Pith. sign in

REVIEW 3 major objections 6 minor 3 cited by

Learning to Detect Objects with a 1 Megapixel Event Camera

T0 review · 3 major / 6 minor · reviewed 2026-08-27 · deepseek-v4-flash

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

Pith's one-line read An event-only detector matches frame cameras at detecting cars.

desk verdict A valuable 1MP event-camera detection dataset and a sound recurrent architecture; the headline parity with frame-based detection is real on this benchmark but weaker than the abstract claims because the ground truth itself comes from a commercial frame detector. read the letter →

arxiv 2009.13436 v2 pith:FR4CTUDR submitted 2020-09-28 cs.CV cs.LG

Etienne Perot , Pierre de Tournemire , Davide Nitti , Jonathan Masci , Amos Sironi This is my paper · ORCID
classification cs.CVcs.LG
keywords eventcameraobjectdetectionrecurrentneuralnetworkConvLSTMautomotivedatasettemporalconsistencylossevent-basedvisionsingle-shotdetector
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that an object detector fed only raw events from a 1-megapixel event camera can match the accuracy of a conventional grayscale frame-camera detector on automotive scenes. To make that comparison possible, the authors release the first large high-resolution event-camera detection dataset: 14.65 hours of driving, 25 million bounding boxes for cars, pedestrians, and two-wheelers, labeled at 60 Hz through an automated transfer from a co-mounted RGB camera. They introduce RED, a recurrent detector whose ConvLSTM memory lets objects keep being detected after they stop generating events. On their dataset, RED reaches COCO mAP 0.43, identical to Gray-RetinaNet, and runs in about 39 ms per step, without reconstructing intensity images. The sympathetic reading is that event-based vision has crossed the threshold where it can replace frame-based perception in a large-scale real-time task.

What carries the argument

The load-bearing mechanism is a recurrent memory inside a single-shot detector: five ConvLSTM layers that accumulate and preserve features across time, fed by Event Volume tensors (10 channels, 50 ms bins) built from the raw event stream. Because an object that stops moving stops emitting events, a feed-forward detector forgets it; the ConvLSTM state keeps it localized. A dual regression head predicts boxes for the next time step, and a temporal consistency loss ties future predictions to current ones, improving box stability. The automated labeling protocol, which transfers detections from a commercial RGB detector into event-camera coordinates via time synchronization and a homography, is the machinery that makes a 25-million-box dataset affordable.

What would settle it

Take a few thousand frames from the released test set, have humans label the event-camera images directly, and recompute RED's mAP against that manual ground truth. If the transferred labels contain systematic misalignment or semantic errors, the parity with Gray-RetinaNet will shift. A second check: record a scene where an object stops and stays static for several seconds; if RED's detection disappears after the memory decays, the claimed memory benefit is bounded by a short time horizon.

Watch

Extended reading notes

Core claim

The central discovery is that a recurrent network trained directly on event volumes reaches the same mAP (0.43) as a grayscale RetinaNet on the same automotive scenes, the first time an event-camera detector matches a frame-camera detector at scale. The memory carried by five ConvLSTM layers is load-bearing: zeroing the internal state drops mAP from 0.41 to 0.29, about 12 points. A temporal consistency loss, implemented as a dual regression head that predicts boxes one step ahead and penalizes inconsistency, adds another 2 mAP points and 4 mAP75 points. On the low-resolution Gen1 dataset, RED scores 0.40 mAP versus 0.44 for Gray-RetinaNet. The authors conclude that reconstructing gray-level images from events before detection, as in E2Vid-RetinaNet, is both slower (21x) and less accurate than direct event-based detection.

Load-bearing premise

The entire evaluation rests on the automated labeling protocol: ground-truth boxes are produced by a commercial frame-based detector on RGB video, then transferred to event-camera coordinates via time synchronization and a homography that assumes distant objects. If the RGB detector misses classes, the homography misaligns, or synchronization drifts, the mAP numbers measure the event detector against a biased ground truth rather than against true object locations.

