Pith. sign in

REVIEW 4 major objections 5 minor 60 references

SkyEV is a new open-source RGB-event dataset for UAV detection that includes camera ego-motion, tiny targets, and uncompressed frames.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 14:27 UTC pith:TQB5HCGR

load-bearing objection Genuinely useful RGB-event UAV dataset with real hardware sync and ego-motion, but the central artifact is unreleased and sync/label accuracy is unquantified; worth serious referee time if the authors commit to releasing data and adding validation. the 4 major comments →

arxiv 2607.18747 v1 pith:TQB5HCGR submitted 2026-07-21 cs.CV

SkyEV: RGB-Event UAV detection and tracking dataset and baseline

Jakub Mandula , Sebastian Heusinger , Julian Moosmann , Christian Vogt , Michele Magno This is my paper
classification cs.CV
keywords UAV detectionevent-based visionRGB-event fusionmulti-modal datasetobject trackingsmall object detectioncamera ego-motionsensor fusion
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper introduces SkyEV, an open-source dataset of 2.17 hours of synchronized, uncompressed RGB and event-camera recordings of UAVs. The dataset is designed to reflect realistic counter-UAV conditions that existing datasets lack: significant camera ego-motion, a wide range of lens fields of view, and small, fast targets (median bounding box 41 px, velocities up to 14,000 px/s). The authors also present a baseline detector that fuses RGB frames with event histograms and reports that it improves detection over either modality alone, reaching a higher mAP50-95 than the individual YOLOX models. The value, if the claims hold, is a public benchmark that lets the community train and compare multimodal algorithms under the motion and scale conditions that real drone detection systems face.

Core claim

The central deliverable is SkyEV, a dataset recorded with two Prophesee EVK4 event cameras and a global-shutter RGB camera mounted in parallel. RGB frames are stored as 10-bit uncompressed TIFFs, the shutter signal is written into the event stream for sub-frame synchronization, and labels are semi-automatically generated and propagated across views via homographies. As a baseline, the authors fuse a SAST event-patch-adaptive transformer with a YOLOX RGB backbone, fusing features at the FPN stage; on SkyEV this fused model reaches 45.8 mAP50-95, outperforming YOLOX on RGB alone (35.5) and on events alone (37.3).

What carries the argument

The load-bearing mechanisms are (1) hardware time synchronization, in which the RGB camera's shutter signal is recorded as an external event into the event stream, allowing precise temporal alignment; (2) homography-based spatial alignment between the RGB and the two event cameras, with different FOV lenses, so labels from one view can be mapped to the others; and (3) the baseline fusion architecture, in which the SAST transformer selects only active event patches that then fuse (by element-wise summation across pyramid scales) with YOLOX RGB features for detection.

Load-bearing premise

The load-bearing assumption is that the homography-based alignment is accurate for the recorded targets: it is exact for one depth plane, and for near, small targets the parallax misalignment could corrupt the ground-truth labels, with no reported quantitative alignment error to rule this out.

What would settle it

Compute the bounding-box overlap (e.g., IoU) between the RGB and event projections for sequences with close targets; if overlap systematically drops below, say, 0.5 as target distance decreases, the homography assumption is violated and labels for those near targets are noisy.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • SkyEV provides a public benchmark where camera ego-motion produces background events, so detectors must separate target motion from ego-motion clutter.
  • The 10-bit uncompressed RGB frames enable research on compression artifacts and dynamic range without the JPEG degradation present in many prior datasets.
  • The fusion baseline shows a clear detection gain from adding the event stream, supporting the case for multimodal UAV detection.
  • With inference under 6 ms on a Jetson AGX Orin (TensorRT), the approach points toward real-time embedded deployment.
  • The dataset includes eight drone types and both wide- and narrow-angle lenses, allowing evaluation of the FOV-range trade-off.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Because the homography mapping is exact for a single plane distance, close-range targets will carry some parallax misalignment; a natural next step is to publish per-sequence alignment error or re-align close targets with a depth-aware method.
  • The absence of a reported synchronization jitter figure leaves the 'highly synchronized' claim partly unquantified; a simple test is to measure the residual offset between the shutter event timestamps and event stream across sequences.
  • The dataset could be reused for tracking and multi-object tracking benchmarks, not just detection, since labels and MOT format are provided; that would let the community test whether the sensor fusion helps association under ego-motion.
  • The reported APsmall remains below the overall mAP, hinting that very small targets are the hardest case; a follow-up could report performance versus target size to expose where fusion helps most.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper presents SkyEV, a proposed open-source RGB-event dataset for UAV detection and tracking, recorded with two Prophesee EVK4 event cameras and one global-shutter Ximea RGB camera. The dataset contains 398 sequences, about 2.17 hours of footage, 8 drone types, uncompressed 10-bit RAW RGB frames, and event data in native EVT3 format. Labels are produced by manual annotation on one view, propagation with RTS, manual correction, and homography mapping to the other camera views. The authors also provide an experimental baseline fusing a YOLOX RGB stream with a modified SAST event-stream module at the FPN stage, reporting inference latencies on an RTX 4090 and a Jetson AGX Orin. The main claims are that SkyEV fills gaps left by prior RGB-event drone datasets: camera ego-motion, extremely small targets, varied lens configurations, and uncompressed frames.

