale-adaptive convolutional neural network,","cited_arxiv_id":null,"evidence_quote":"It is the crowd-counting work that motivates training on Gaussian-smoothed density labels and a relative-count loss."},{"cited_title":"Single-image crowd counting via multi-column convolutional neural network,","cited_arxiv_id":null,"evidence_quote":"It provides the density-map CNN design that motivates using a one-by-one convolution to weight feature maps per distance bin."},{"cited_title":"Efficientdet: Scalable and efficient object detection,","cited_arxiv_id":null,"evidence_quote":"It is mentioned as another state-of-the-art detection architecture of the kind that the proposed method outperforms."}],"review_version":1}