Editorial extensions

If this is right

  • An automotive event-based detector can be used without any intensity-image reconstruction step, saving computation and latency while matching frame-based accuracy.
  • Recurrent memory is required for event-based detection; feed-forward event detectors lose about 12 mAP points on this task.
  • The temporal consistency loss improves localization, raising mAP75 by 4 points, so similar auxiliary future-prediction losses may help other recurrent vision tasks.
  • The released 1-megapixel dataset, with 25M boxes at 60 Hz, gives the field a benchmark on which event-camera detectors can be compared fairly.
  • Because event data is invariant to absolute illumination, a detector trained on daylight transfers to night driving qualitatively, a property frame detectors lack.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the automated labels are noisy in ways that favor neither method, the mAP parity suggests the residual gap between event and frame perception is now about sensor limitations (resolution, color, noise) rather than algorithmic immaturity; adding color to the frame detector raises its mAP to 0.56, so a color event sensor could plausibly close or reverse that gap.
  • The same automated transfer protocol could be applied to other tasks such as face detection or pose estimation, letting researchers build large event datasets at a fraction of manual labeling cost.
  • A quantitative night evaluation with human labels would test the claimed illumination invariance more rigorously than the qualitative figures; if confirmed, it is a practical reason to prefer event cameras in low-light automotive systems.
  • Because the ground truth itself comes from a frame-based detector, any class the RGB detector systematically misses is invisible to both training and evaluation; future datasets should include a manual audit subset.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Request a human review

A listed scientist reviews the paper for a fee and the review publishes here regardless of verdict. See the reviewers or get listed.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. This paper introduces a 1-megapixel event-camera dataset for automotive object detection, with 14.65 hours of recordings and 25M bounding boxes generated by an automated protocol that transfers detections from a commercial RGB-frame detector into event-camera coordinates via time synchronization and a homography. The paper also proposes RED, a recurrent ConvLSTM object detector with a temporal-consistency loss, and reports that on the new dataset RED reaches a COCO mAP of 0.43, matching Gray-RetinaNet, while outperforming feed-forward event-based baselines and an event-to-video reconstruction plus RetinaNet pipeline. The paper includes ablations isolating the contributions of the event-volume input representation, recurrent memory, and consistency loss, and shows qualitative generalization to night recordings and to a different event sensor.

Significance. If the parity result holds up to independent scrutiny, the paper is a landmark contribution for event-based vision: it demonstrates for the first time that an event-based detector can match a grayscale frame-based detector on a large-scale automotive detection task, and the released dataset will be a valuable community resource. The ablation study is cleanly designed and convincingly shows that recurrent memory contributes 12 mAP points and the temporal-consistency loss contributes 2 points (Sec. 5.1, Table 1). The comparison against E2Vid-RetinaNet is useful and shows that training directly from events is both faster and more accurate than passing through an intermediate intensity image. The evaluation code is made public, which supports reproducibility. The main caveats are that the parity claim rests entirely on automated, unquantified label generation and on a single run with no error bars, which tempers the strength of the claims.