Significance. If the dataset is released and its label and synchronization quality are quantitatively validated, SkyEV would be a valuable community resource. The combination of raw uncompressed RGB, two event cameras with different fields of view, strong ego-motion, and high-velocity small targets addresses a real gap in current benchmarks, most of which use compressed frames, stationary cameras, or limited lens diversity. The hardware-triggered synchronization design, the use of MOT-format labels, and the inclusion of a unified data loader are useful contributions. The baseline, while not the primary contribution, shows a plausible fusion architecture and reports embedded latency figures that are relevant for real-time counter-UAV systems. The main risks are that the dataset is not currently accessible, the synchronization accuracy is not measured, and the spatial alignment error from the homography mapping is not quantified, all of which directly affect the validity of the benchmark labels.

major comments (4)
  1. [Abstract and Sec. 1] The paper's central artifact is described as an 'open-source dataset' and 'open-source data loader,' but no URL, repository, license, or download procedure is provided anywhere in the manuscript. For a dataset paper, accessibility is load-bearing; without a release mechanism the contribution cannot be evaluated or used. The authors should supply a permanent URL, a license, and a checksum or release version, or explicitly state that the dataset will be made available upon acceptance.
  2. [Sec. 3.2, Eq. (1)] The label-alignment pipeline is not quantitatively validated. Eq. (1) is a plane-induced homography that is exact only for a single distance d, and the paper itself notes this. The dataset deliberately includes close, small, fast targets (median box 41 px, velocities up to 14,000 px/s), for which the infinite-distance assumption will fail with a 5 cm baseline. The authors say manual adjustments and edge-overlap checks are made, but they report no residual alignment error, no fraction of frames or boxes that required manual correction, and no independent verification of labels after homography mapping. Because labels are propagated to both event views from a single annotated view, this is the main quality-control step for the benchmark. The paper should include quantitative alignment statistics, e.g., re-projection error on calibration targets, agreement between independently labeled view
  3. [Sec. 3.1] The 'highly synchronized' claim is not supported by any measured quantity. The paper states that the RGB shutter signal is timestamped as an external event in the EVK4 stream, which is a reasonable hardware mechanism, but no sync jitter, timestamp resolution, or drift figure is reported. For an event-RGB dataset, the temporal correspondence between frame exposure intervals and event timestamps is critical: a 1 ms error can displace a high-velocity target by hundreds of pixels. The authors should report the timestamp resolution of the EVK4, the measured distribution of shutter-signal latency, and how the ±25 ms event-histogram window in Sec. 4 relates to the actual exposure intervals.
  4. [Sec. 5 and Table 3] The baseline results are presented as demonstrating the dataset's efficacy, but they rest on a single 80:20 split, 20 training epochs, and no seed or repeated-run statistics. No optimizer, learning-rate schedule, batch size, or validation-set sequence list is given, so the results in Table 3 are not reproducible and the claimed improvement of SAST+RGB over single-modality models is not statistically supported. For a benchmark paper, the evaluation protocol should include either multiple seeds with mean/std, a held-out split that is fully specified, or a clear statement that the numbers are indicative only and not a competitive comparison. At minimum, the training configuration and split metadata must be released with the dataset.
minor comments (5)
  1. [Sec. 3.4 vs Table 1] The number of frames is inconsistent: Sec. 3.4 says 243,823 frames while Table 1 lists 242,465. The abstract states an average pixel size of 83x49, while Table 1 gives an average scale of 83x49 (63.8); the distinction between 'average' and 'median' (41 px, Fig. 7) should be stated consistently.
  2. [Fig. 7] The x-axis is labeled 'Area (px)' but the caption text discusses bounding-box sizes in terms of sqrt(wh). Please clarify whether the histogram is of area, width×height, or sqrt(area), and ensure the axis label matches the metric reported in Table 1.
  3. [Title] The title contains a spacing issue: 'UA V detection' should be 'UAV detection'.
  4. [Sec. 3.1 and Fig. 3] The lens configurations are said to be 'listed in the table of Fig. 3,' but the figure in the text displays a schematic rather than a readable table. Please move the exact lens/focal-length configuration into a separate table with sequence-level details.
  5. [General] Several references use inconsistent formatting (e.g., missing page ranges, inconsistent DOI/arXiv presentation). The reference list should be normalized to the journal style.

