{"id":"677648bf-d6ab-4259-9464-d044907a8b7b","arxiv_id":"2512.04883","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A heterogeneous observer-follower tracker achieves ~35 FPS on a Jetson Orin Nano at 1080p with ~3% precision loss versus frame-by-frame YOLO11l detection on UAV tracking benchmarks.","lead":"SDG-Track is a drone-tracking system for small edge computers that pairs a slow high-resolution detector on the GPU with fast optical-flow interpolation on the CPU. It reports about 35 frames per second at 1080p while keeping roughly 97% of the precision of running the detector on every frame.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central real-time claim is unsupported: 35.1 FPS is measured offline (Sec 3.3), but live closed-loop throughput/latency with camera, PTZ, and concurrent GPU/CPU load is never measured.","rationale":"The paper's core numbers are internally consistent and the architecture is plausible: a 1080p YOLO11l observer at ~6.7 FPS plus CPU sparse flow can in principle bridge detection gaps, and the ablations show each module helps. No contradiction with existing consensus; this is a systems-integration claim. The most load-bearing weakness is that the headline throughput and \"30 Hz control\" claim depend on offline processing: Sec 3.3 explicitly scopes Sys FPS as offline rate on recorded streams, but abstract/conclusion treat it as real-time evidence. No latency or live-loop measurements are provided, and on a shared-memory SoC the GPU detector can throttle CPU flow and vice versa. A direct live closed-loop test would settle whether the 35.1 FPS transfers. The manuscript also contains a lorem-ipsum placeholder in Section 1 and does not release the self-collected dataset; these reinforce \"conditional\" rather than \"accept\" but are secondary. Verdict stays CONDITIONAL/UNCHANGED.","tokens_in":9098,"tokens_out":6944,"duration_ms":60877,"concrete_test":"Run the full system closed-loop: stream 1080p from the Hikvision PTZ camera into the Orin Nano, have SDG-Track output gimbal commands, and record (a) actual control-update rate, (b) p50/p95 end-to-end latency from camera frame timestamp to command issuance, and (c) tracking precision on a labeled live sequence (or with a motion-capture ground truth) over ≥5 minutes. If the live update rate is below 30 Hz, p95 latency exceeds ~40 ms, or precision drops significantly below 0.889, the 35.1 FPS/30 Hz claim is not established and the paper should be revised to scope the claim as offline throughput only.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Sec 3.3 defines Sys FPS as \"the average offline state-update/output rate of the pipeline on recorded streams,\" and states the asynchronous detector \"does not throttle the per-frame update loop\" by assumption. Yet the abstract and conclusion translate this into \"real-time\" 35.1 FPS / 30 Hz gimbal-control capability. The paper reports no closed-loop experiment: no end-to-end latency from camera frame to PTZ command, no live camera decoding, no measurement of GPU/CPU contention on the Jetson Orin Nano (shared memory bandwidth/power), and no quantitative live-tracking accuracy. The \"successfully tracks agile FPV drones under real-world operational conditions\" sentence is qualitative; Table 2's self-collected results are offline metrics on recorded frames. Because the core contribution is a deployable edge system, a throughput number computed by replaying video does not establish that the pipeline can sustain 30 Hz control with acceptable latency when detections arrive ~150 ms late and flow must run concurrently with GPU inference. This is the load-bearing gap; the internal arithmetic (0.889/0.915 = 97.2%) is consistent.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes SDG-Track, an observer-follower framework for UAV tracking on edge devices. A high-capacity YOLO11l detector runs on the GPU at native 1080p resolution at low frequency, while a CPU-based sparse optical-flow follower interpolates the target state at high frequency within an adaptive ROI. A recovery module combines Lab/HSV color histograms with geometric constraints to re-acquire the target after loss. Experiments on LRDD v2 and a self-collected dataset report 0.889 precision and 0.647 AUC at 35.1 FPS on a Jetson Orin Nano, retaining 97.2% of the frame-by-frame detection precision of the YOLO11l backbone. The paper claims real-time gimbal control and successful real-world tracking of agile FPV drones.","tokens_in":9378,"tokens_out":4528,"duration_ms":46740,"significance":"If the claims are