major comments (3)
  1. [Sec. 4, Sec. 5.2, Table 2] The headline parity between RED and Gray-RetinaNet (both 0.43 mAP on the 1Mpx Dataset) is measured exclusively against the ground truth produced by the automated labeling protocol of Sec. 4. That protocol transfers boxes from an unnamed commercial RGB detector to event coordinates via temporal synchronization and a single homography, and Sec. 5.3 explicitly acknowledges geometric errors (misalignment) and semantic errors (label swaps, erroneous boxes) in this ground truth, but provides no quantification of their frequency or magnitude. Consequently the equal mAP shows only that RED can reproduce the commercial detector's output from events on this benchmark; it does not establish that RED has comparable precision to frame-based detectors on independently established ground truth, as claimed in Sec. 5.2. The authors should either quantify the label noise (e.g., with a human-annotated subset or cross-detector agreement statistics) and show that both detectors are affected symmetrically, or explicitly restate the claim as being benchmark-relative.
  2. [Sec. 5.2, Table 2] All mAP numbers in Table 2 appear to come from a single training run, selected by validation performance, with no error bars or multiple-seed statistics. When the central claim is an equality (0.43 = 0.43), run-to-run variance could easily be as large as the difference. The authors should report the mean and standard deviation over at least several seeds for RED and Gray-RetinaNet on the 1Mpx dataset, or otherwise provide evidence that the parity is not a product of noise.
  3. [Sec. 5.2, paragraph 4] The claim of parity with 'commonly used frame-camera detectors' is conditioned on grayscale input throughout, and the paper itself reports that RetinaNet on color reaches 0.56 mAP, which is 13 points above RED. The paragraph in Sec. 5.2 acknowledges this, but the abstract and conclusion phrase the result as general parity. The authors should qualify the claim in the abstract and conclusion to state explicitly that parity is with a grayscale frame-based detector on the released benchmark, since the color result indicates that commonly used frame-camera detectors have access to information that event cameras currently lack.
minor comments (6)
  1. [Sec. 5.1, Table 1] The table would be clearer if the left and right halves had explicit mAP and mAP75 column headers; as printed, the reader must infer that the two numbers in each cell are mAP and mAP75.
  2. [Sec. 4, first paragraph] The sentence 'the homography assumption is good enough for our case, since objects encountered in automotive scenarios are relatively far compared to the cameras baseline' is a qualitative assertion; a quantitative analysis of residual parallax error as a function of depth would strengthen the protocol description.
  3. [Sec. 5.2, Table 2] The runtime 'realtime' claim for RED (39.33 ms on 1Mpx) is stated without an explicit frame-rate context; since the input event volume is 50 ms, the paper should state that this corresponds to about 25 Hz.
  4. [Abstract] The phrase '25M bounding boxes ... labeled at high frequency' could be misread as human-annotated; the abstract should note that labels are generated by an automated protocol using a commercial frame-based detector.
  5. [Sec. 5.4] The night and cross-camera generalization experiments are qualitative only; the paper does state that no quantitative evaluation is possible due to lack of night ground truth, but a small human-evaluated quantitative study or a larger set of qualitative examples would strengthen the claims.
  6. [Fig. 3(b)] The IoU-vs-track-duration plot would benefit from error bands or an indication of the number of tracks used; as drawn, the magnitude of the improvement with the consistency loss is hard to assess.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the mAP parity is an empirical benchmark result; the automated labeling protocol is a data-quality concern, not a circular derivation.

full rationale

The paper's derivation chain is empirical rather than analytic: RED is trained with a supervised loss (Eq. 3 and Eq. 4) on labels produced by the Sec. 4 protocol, and its mAP is measured on held-out test sequences. Nothing in the loss or architecture definitionally forces RED's mAP to equal Gray-RetinaNet's 0.43; that equality is a measured outcome. The automated labeling protocol (commercial RGB detector plus homography) does mean both RED and Gray-RetinaNet are trained and evaluated against the same machine-generated ground truth, and Sec. 5.3 explicitly concedes geometric and semantic label errors. That is a legitimate benchmark-validity concern about whether the labels constitute independent ground truth, but it is not circular: the paper does not define the target quantity in terms of its own outputs, fit a parameter and then rename it a prediction, or import a uniqueness theorem from the authors' prior work. The self-citation to [5] (Gen1 dataset) provides a public benchmark, not a load-bearing argument. No equation reduces to its own input, so no significant circularity is present.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a new dataset whose ground truth is generated by an automated frame-based pipeline, plus standard deep learning components with hand-set hyperparameters. No new physical entities are introduced.

free parameters (5)
  • focal loss gamma = 2
    Set to 2 following RetinaNet practice; chosen for class imbalance, no tuning shown.
  • smooth L1 beta = 0.11
    Used in regression losses Lr and Lt; value from SSD-style default.
  • event volume bins B = 5
    Number of temporal bins in input representation, chosen in ablation.
  • time interval delta t = 50 ms
    Detection period and event volume window; set for real-time operation, affects all results.
  • learning rate and decay = 0.0002 with 0.98 decay per epoch
    ADAM optimizer hyperparameters selected via validation.
assumptions (4)
  • domain assumption Homography mapping between RGB and event cameras is accurate for automotive scenes.
    Section 4 assumes objects are far compared to the camera baseline; misalignments are acknowledged as label noise.
  • domain assumption The commercial automotive detector's bounding boxes on RGB frames are reliable enough to serve as ground truth.
    The entire dataset labels come from this detector; errors like label swaps are discussed in Section 5.3.
  • domain assumption Event data alone contains sufficient information for object detection.
    The paper assumes no gray-level reconstruction is needed; supported by experiments but not by a theoretical bound.
  • standard math Standard deep learning components (ConvLSTM, SSD, focal loss) train and generalize as expected with backpropagation through time.
    Treated as established background.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning to Detect Objects with a 1 Megapixel Event Camera." pith.science (2026). https://pith.science/paper/FR4CTUDR