Circularity Check

0 steps flagged

No circularity: SkyEV's dataset statistics and baseline results are empirical measurements, not predictions derived from the paper's own inputs.

full rationale

The paper is a dataset-and-baseline contribution. The dataset statistics (frame counts, box sizes, velocities) are measurements of recorded data; the baseline mAP values in Table 3 are results of training and evaluating published external models (SAST, YOLOX) on the dataset, not quantities derived from the dataset's defining assumptions. The homography in Eq. (1) is used for label transfer between cameras and is explicitly stated to be accurate only for large distances; this is a stated approximation about label quality, not a predicted result that reduces to its own input. The 'highly synchronized' claim rests on a hardware shutter signal mechanism, and any missing jitter quantification is a measurement-quality limitation, not a circular derivation. The self-citations (refs [2], [30], [31]) are used only to motivate a standard multi-camera rig and fusion background; they do not supply a load-bearing theorem or fitted parameter, and the central contribution is externally grounded in the physical recordings and in external model implementations. No fitted quantity is renamed as a prediction, and no uniqueness claim from the authors' prior work is imported to force a choice. Thus there is no significant circularity.

Axiom & Free-Parameter Ledger

2 free parameters · 3 axioms · 0 invented entities

The central claim is an empirical resource, so the load-bearing assumptions are about alignment, synchronization, and label quality. The baseline depends on hand-chosen event-histogram and training parameters. No new physical entities or theoretical constructs are introduced.

free parameters (2)
  • Event histogram representation parameters = ±25 ms window, 10 temporal bins, 2 polarity bins
    Hand-chosen input representation for the baseline (Sec. 4). These parameters affect event features and the reported fusion benefit, but are not fitted to the target result.
  • Training hyperparameters = 20 epochs, 50% horizontal flip, 80% zoom, zoom factors 1.5 and 1.2
    Chosen for the baseline without a reported tuning procedure (Sec. 5). The reported mAP values are conditional on these choices.
axioms (3)
  • domain assumption Homography maps labels across views exactly for targets at infinite distance; parallax is negligible for recorded targets.
    Eq. (1) and Sec. 3.2: the tn^T/d term vanishes only when d approaches infinity. The dataset includes close targets, so manual adjustments are needed; no alignment error is quantified.
  • domain assumption Hardware shutter signal recorded as an external event gives sub-frame synchronization between RGB and event cameras.
    Sec. 3.1 describes the mechanism but reports no measured sync jitter or validation that the sync holds across all sequences.
  • domain assumption Semi-automatic labels (RTS propagation plus manual correction) are accurate enough to train and evaluate detectors.
    Sec. 3.2; no inter-annotator agreement, label quality metric, or bounding-box noise analysis is reported.

pith-pipeline@v1.3.0-alltime-deepseek · 13243 in / 10049 out tokens · 90807 ms · 2026-08-01T14:27:47.646378+00:00 · methodology

0 comments
read the original abstract

Detecting UAVs in air spaces has become increasingly important due to UAVs widespread availability and easy usage. However, due to their small size, they are typically difficult to detect at a sufficient range. For the training of optimized detection algorithms, datasets have been published, covering optical sensing methods ranging from infrared to regular RGB to event-sensor-based. However, these datasets often fail to reflect realistic counter-UAV scenarios, lacking critical factors such as camera ego-motion, extremely small target scales, and diverse lens configurations, and introduce compression artefacts on the frame images. To address this gap, we introduce SkyEV, an open-source dataset featuring highly synchronized uncompressed RGB and event-based data. SkyEV distinguishes itself by capturing complex real-world conditions, including significant camera motion and varied optical setups, which are essential for testing the fundamental trade-off between Field of View and detection range. Furthermore, we provide a unified data loader and establish an experimental baseline using a multi-modal architecture, demonstrating the dataset's efficacy in detecting challenging, small-scale targets.

Figures

Figures reproduced from arXiv: 2607.18747 by Christian Vogt, Jakub Mandula, Julian Moosmann, Michele Magno, Sebastian Heusinger.

Figure 1
Figure 1. Figure 1: Example image from the SkyEV dataset containing event data overlayed on top of an RGB frame with multiple annotated UAV tracks. Ego-motion of the camera causes background clutter also in the event domain. identification and localization, making it a prominent focus for medium-range detection [8, 9, 56]. Despite being a passive and relatively low-cost solution, traditional optical UAV detection suffers from… view at source ↗
Figure 2
Figure 2. Figure 2: A sample of challenging scenarios from the dataset 3.1 Dataset Recording The dataset was recorded using the multi-sensor setup shown in [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Schematic setup of the dataset recording setup, including lenses used on each camera. Global shutter RGB camera in center (Ximea) has its shutter signal hardware￾timestamped using the EVK4 Trigger-in. between 8 mm and 100 mm was used, with the exact configurations listed in the table of [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: LightGlue Feature matching be￾tween RGB images and Synthetic frames from events using E2VID [40] [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Bounding box heatmap for the dataset, split for Wide-angle event camera FOV (a) as well as zoomed Narrow-angle FOV (b). The centered distribution helps by allowing more aggressive dataset augmentations without loss of training labels. igate the scarcity of small-scale bounding box instances to an extent, it often fails to capture the stochastic nature of sensor noise [22]. The distribution of bounding box … view at source ↗
Figure 7
Figure 7. Figure 7: Distribution of bounding sizes in the SkyEV dataset. 0 2000 4000 6000 8000 10000 12000 14000 Speed (px/s) 10 0 10 1 10 2 10 3 10 4 10 5 Frequency [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: The SAST model can selectively mask inactive event patches, thereby improving inference performance. In tasks such as UAV detection, large areas of the scene that lack activity allow the system to benefit greatly from this sparsity. In terms of network architecture, only event histogram image tiles with po￾tentially interesting features are extracted with an SAST [37] module. In our approach, the SAST mode… view at source ↗
Figure 10
Figure 10. Figure 10: Overview of the SAST+RGB architecture used as experimental baseline. (Bounding boxes smaller than 32×32 px). Otherwise, the evaluation of the SAST implementation would filter those boxes. The SAST module is combined with a YOLOX [14] backbone that takes RGB images as input. To extract the same amount of semantic information from RGB data as SAST does from event data, YOLOX is used as the backbone for the … view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