supported, the architecture is a practical engineering solution to the resolution-speed trade-off for small-UAV tracking on embedded platforms, with relevance to counter-UAV systems. The paper provides public code, an ablation study, and a comparison against baseline trackers; the retention arithmetic (0.889/0.915 ≈ 0.972) is correct. However, the central real-time claim rests on an offline throughput definition, and no closed-loop validation is reported. This substantially limits the current evidence for the paper's headline contribution.","major_comments":[{"comment":"The central real-time claim is not supported by the reported experiment. Sys FPS is explicitly defined as 'the average offline state-update/output rate of the pipeline on recorded streams' (Sec. 3.3), and the statement that the asynchronous detector 'does not throttle the per-frame update loop' is an assumption, not a measurement. No closed-loop experiment is provided: no end-to-end latency from camera frame to PTZ command, no live camera decoding, no measurement of GPU/CPU contention or power/thermal effects on the Jetson Orin Nano. The abstract and conclusion translate this offline number into '35.1 FPS system throughput' and 'real-time gimbal control,' which is unsupported. A closed-loop experiment, or at least a latency/contention measurement on recorded streams, is necessary to substantiate the deployability claim.","section":"Sec. 3.3, Table 2; Abstract"},{"comment":"The comparison reports only aggregate precision and AUC for LRDD v2 and the self-collected dataset, with no per-sequence breakdown, standard deviation, or confidence intervals. The claim that SDG-Track 'retains 97.2% of the frame-by-frame detection precision' is based on a single aggregate comparison to one baseline. The baseline set is also narrow—ByteTrack with two backbones, a Kalman predictor, and KCF; modern lightweight trackers are absent. Per-sequence results would strengthen the robustness claim and clarify whether the aggregate numbers are dominated by easy sequences.","section":"Sec. 3.3 / Table 2"}],"minor_comments":[{"comment":"A long placeholder Latin passage ('Nam dui ligula, fringilla a, ...') appears in the Introduction. This is clearly unfinished text and must be removed before submission.","section":"Sec. 1 (after contributions)"},{"comment":"References [8]–[15] are listed in the bibliography but never cited in the main text. Citation numbering should be checked and corrected.","section":"References"},{"comment":"The adaptive weighting αM_HSV + βM_Lab is mentioned, but the function that maps target saturation to α is not specified. Since this is a core part of the recovery module, a formula or a table of values should be provided.","section":"Sec. 2.3.1"},{"comment":"OTSU segmentation is described as 'without manual tuning,' but a fixed 'strict confidence threshold (set to 0.4)' is introduced. Please clarify whether this threshold is fixed or automatically adapted.","section":"Sec. 2.3.2"},{"comment":"The definition of Sys FPS is clear for frame-by-frame baselines, but for sparse-detection methods it is not fully stated whether the reported value includes the detector's parallel execution time or only the follower/recovery per-frame loop. Make the definition uniform across all compared methods.","section":"Sec. 3.3 / Table 2"}],"recommendation":"major_revision","confidential_remarks":"The manuscript appears to be an incomplete draft: it contains placeholder text and uncited references. The central technical concern is that the headline throughput is measured offline, and the paper's real-time control claim is not backed by closed-loop experiments. If the authors can provide such experiments, or at least a careful latency and contention analysis on the live system, the paper could become acceptable. The architecture is plausible and the code availability is a positive factor."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"What you should know: this is a legitimate engineering-systems paper, not a methodological breakthrough. The new thing is the specific split—GPU YOLO11-l at 1080p as a sparse observer, CPU ROI-constrained sparse optical flow as a high-rate follower, plus a training-free color/geometry recovery module—and the measured precision-retention/throughput trade-off on LRDD v2. Each block is standard, but the integration is real and the ablations support it: adding the follower and recovery modules raises LRDD precision from 0.652 to 0.889, and 0.889/0.915 = 97.2% checks out.