@misc{pith2026200913436,
  author       = {Pith},
  title        = {Pith review of: Learning to Detect Objects with a 1 Megapixel Event Camera},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FR4CTUDR}},
  note         = {Machine review of arXiv:2009.13436}
}
read the original abstract

Event cameras encode visual information with high temporal precision, low data-rate, and high-dynamic range. Thanks to these characteristics, event cameras are particularly suited for scenarios with high motion, challenging lighting conditions and requiring low latency. However, due to the novelty of the field, the performance of event-based systems on many vision tasks is still lower compared to conventional frame-based solutions. The main reasons for this performance gap are: the lower spatial resolution of event sensors, compared to frame cameras; the lack of large-scale training datasets; the absence of well established deep learning architectures for event-based processing. In this paper, we address all these problems in the context of an event-based object detection task. First, we publicly release the first high-resolution large-scale dataset for object detection. The dataset contains more than 14 hours recordings of a 1 megapixel event camera, in automotive scenarios, together with 25M bounding boxes of cars, pedestrians, and two-wheelers, labeled at high frequency. Second, we introduce a novel recurrent architecture for event-based detection and a temporal consistency loss for better-behaved training. The ability to compactly represent the sequence of events into the internal memory of the model is essential to achieve high accuracy. Our model outperforms by a large margin feed-forward event-based architectures. Moreover, our method does not require any reconstruction of intensity images from events, showing that training directly from raw events is possible, more efficient, and more accurate than passing through an intermediate intensity image. Experiments on the dataset introduced in this work, for which events and gray level images are available, show performance on par with that of highly tuned and studied frame-based detectors.

Figures

Figures reproduced from arXiv: 2009.13436 by the authors.

Figure 1
Figure 1. Results of our event-camera detector on examples of the released 1Mpx Automotive [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of the proposed architecture. Input events are used to build a tensor map [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. (a) Detail of the box regression heads. In order to regularize temporally our network, we [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Detections on a 1 Mpx Dataset sequence. From top to bottom: [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Labeling and detector failure cases. White boxes correspond to labels, colored boxes to [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Top: Gray-Retinanet applied to night recordings of a HDR automotive camera. Bottom: Our detector RED applied to recordings of a 1 Mpx event camera of the same scene. The detectors were trained on day light data. Gray-Retinanet does not generalize well on night images. …
Figure 7
Figure 7. Figure 7: RED detector trained on ATIS data and applied to DAVIS sequences. Even if our model was trained on a different camera, it generalizes to other sensors, points of view and light conditions. 5.4 Generalization to Night Recordings and Other Event Cameras We now study the …
Figure 8
Figure 8. Figure 8: Left: RGB frames used for labelling, with overlaid bounding boxes returned by the automotive labeling software. Right: sample snapshots from the 1Mpx Detection Dataset, with transferred bounding boxes. The bounding boxes are transferred from the RGB camera using the au…
Figure 9
Figure 9. Figure 9: Visualization of a timesurface with a constant decay of 100 ms. Large values are represented [PITH_FULL_IMAGE:figures/full_fig_p014_9.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. OpenAlex reports about 142 citations worldwide. Full citation record

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

    cs.CV 2026-07 conditional novelty 6.0 of 10

    The paper introduces SkyEV, a 2.17-hour RGB-event drone detection dataset with ego-motion and varied optics, plus a SAST+YOLOX fusion baseline.

  2. WD-DETR: Wavelet Denoising-Enhanced Real-Time Object Detection Transformer for Robot Perception with Event Cameras

    cs.RO 2025-06 conditional novelty 6.0 of 10

    WD-DETR combines a time-decay event representation, wavelet-based denoising in the backbone, and a transformer head to reach new state-of-the-art mAP on DSEC, Gen1, and 1Mpx event camera detection benchmarks.