60 extracted references · 18 canonical work pages · 15 internal anchors

  1. [1]

    prophesee.ai/stable/metavision_sdk/index.html

    Metavision SDK — Metavision SDK Docs 4.0.0 documentation,https://docs. prophesee.ai/stable/metavision_sdk/index.html

  2. [2]

    In: 2025 International Joint Conference on Neural Networks (IJCNN)

    Bonazzi, P., Vogt, C., Jost, M., Qin, H., Khacef, L., Paredes-Valles, F., Magno, M.: RGB-Event Fusion with Self-Attention for Collision Prediction. In: 2025 International Joint Conference on Neural Networks (IJCNN). pp. 1–6.https: //doi.org/10.1109/IJCNN64981.2025.11227211,https://ieeexplore.ieee. org/document/11227211/

    arXiv 2025

  3. [3]

    https://doi.org/10.1109/TPAMI.1986.4767851,https://ieeexplore.ieee

    Canny, J.: A Computational Approach to Edge DetectionP AMI-8(6), 679–698. https://doi.org/10.1109/TPAMI.1986.4767851,https://ieeexplore.ieee. org/abstract/document/4767851

    arXiv 1986

  4. [4]

    13627,http://arxiv.org/abs/2408.13627

    Chakravarthi, B., Verma, A.A., Daniilidis, K., Fermuller, C., Yang, Y.: Recent Event Camera Innovations: A Survey.https://doi.org/10.48550/arXiv.2408. 13627,http://arxiv.org/abs/2408.13627

    Pith/arXiv arXiv doi:10.48550/arxiv.2408.13627

  5. [5]

    2506.23575,http://arxiv.org/abs/2506.23575

    Chen, N., Xiao, C., Dai, Y., He, S., Li, M., An, W.: Event-based Tiny Object De- tection: A Benchmark Dataset and Baseline.https://doi.org/10.48550/arXiv. 2506.23575,http://arxiv.org/abs/2506.23575

    Pith/arXiv arXiv doi:10.48550/arxiv.2506.23575

  6. [6]

    Chen, Y., Aggarwal, P., Choi, J., Kuo, C.C.J.: A Deep Learning Approach to Drone Monitoring,http://arxiv.org/abs/1712.00863

    Pith/arXiv arXiv

  7. [7]

    100511,https://linkinghub.elsevier.com/retrieve/pii/S2590005625001389

    DeMacedo,S.O.,Caetano,M.,DaCosta,R.M.:Dronedetectioninairportenviron- ments: A literature review28, 100511.https://doi.org/10.1016/j.array.2025. 100511,https://linkinghub.elsevier.com/retrieve/pii/S2590005625001389

    work page doi:10.1016/j.array.2025 2025

  8. [8]

    Vision-Based Anti Unmanned Aerial Technology: Opportunities and Challenges

    Ding, G., Ren, Y., Liu, Y., Zhao, Q., Li, S.: Vision-Based Anti Unmanned Aerial Technology: Opportunities and Challenges.https://doi.org/10.48550/arXiv. 2507.10006,http://arxiv.org/abs/2507.10006

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2507.10006

  9. [9]

    48550/arXiv.2504.11967,http://arxiv.org/abs/2504.11967

    Dong, Y., Wu, F., Zhang, S., Chen, G., Hu, Y., Yano, M., Sun, J., Huang, S., Liu, F., Dai, Q., Cheng, Z.Q.: Securing the Skies: A Comprehensive Survey on Anti-UAV Methods, Benchmarking, and Future Directions.https://doi.org/10. 48550/arXiv.2504.11967,http://arxiv.org/abs/2504.11967

    doi:10.48550/arxiv.2504.11967

  10. [10]

    https://openreview.net/forum?id=YicbFdNTTy

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., Uszkoreit, J., Houlsby, N.: An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale. https://openreview.net/forum?id=YicbFdNTTy

  11. [11]