\n\nWhat it does well: it profiles the detector trade-off honestly (Table 1), gives a plausible decomposition of compute, compares against reasonable baselines (ByteTrack, Kalman, KCF), and makes code public. The claims about component contribution are backed by the ablation trend.\n\nSoft spots: the biggest is that real-time is asserted but not demonstrated. Sys FPS is defined as offline state-update/output rate on recorded streams (Sec 3.3), and the abstract/conclusion generalize that to \"real-time\" and 30 Hz gimbal control. No closed-loop experiment: no end-to-end latency, no live camera decoding, no measurement of GPU/CPU contention on the Jetson. A detection arriving ~150 ms late in a live loop may not behave like the replayed stream. This is a gap, not a fatal flaw, because the offline number is defensible as stated. Second, there's no variance or per-sequence breakdown, so it's unclear how stable the 35 FPS and 0.889 are. Third, the self-collected dataset is not released and is used only for a qualitative claim; that's weak but not central. Fourth, several thresholds (τ_det, τ_area, α/β, recovery threshold) are hand-set and not sensitivity-analyzed. Fifth, and unrelated to the science: the manuscript contains a lorem-ipsum placeholder paragraph in the introduction (the \"Nam dui ligula\" block). That should have been caught before posting. Also, a direct comparison against a sparse-flow-only baseline (like the cited moving-window flow tracker) would strengthen the case; KCF and constant-velocity Kalman are less relevant competitors.\n\nBottom line: the core result holds on its own terms—offline measured precision retention and throughput are consistent—and the architecture is a plausible deployable template. The live-loop claim needs support. This is worth refereeing; with a closed-loop test and parameter sensitivity it could be a solid venue paper. I'd bring it to reading group as an example of pragmatic edge-systems engineering.","headline":"Useful edge-systems integration with honest ablations and consistent arithmetic, but the 'real-time' claim rests on offline throughput; needs a closed-loop experiment and cleanup.","tokens_in":9937,"tokens_out":1744,"would_cite":true,"duration_ms":17191,"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":"SDG-Track claims that on a Jetson-class edge device, a sparse detector at ~5 Hz combined with ROI-constrained CPU optical flow can sustain 35.1 FPS UAV tracking while preserving 97.2% of the full-frame detector's precision.","keywords":["UAV tracking","edge computing","small object detection","sparse optical flow","observer-follower architecture","tracking-by-detection","re-acquisition","embedded systems"],"falsifier":"A live closed-loop test on the ground-to-air station: track an agile FPV drone and measure the sustained gimbal command rate and end-to-end latency from frame capture to command output; if the live system cannot hold at least 30 Hz updates while keeping precision within about 3% of the offline frame-by-frame detector, the paper's central claim fails.","tokens_in":8948,"feed_emoji":"🛸","tokens_out":7361,"duration_ms":58174,"temperature":0.7,"pith_summary":"The paper tackles the vanishing pixel problem in edge-based UAV tracking: downsampling 1080p frames to detector input size makes distant drones disappear, while processing native resolution is too slow for gimbal control. It proposes SDG-Track, an observer-follower architecture where a GPU runs YOLO11l at 1080p only about five times per second to supply precise position anchors, and a CPU follower interpolates the target's motion between detections with sparse optical flow inside an adaptive region of interest. A training-free recovery module re-acquires the target after occlusion or drift. The reported result is 35.1 FPS system throughput with 0.889 precision on LRDD v2, retaining 97.2% of the frame-by-frame detector's precision. If the offline measurements hold up live, this gives portable counter-UAV systems a way to reconcile high-resolution small-target detection with smooth real-time control.","feed_headline":"Sparse detector with flow achieves 35 FPS tracking at 97% precision","feed_subtitle":"A heavy GPU detector anchors at 1080p every ~150 ms; CPU optical flow fills the gaps for smooth gimbal control.","key_machinery":"The Observer-Follower architecture is the load-bearing design: a GPU-bound detector (YOLO11l) provides sparse, high-accuracy anchors at native resolution, and a CPU-bound follower performs high-frequency state interpolation with ROI-constrained pyramidal Lucas-Kanade optical flow, refined by median-flow outlier