  3. LiteEvent-AE: Lightweight Autoencoder for Event-Based Vision on Low-Latency Energy-Constrained Edge Devices

    cs.CV 2026-08 conditional novelty 4.0 of 10

    A lightweight event-frame autoencoder classifier consumes 16.19 J per 100 inferences on a Raspberry Pi 4B, about 726 times less energy than a YOLOv9 baseline, with a 4 to 6 percent accuracy gap.

Reference graph

Works this paper leans on

75 extracted references · 62 canonical work pages · cited by 3 Pith papers

  1. [1]

    A 128×128 120 dB 15µs latency asynchronous temporal contrast vision sensor.IEEE J

    P. Lichtsteiner, C. Posch, and T. Delbruck. A 128 × 128 120 db 15 µs latency asynchronous temporal contrast vision sensor. IEEE Journal of Solid-State Circuits, 43(2):566–576, Feb 2008. ISSN 1558-173X. doi: 10.1109/JSSC.2007.914337

    work page arXiv 2008

  2. [2]

    Retinomor- phic event-based vision sensors: bioinspired cameras with spiking output

    Christoph Posch, Teresa Serrano-Gotarredona, Bernabe Linares-Barranco, and Tobi Delbruck. Retinomor- phic event-based vision sensors: bioinspired cameras with spiking output. Proceedings of the IEEE, 102 (10):1470–1484, 2014

    work page 2014

  3. [3]

    4.1 a 640× 480 dynamic vision sensor with a 9µm pixel and 300meps address-event representation

    Bongki Son, Yunjae Suh, Sungho Kim, Heejae Jung, Jun-Seok Kim, Changwoo Shin, Keunju Park, Kyoobin Lee, Jinman Park, Jooyeon Woo, et al. 4.1 a 640× 480 dynamic vision sensor with a 9µm pixel and 300meps address-event representation. In 2017 IEEE International Solid-State Circuits Conference (ISSCC), pages 66–67. IEEE, 2017

    work page 2017

  4. [4]

    Thomas Finateu, Atsumi Niwa, Daniel Matolin, Koya Tsuchimoto, Andrea Mascheroni, Etienne Reynaud, Pooria Mostafalu, Frederick Brady, Ludovic Chotard, Florian LeGoff, et al. 5.10 a 1280 × 720 back- illuminated stacked temporal contrast event-based vision sensor with 4.86µm pixels, 1.066 geps readout, programmable event-rate controller and compressive data-...

    work page 2020

  5. [5]

    A large scale event-based detection dataset for automotive

    Pierre de Tournemire, Davide Nitti, Etienne Perot, Davide Migliore, and Amos Sironi. A large scale event-based detection dataset for automotive. arXiv, pages arXiv–2001, 2020

    work page 2001

  6. [6]

    Event-based Vision: A Survey

    Guillermo Gallego, Tobi Delbruck, Garrick Orchard, Chiara Bartolozzi, Brian Taba, Andrea Censi, Stefan Leutenegger, Andrew Davison, Joerg Conradt, Kostas Daniilidis, et al. Event-based vision: A survey. arXiv preprint arXiv:1904.08405, 2019

    work page Pith review arXiv 1904

  7. [7]

    Simultaneous mosaicing and tracking with an event camera

    Hanme Kim, Ankur Handa, Ryad Benosman, Sio-Hoi Ieng, and Andrew J Davison. Simultaneous mosaicing and tracking with an event camera. J. Solid State Circ, 43:566–576, 2008

    work page 2008

  8. [8]

    Simultaneous optical flow and intensity estimation from an event camera

    Patrick Bardow, Andrew J Davison, and Stefan Leutenegger. Simultaneous optical flow and intensity estimation from an event camera. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 884–892, 2016

    work page 2016

Show all 75 references
  1. [9]

    Real-time intensity-image reconstruction for event cameras using manifold regularisation

    Gottfried Munda, Christian Reinbacher, and Thomas Pock. Real-time intensity-image reconstruction for event cameras using manifold regularisation. International Journal of Computer Vision , 126(12): 1381–1393, 2018

    2018

  2. [10]