    2025.3568333,https://ieeexplore.ieee.org/document/10994378

    Duan, P., Li, B., Yang, Y., Lou, H., Teng, M., Zhou, X., Ma, Y., Shi, B.: EventAid: Benchmarking Event-Aided Image/Video Enhancement Algorithms With Real- Captured Hybrid Dataset47(8), 6959–6973.https://doi.org/10.1109/TPAMI. 2025.3568333,https://ieeexplore.ieee.org/document/10994378

    arXiv 2025

  12. [12]

    In: Gasparovic, SkyEV: RGB-Event Dataset 15 E., Robins, V., Turner, K

    Emerson, T.H., Tymochko, S., Stantchev, G., Edelberg, J.A., Wilson, M., Olson, C.C.: A Topological Approach for Motion Track Discrimination. In: Gasparovic, SkyEV: RGB-Event Dataset 15 E., Robins, V., Turner, K. (eds.) Research in Computational Topology 2, pp. 211–

  13. [13]

    Gallego, G., Delbrück, T., Orchard, G., Bartolozzi, C., Taba, B., Censi, A., Leutenegger, S., Davison, A.J., Conradt, J., Daniilidis, K., Scaramuzza, D.: Event- Based Vision: A Survey44(1), 154–180.https://doi.org/10.1109/TPAMI.2020. 3008413

    doi:10.1109/tpami.2020 2020

  14. [14]

    Ge, Z., Liu, S., Wang, F., Li, Z., Sun, J.: YOLOX: Exceeding YOLO Series in 2021.https://doi.org/10.48550/arXiv.2107.08430,http://arxiv.org/abs/ 2107.08430

    Pith/arXiv arXiv doi:10.48550/arxiv.2107.08430 2021

  15. [15]

    DSEC: A Stereo Event Camera Dataset for Driving Scenarios

    Gehrig, M., Aarents, W., Gehrig, D., Scaramuzza, D.: DSEC: A Stereo Event Camera Dataset for Driving Scenarios.https://doi.org/10.48550/arXiv.2103. 06011,http://arxiv.org/abs/2103.06011

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2103.06011

  16. [16]

    In: 2023 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR)

    Gehrig, M., Scaramuzza, D.: Recurrent vision transformers for object detection with event cameras. In: 2023 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR). pp. 13884–13893 (2023).https://doi.org/10.1109/ CVPR52729.2023.01334

    arXiv 2023

  17. [17]

    In: Emerging Imaging and Sensing Technologies for Secu- rity and Defence III; and Unmanned Sensors, Systems, and Countermeasures

    Hammer, M., Hebel, M., Laurenzis, M., Arens, M.: Lidar-based detection and tracking of small UAVs. In: Emerging Imaging and Sensing Technologies for Secu- rity and Defence III; and Unmanned Sensors, Systems, and Countermeasures. vol. 10799, pp. 177–185. SPIE.https://doi.org/10.1117/12.2325702,https://www. spiedigitallibrary.org/conference- proceedings- of...

    work page doi:10.1117/12.2325702

  18. [18]

    2026.3650835,https://ieeexplore.ieee.org/abstract/document/11328061

    Jekateryńczuk, G., Piotrowski, Z.: The Sound of UAVs: A Review of Algorithms, Applications, and Databases14, 4222–4245.https://doi.org/10.1109/ACCESS. 2026.3650835,https://ieeexplore.ieee.org/abstract/document/11328061

    arXiv 2026

  19. [19]

    Jiang, N., Wang, K., Peng, X., Yu, X., Wang, Q., Xing, J., Li, G., Guo, G., Ye, Q., Jiao, J., Zhao, J., Han, Z.: Anti-UAV: A Large-Scale Benchmark for Vision- Based UAV Tracking25, 486–500.https://doi.org/10.1109/TMM.2021.3128047, https://ieeexplore.ieee.org/document/9615243/

    arXiv 2021

  20. [20]

    Khawaja, W., Ezuma, M., Semkin, V., Erden, F., Ozdemir, O., Guvenc, I.: A Sur- vey on Detection,Classification, andTracking ofAAVs UsingRadar andCommuni- cations Systems28, 3272–3310.https://doi.org/10.1109/COMST.2025.3554613, https://ieeexplore.ieee.org/abstract/document/10938573

    arXiv 2025

  21. [21]

    Leal-Taixé, L., Milan, A., Reid, I., Roth, S., Schindler, K.: MOTChallenge 2015: Towards a Benchmark for Multi-Target Tracking.https://doi.org/10.48550/ arXiv.1504.01942,http://arxiv.org/abs/1504.01942

    Pith/arXiv arXiv doi:10.48550/arxiv.1504.01942 2015

  22. [22]

    1109/JSSC.2007.914337,https://ieeexplore.ieee.org/document/4444573

    Lichtsteiner, P., Posch, C., Delbruck, T.: A 128x128 120 dB 15µsLatency Asyn- chronous Temporal Contrast Vision Sensor43(2), 566–576.https://doi.org/10. 1109/JSSC.2007.914337,https://ieeexplore.ieee.org/document/4444573

    arXiv 2007

  23. [23]