rejection, scale-aware search-window and feature-quality tuning, drift-correction template matching, and a linear Kalman filter. A Dual-Space Recovery module fuses Lab and HSV color probability maps with a geometric safety valve (size, position, shape, spectral constraints) to re-capture the target without learned re-identification. The flow follower is what up-samples the ~5 Hz detec","core_discovery":"On an NVIDIA Jetson Orin Nano, the paper shows that feeding YOLO11l native 1920x1080 frames yields strong small-target detection (0.915 precision) but only 6.1 FPS, while a frame-by-frame lightweight detector is fast but inaccurate (0.412 precision). SDG-Track reallocates the work: the same heavy detector runs asynchronously at roughly 5–7 Hz as an observer producing absolute anchors, and a CPU follower runs sparse pyramidal Lucas-Kanade optical flow within a dynamically sized ROI to estimate target state in between detections. The system also includes median-flow outlier rejection, scale-aware parameter tuning, a lazy template update, Kalman smoothing, and a dual-color-space geometric recov","pith_inferences":["The reported Sys FPS is an offline state-update rate on recorded streams; a live closed-loop evaluation that includes camera decode, detector/flow synchronization, and PTZ command latency would be the natural next validation of the real-time claim.","The same sparse-detection-plus-flow pattern may transfer to other small-object tracking domains (maritime, ground vehicles, wildlife) where a strong detector exists but must run at low frequency.","The adaptive ROI and scale-aware thresholds are set by hand; a testable extension would estimate them from target range or apparent size to remove the remaining tuning parameters.","Because recovery relies on color histograms, targets with appearance change or distractors with similar spectral content could break it; a lightweight descriptor could complement the color cue when compute permits."],"forward_implications":["A portable counter-UAV station can operate a high-capacity detector at full resolution without dropping below the 30 Hz update rate that PTZ gimbal control needs.","Sparse detection plus flow interpolation preserves most of the detector's accuracy (97.2% precision) while increasing output rate by about a factor of five.","Training-free re-acquisition removes the need for a learned re-identification model, keeping the system lightweight and easy to deploy.","The observer-follower split could become a general recipe for any edge vision task that needs both small-object accuracy and a high-rate control loop.","The paper's own future direction—learned motion prediction—suggests that the remaining 2.8% precision gap and aggressive-maneuver failures are the next targets."],"fun_headline_variants":["SDG-Track hits 35 FPS on 1080p UAV tracking","Observer-follower framework: 35 FPS, 97% precision on edge","Heavy detector + CPU flow: 35 FPS UAV tracking on Jetson","Sparse detection, dense flow: real-time UAV tracking on edge","1080p UAV tracking at 35 FPS on Jetson Orin Nano"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The real-time claim rests on the offline Sys FPS measurement; if the 35.1 FPS throughput does not transfer to the live closed loop with camera decoding, asynchronous detector-flow synchronization, and PTZ command latency, the 30 Hz control claim is unsupported.","fun_headline_variants_meta":{"raw":{"variants":["SDG-Track hits 35 FPS on 1080p UAV tracking","Observer-follower framework: 35 FPS, 97% precision on edge","Heavy detector + CPU flow: 35 FPS UAV tracking on Jetson","Sparse detection, dense flow: real-time UAV tracking on edge","1080p UAV tracking at 35 FPS on Jetson Orin Nano"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000299,"raw_usage":{"total_tokens":1591,"prompt_tokens":795,"completion_tokens":796,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":539,"completion_tokens_details":{"reasoning_tokens":693}},"tokens_in":539,"tokens_out":796,"duration_ms":6976,"temperature":1.0,"reasoning_tokens":693,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T18:29:30.533811+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A live closed-loop test on the ground-to-air station: track an agile FPV drone and measure the sustained gimbal command rate and end-to-end latency from frame capture to command output; if the live system cannot hold at least 30 Hz updates while keeping precision within about 3% of the offline frame-by-frame detector, the paper's central claim fails.","supporting_citations":[],"review_version":1}