    High speed and high dynamic range video with an event camera

    Henri Rebecq, René Ranftl, Vladlen Koltun, and Davide Scaramuzza. High speed and high dynamic range video with an event camera. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019

    2019

  3. [11]

    Space-time event clouds for gesture recognition: From rgb cameras to event cameras

    Qinyi Wang, Yexin Zhang, Junsong Yuan, and Yilong Lu. Space-time event clouds for gesture recognition: From rgb cameras to event cameras. In 2019 IEEE Winter Conference on Applications of Computer Vision (WACV), pages 1826–1835. IEEE, 2019

    2019

  4. [12]

    Eventnet: Asynchronous recursive event processing

    Yusuke Sekikawa, Kosuke Hara, and Hideo Saito. Eventnet: Asynchronous recursive event processing. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 3887–3896, 2019

    2019

  5. [13]

    Event-based asyn- chronous sparse convolutional networks

    Nico Messikommer, Daniel Gehrig, Antonio Loquercio, and Davide Scaramuzza. Event-based asyn- chronous sparse convolutional networks. In European Conference on Computer Vision. (ECCV), 2020

    2020

  6. [14]

    Dynamic evolving spiking neural networks for on-line spatio-and spectro-temporal pattern recognition

    Nikola Kasabov, Kshitij Dhoble, Nuttapod Nuntalid, and Giacomo Indiveri. Dynamic evolving spiking neural networks for on-line spatio-and spectro-temporal pattern recognition. Neural Networks, 41:188–201, 2013

    2013

  7. [15]

    Training deep spiking neural networks using backpropagation

    Jun Haeng Lee, Tobi Delbruck, and Michael Pfeiffer. Training deep spiking neural networks using backpropagation. Frontiers in neuroscience, 10:508, 2016

    2016

  8. [16]

    Slayer: Spike layer error reassignment in time

    Sumit Bam Shrestha and Garrick Orchard. Slayer: Spike layer error reassignment in time. In Advances in Neural Information Processing Systems, pages 1412–1421, 2018

    2018

  9. [17]

    Deep learning in spiking neural networks

    Amirhossein Tavanaei, Masoud Ghodrati, Saeed Reza Kheradpisheh, Timothee Masquelier, and Anthony Maida. Deep learning in spiking neural networks. Neural Networks, 111:47–63, 2019

    2019

  10. [18]

    Asynchronous convolutional networks for object detection in neuromorphic cameras

    Marco Cannici, Marco Ciccone, Andrea Romanoni, and Matteo Matteucci. Asynchronous convolutional networks for object detection in neuromorphic cameras. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, pages 0–0, 2019. 17

    2019

  11. [19]

    Hats: Histograms of averaged time surfaces for robust event-based object classification

    Amos Sironi, Manuele Brambilla, Nicolas Bourdis, Xavier Lagorce, and Ryad Benosman. Hats: Histograms of averaged time surfaces for robust event-based object classification. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 1731–1740, 2018

    2018

  12. [20]

    Phased lstm: Accelerating recurrent network training for long or event-based sequences

    Daniel Neil, Michael Pfeiffer, and Shih-Chii Liu. Phased lstm: Accelerating recurrent network training for long or event-based sequences. In Advances in neural information processing systems, pages 3882–3890, 2016

    2016

  13. [21]

    Ev-segnet: semantic segmentation for event-based cameras

    Inigo Alonso and Ana C Murillo. Ev-segnet: semantic segmentation for event-based cameras. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, pages 0–0, 2019

    2019

  14. [22]

    Event- based vision meets deep learning on steering prediction for self-driving cars

    Ana I Maqueda, Antonio Loquercio, Guillermo Gallego, Narciso García, and Davide Scaramuzza. Event- based vision meets deep learning on steering prediction for self-driving cars. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 5419–5427, 2018

    2018

  15. [23]

    Unsupervised event-based learning of optical flow, depth, and egomotion

    Alex Zihao Zhu, Liangzhe Yuan, Kenneth Chaney, and Kostas Daniilidis. Unsupervised event-based learning of optical flow, depth, and egomotion. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 989–997, 2019

    2019

  16. [24]