    Lin, T.Y., Maire, M., Belongie, S., Bourdev, L., Girshick, R., Hays, J., Perona, P., Ramanan, D., Zitnick, C.L., Dollár, P.: Microsoft COCO: Common Objects in Context.https://doi.org/10.48550/arXiv.1405.0312,http://arxiv.org/ abs/1405.0312

    Pith/arXiv arXiv doi:10.48550/arxiv.1405.0312

  24. [24]

    org/abs/2306.13643

    Lindenberger, P., Sarlin, P.E., Pollefeys, M.: LightGlue: Local Feature Matching at Light Speed.https://doi.org/10.48550/arXiv.2306.13643,http://arxiv. org/abs/2306.13643

    Pith/arXiv arXiv doi:10.48550/arxiv.2306.13643

  25. [25]

    Mandula et al

    Liu, Z., Lin, Y., Cao, Y., Hu, H., Wei, Y., Zhang, Z., Lin, S., Guo, B.: Swin Transformer: Hierarchical Vision Transformer using Shifted Windows.https:// doi.org/10.48550/arXiv.2103.14030,http://arxiv.org/abs/2103.14030 16 J. Mandula et al

    Pith/arXiv arXiv doi:10.48550/arxiv.2103.14030

  26. [26]

    EV-Flying: an Event-based Dataset for In-The-Wild Recognition of Flying Objects

    Magrini, G., Becattini, F., Colombo, G., Pala, P.: EV-Flying: An Event-based Dataset for In-The-Wild Recognition of Flying Objects.https://doi.org/10. 48550/arXiv.2506.04048,http://arxiv.org/abs/2506.04048

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2506.04048

  27. [27]

    Magrini, G., Becattini, F., Pala, P., Bimbo, A.D., Porta, A.: Neuromorphic Drone Detection: An Event-RGB Multimodal Approach.https://doi.org/10.48550/ arXiv.2409.16099,http://arxiv.org/abs/2409.16099

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2409.16099

  28. [28]

    Magrini, G., Berlincioni, L., Becattini, F., Cultrera, L., Pala, P.: Drone Detection with Event Cameras

  29. [29]

    48550/arXiv.2506.05163,http://arxiv.org/abs/2506.05163

    Magrini, G., Marini, N., Becattini, F., Berlincioni, L., Biondi, N., Pala, P., Bimbo, A.D.: FRED: The Florence RGB-Event Drone Dataset.https://doi.org/10. 48550/arXiv.2506.05163,http://arxiv.org/abs/2506.05163

    Pith/arXiv arXiv doi:10.48550/arxiv.2506.05163

  30. [30]

    org/10.48550/arXiv.2403.11875,http://arxiv.org/abs/2403.11875

    Mandula, J., Kühne, J., Pascarella, L., Magno, M.: Towards Real-Time Fast Un- manned Aerial Vehicle Detection Using Dynamic Vision Sensors.https://doi. org/10.48550/arXiv.2403.11875,http://arxiv.org/abs/2403.11875

    Pith/arXiv arXiv doi:10.48550/arxiv.2403.11875

  31. [31]

    In: 2023 IEEE SENSORS

    Moosmann, J., Mandula, J., Mayer, P., Benini, L., Magno, M.: Quantitative Eval- uation of a Multi-Modal Camera Setup for Fusing Event Data with RGB Images. In: 2023 IEEE SENSORS. pp. 1–4.https://doi.org/10.1109/SENSORS56945. 2023.10325041,https://ieeexplore.ieee.org/abstract/document/10325041

    arXiv 2023

  32. [32]

    In: 2021 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition Workshops (CVPRW)

    Muglikar, M., Gehrig, M., Gehrig, D., Scaramuzza, D.: How to Calibrate Your Event Camera. In: 2021 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition Workshops (CVPRW). pp. 1403–1409. IEEE.https://doi.org/ 10.1109/CVPRW53098.2021.00155,https://ieeexplore.ieee.org/document/ 9523174/

    arXiv 2021

  33. [33]

    Park, J., Hong, M.: Continuous Histogram for Event-based Vision Camera Systems

  34. [34]

    Paul, M., Danelljan, M., Mayer, C., Gool, L.V.: Robust Visual Tracking by Seg- mentation.https://doi.org/10.48550/arXiv.2203.11191,http://arxiv.org/ abs/2203.11191

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2203.11191

  35. [35]

    Peng, S., Zhou, H., Dong, H., Shi, Z., Liu, H., Duan, Y., Chang, Y., Yan, L.: CoSEC: A Coaxial Stereo Event Camera Dataset for Autonomous Driving.https: //doi.org/10.48550/arXiv.2408.08500,http://arxiv.org/abs/2408.08500

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2408.08500

  36. [36]

    Specific implementation: utils/evaluation/prophesee/evaluation.py

    Peng, Y.: Sast: Scene adaptive sparse transformer for event-based object detec- tion.https://github.com/Peterande/SAST(2024), accessed: 2026-03-04. Specific implementation: utils/evaluation/prophesee/evaluation.py

    2024

  37. [38]

    Learning to Detect Objects with a 1 Megapixel Event Camera

    Perot, E., de Tournemire, P., Nitti, D., Masci, J., Sironi, A.: Learning to Detect Objects with a 1 Megapixel Event Camera.https://doi.org/10.48550/arXiv. 2009.13436,http://arxiv.org/abs/2009.13436

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2009.13436 2009

  38. [39]

    In: Proceedings of the 38th International Conference on Machine Learning

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., Krueger, G., Sutskever, I.: Learning Transferable Visual Models From Natural Language Supervision. In: Proceedings of the 38th International Conference on Machine Learning. pp. 8748–8763. PMLR,https: //proceedings.mlr.press/v139/radford21a.html

  39. [40]

    1109/TPAMI.2019.2963386,https://ieeexplore.ieee.org/document/8946715/

    Rebecq, H., Ranftl, R., Koltun, V., Scaramuzza, D.: High Speed and High Dynamic Range Video with an Event Camera43(6), 1964–1980.https://doi.org/10. 1109/TPAMI.2019.2963386,https://ieeexplore.ieee.org/document/8946715/

    arXiv 1964

  40. [41]

    Rouhi, A., Patel, S., McCarthy, N., Khan, S., Khorsand, H., Lefkowitz, K., Han, D.K.: LRDDv2: Enhanced Long-Range Drone Detection Dataset with Range Infor- SkyEV: RGB-Event Dataset 17 mation and Comprehensive Real-World Challenges.https://doi.org/10.48550/ arXiv.2508.03331,http://arxiv.org/abs/2508.03331

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2508.03331

  41. [42]

    https://doi.org/10.48550/arXiv.2509.25164,http://arxiv.org/abs/2509

    Sapkota, R., Cheppally, R.H., Sharda, A., Karkee, M.: YOLO26: Key Architectural Enhancements and Performance Benchmarking for Real-Time Object Detection. https://doi.org/10.48550/arXiv.2509.25164,http://arxiv.org/abs/2509. 25164

    doi:10.48550/arxiv.2509.25164

  42. [43]

    Steinhauser, C., Reis, P., Padusinski, H., Langner, J., Sax, E.: Data Quality Mat- ters: Quantifying Image Quality Impact on Machine Learning Performance.https: //doi.org/10.48550/arXiv.2503.22375,http://arxiv.org/abs/2503.22375

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2503.22375

  43. [44]

    2760/02357,https://publications.jrc.ec.europa.eu/repository/handle/ JRC137334

    Stepniak, M., Cheimariotis, I., Lodi, C., Rataj, M., Zawieska, J., Grosso, M., Marotta, A.: Research and Innovation on Drones in Europe.https://doi.org/10. 2760/02357,https://publications.jrc.ec.europa.eu/repository/handle/ JRC137334

  44. [45]

    Tkachenko, M., Malyuk, M., Holmanyuk, A., Liubimov, N.: Label Studio: Data labeling software (2020-2025),https://github.com/HumanSignal/label-studio, opensourcesoftwareavailablefromhttps://github.com/HumanSignal/label-studio

    2020

  45. [46]

    Wang, S., Zhao, M., Dou, R., Yu, S., Liu, L., Wu, N.: A Compact High- Quality Image Demosaicking Neural Network for Edge-Computing Devices21(9), 3265.https://doi.org/10.3390/s21093265,https://pmc.ncbi.nlm.nih.gov/ articles/PMC8125912/

    work page doi:10.3390/s21093265

  46. [47]

    Wang, X., Jin, Y., Wu, W., Zhang, W., Zhu, L., Jiang, B., Tian, Y.: Object Detection using Event Camera: A MoE Heat Conduction based Detector and A New Benchmark Dataset.https://doi.org/10.48550/arXiv.2412.06647, http://arxiv.org/abs/2412.06647

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2412.06647

  47. [48]

    VisEvent: Reliable Object Tracking via Collaboration of Frame and Event Flows

    Wang, X., Li, J., Zhu, L., Zhang, Z., Chen, Z., Li, X., Wang, Y., Tian, Y., Wu, F.: VisEvent: Reliable Object Tracking via Collaboration of Frame and Event Flows. https://doi.org/10.48550/arXiv.2108.05015,http://arxiv.org/abs/2108. 05015

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2108.05015

  48. [49]

    Wang, X., Lou, X., Wang, S., Huang, J., Chen, L., Jiang, B.: Long-Term Visual Object Tracking with Event Cameras: An Associative Memory Augmented Tracker and A Benchmark Dataset.https://doi.org/10.48550/arXiv.2403.05839, http://arxiv.org/abs/2403.05839

    Pith/arXiv arXiv doi:10.48550/arxiv.2403.05839

  49. [50]

    Wang, X., Wang, S., Tang, C., Zhu, L., Jiang, B., Tian, Y., Tang, J.: Event Stream- based Visual Object Tracking: A High-Resolution Benchmark Dataset and A Novel Baseline.https://doi.org/10.48550/arXiv.2309.14611,http://arxiv.org/ abs/2309.14611