    Ev-flownet: Self-supervised optical flow estimation for event-based cameras

    Alex Zhu, Liangzhe Yuan, Kenneth Chaney, and Kostas Daniilidis. Ev-flownet: Self-supervised optical flow estimation for event-based cameras. In Proceedings of Robotics: Science and Systems, Pittsburgh, Pennsylvania, June 2018. doi: 10.15607/RSS.2018.XIV .062

    2018 doi

  17. [25]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical image computing and computer-assisted intervention, pages 234–241. Springer, 2015

    2015

  18. [26]

    You only look once: Unified, real-time object detection

    Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. You only look once: Unified, real-time object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 779–788, 2016

    2016

  19. [27]

    Adaptive temporal pooling for object detection using dynamic vision sensor

    Jia Li, Feng Shi, Wei-Heng Liu, Dongqing Zou, Qiang Wang, Paul KJ Park, and Hyunsurk Ryu. Adaptive temporal pooling for object detection using dynamic vision sensor. In BMVC, 2017

    2017

  20. [28]

    Faster r-cnn: Towards real-time object detection with region proposal networks

    Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object detection with region proposal networks. In Advances in neural information processing systems, pages 91–99, 2015

    2015

  21. [29]

    Converting static image datasets to spiking neuromorphic datasets using saccades

    Garrick Orchard, Ajinkya Jayawant, Gregory K Cohen, and Nitish Thakor. Converting static image datasets to spiking neuromorphic datasets using saccades. Frontiers in neuroscience, 9:437, 2015

    2015

  22. [30]

    A low power, fully event- based gesture recognition system

    Arnon Amir, Brian Taba, David Berg, Timothy Melano, Jeffrey McKinstry, Carmelo Di Nolfo, Tapan Nayak, Alexander Andreopoulos, Guillaume Garreau, Marcela Mendoza, et al. A low power, fully event- based gesture recognition system. In Proceedings of the IEEE Conference on Compute...

    2017

  23. [31]

    Ddd17: End-to-end davis driving dataset

    Jonathan Binas, Daniel Neil, Shih-Chii Liu, and Tobi Delbruck. Ddd17: End-to-end davis driving dataset. arXiv preprint arXiv:1711.01458, 2017

    2017 arXiv

  24. [32]

    The multivehicle stereo event camera dataset: An event camera dataset for 3d perception

    Alex Zihao Zhu, Dinesh Thakur, Tolga Özaslan, Bernd Pfrommer, Vijay Kumar, and Kostas Daniilidis. The multivehicle stereo event camera dataset: An event camera dataset for 3d perception. IEEE Robotics and Automation Letters, 3(3):2032–2039, 2018

    2018

  25. [33]

    Neuro- morphic benchmark datasets for pedestrian detection, action recognition, and fall detection

    Shu Miao, Guang Chen, Xiangyu Ning, Yang Zi, Kejia Ren, Zhenshan Bing, and Alois C Knoll. Neuro- morphic benchmark datasets for pedestrian detection, action recognition, and fall detection. Frontiers in neurorobotics, 13:38, 2019

    2019

  26. [34]

    Esim: an open event camera simulator

    Henri Rebecq, Daniel Gehrig, and Davide Scaramuzza. Esim: an open event camera simulator. In Conference on Robot Learning, pages 969–982, 2018

    2018

  27. [35]

    Video to events: Bringing modern computer vision closer to event cameras

    Daniel Gehrig, Mathias Gehrig, Javier Hidalgo-Carrió, and Davide Scaramuzza. Video to events: Bringing modern computer vision closer to event cameras. arXiv preprint arXiv:1912.03095, 2019

    1912 arXiv

  28. [36]

    Imagenet classification with deep convolutional neural networks

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. In Advances in neural information processing systems, pages 1097–1105, 2012

    2012

  29. [37]

    Ssd: Single shot multibox detector

    Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, and Alexander C Berg. Ssd: Single shot multibox detector. In European conference on computer vision, pages 21–37. Springer, 2016. 18

    2016

  30. [38]

    Squeeze-and-excitation networks, 2017

    Jie Hu, Li Shen, Samuel Albanie, Gang Sun, and Enhua Wu. Squeeze-and-excitation networks, 2017

    2017

  31. [39]