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2309.14611

  50. [51]

    In: 2025 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW)

    Xie, B., Zhang, C., Wang, F., Liu, P., Lu, F., Chen, Z., Hu, W.: CST Anti-UAV: A Thermal Infrared Benchmark for Tiny UAV Tracking in Complex Scenes. In: 2025 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW). pp. 6216–6225.https://doi.org/10.1109/ICCVW69036.2025.00647,https:// ieeexplore.ieee.org/document/11375612

    arXiv 2025

  51. [52]

    Yang, N., Wang, Y., Liu, Z., Dai, Y., Liu, Y., Zhao, X.: Focus Through Mo- tion: RGB-Event Collaborative Token Sparsification for Efficient Object Detec- tion.https://doi.org/10.48550/arXiv.2509.03872,http://arxiv.org/abs/ 2509.03872

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2509.03872

  52. [53]

    Visible-Thermal Tiny Object Detection: A Benchmark Dataset and Baselines

    Ying, X., Xiao, C., Li, R., He, X., Li, B., Cao, X., Li, Z., Wang, Y., Hu, M., Xu, Q., Lin, Z., Li, M., Zhou, S., An, W., Sheng, W., Liu, L.: Visible-Thermal Tiny Object Detection: A Benchmark Dataset and Baselines.https://doi.org/10. 48550/arXiv.2406.14482,http://arxiv.org/abs/2406.14482 18 J. Mandula et al

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2406.14482

  53. [54]

    In: 2020 39th Chinese Control Conference (CCC)

    Yuan, X., Xia, J., Wu, J., Shi, J., Deng, L.: Low Altitude Small UAV Detec- tion Based on YOLO model. In: 2020 39th Chinese Control Conference (CCC). pp. 7362–7366.https://doi.org/10.23919/CCC50068.2020.9188588,https: //ieeexplore.ieee.org/abstract/document/9188588

    arXiv 2020

  54. [55]

    https://doi.org/10.48550/arXiv.2412.04149,http://arxiv.org/abs/2412

    Zhang, H., Wang, X., Xu, C., Wang, X., Xu, F., Yu, H., Yu, L., Yang, W.: Frequency-Adaptive Low-Latency Object Detection Using Events and Frames. https://doi.org/10.48550/arXiv.2412.04149,http://arxiv.org/abs/2412. 04149

    Pith/arXiv arXiv doi:10.48550/arxiv.2412.04149

  55. [56]

    Zhao, J., Zhang, J., Li, D., Wang, D.: Vision-Based Anti-UAV Detection and Tracking23(12), 25323–25334.https://doi.org/10.1109/TITS.2022.3177627, https://ieeexplore.ieee.org/abstract/document/9785379

    arXiv 2022

  56. [57]

    In: 2024 IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR)

    Zhao, Y., Lv, W., Xu, S., Wei, J., Wang, G., Dang, Q., Liu, Y., Chen, J.: DE- TRs Beat YOLOs on Real-time Object Detection. In: 2024 IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR). pp. 16965–16974. https://doi.org/10.1109/CVPR52733.2024.01605,https://ieeexplore.ieee. org/document/10657220

    arXiv 2024

  57. [58]

    In: 2023 IEEE International Conference on Robotics and Automation (ICRA)

    Zhou, Z., Wu, Z., Boutteau, R., Yang, F., Demonceaux, C., Ginhac, D.: RGB- Event Fusion for Moving Object Detection in Autonomous Driving. In: 2023 IEEE International Conference on Robotics and Automation (ICRA). pp. 7808–7815. https://doi.org/10.1109/ICRA48891.2023.10161563,https://ieeexplore. ieee.org/abstract/document/10161563

    arXiv 2023

  58. [59]

    Zhu, X.F., Xu, T., Zhao, J., Liu, J.W., Wang, K., Wang, G., Li, J., Wang, Q., Jin, L., Zhu, Z., Xing, J., Wu, X.J.: Evidential Detection and Tracking Collaboration: New Problem, Benchmark and Algorithm for Robust Anti-UAV System.https: //doi.org/10.48550/arXiv.2306.15767,http://arxiv.org/abs/2306.15767

    Pith/arXiv arXiv doi:10.48550/arxiv.2306.15767

  59. [60]

    CRSOT: Cross-Resolution Object Tracking using Unaligned Frame and Event Cameras

    Zhu,Y.,Wang,X.,Li,C.,Jiang,B.,Zhu,L.,Huang,Z.,Tian,Y.,Tang,J.:CRSOT: Cross-Resolution Object Tracking using Unaligned Frame and Event Cameras. https://doi.org/10.48550/arXiv.2401.02826,http://arxiv.org/abs/2401. 02826

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2401.02826

  60. [222]

    Springer International Publishing.https://doi.org/10.1007/978-3-030- 95519-9_9,https://doi.org/10.1007/978-3-030-95519-9_9

    doi:10.1007/978-3-030-