    Convo- lutional lstm network: A machine learning approach for precipitation nowcasting

    SHI Xingjian, Zhourong Chen, Hao Wang, Dit-Yan Yeung, Wai-Kin Wong, and Wang-chun Woo. Convo- lutional lstm network: A machine learning approach for precipitation nowcasting. In Advances in neural information processing systems, pages 802–810, 2015

    2015

  32. [40]

    Unsupervised learning for physical interaction through video prediction

    Chelsea Finn, Ian Goodfellow, and Sergey Levine. Unsupervised learning for physical interaction through video prediction. In Advances in neural information processing systems, pages 64–72, 2016

    2016

  33. [41]

    Mobile video object detection with temporally-aware feature maps

    Mason Liu and Menglong Zhu. Mobile video object detection with temporally-aware feature maps. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages 5686–5695, 2018

    2018

  34. [42]

    Focal loss for dense object detection

    Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Dollár. Focal loss for dense object detection. In Proceedings of the IEEE international conference on computer vision, pages 2980–2988, 2017

    2017

  35. [43]

    Efficient estimation of word representations in vector space

    Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. Efficient estimation of word representations in vector space. In Yoshua Bengio and Yann LeCun, editors,1st International Conference on Learning Representations, ICLR 2013, Scottsdale, Arizona, USA, May 2-4, 2013, Workshop...

    2013

  36. [44]

    Representation learning with contrastive predictive coding

    Aäron van den Oord, Yazhe Li, and Oriol Vinyals. Representation learning with contrastive predictive coding. CoRR, abs/1807.03748, 2018. URL http://arxiv.org/abs/1807.03748

    2018 arXiv

  37. [45]

    Backpropagation through time: what it does and how to do it

    Paul J Werbos. Backpropagation through time: what it does and how to do it. Proceedings of the IEEE, 78 (10):1550–1560, 1990

    1990

  38. [46]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In European conference on computer vision, pages 740–755. Springer, 2014

    2014

  39. [47]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014

    2014 arXiv

  40. [48]

    Steering a predator robot using a mixed frame/event-driven convolutional neural network

    Diederik Paul Moeys, Federico Corradi, Emmett Kerr, Philip Vance, Gautham Das, Daniel Neil, Dermot Kerr, and Tobi Delbrück. Steering a predator robot using a mixed frame/event-driven convolutional neural network. In 2016 Second International Conference on Event-based Control, ...

    2016

  41. [49]

    Hots: a hierarchy of event-based time-surfaces for pattern recognition

    Xavier Lagorce, Garrick Orchard, Francesco Galluppi, Bertram E Shi, and Ryad B Benosman. Hots: a hierarchy of event-based time-surfaces for pattern recognition. IEEE transactions on pattern analysis and machine intelligence, 39(7):1346–1359, 2016

    2016

  42. [50]

    A differentiable recurrent surface for asynchronous event-based data

    Marco Cannici, Marco Ciccone, Andrea Romanoni, and Matteo Matteucci. A differentiable recurrent surface for asynchronous event-based data. In The European Conference on Computer Vision (ECCV), August 2020

    2020

  43. [51]

    Learning an event sequence embedding for dense event-based deep stereo

    Stepan Tulyakov, Francois Fleuret, Martin Kiefel, Peter Gehler, and Michael Hirsch. Learning an event sequence embedding for dense event-based deep stereo. In Proceedings of the IEEE International Conference on Computer Vision, pages 1527–1537, 2019

    2019

  44. [52]

    End-to-end learning of representations for asynchronous event-based data

    Daniel Gehrig, Antonio Loquercio, Konstantinos G Derpanis, and Davide Scaramuzza. End-to-end learning of representations for asynchronous event-based data. InProceedings of the IEEE International Conference on Computer Vision, pages 5633–5643, 2019

    2019

  45. [53]

    A qvga 143 db dynamic range frame-free pwm image sensor with lossless pixel-level video compression and time-domain cds

    Christoph Posch, Daniel Matolin, and Rainer Wohlgenannt. A qvga 143 db dynamic range frame-free pwm image sensor with lossless pixel-level video compression and time-domain cds. IEEE Journal of Solid-State Circuits, 46(1):259–275, 2010

    2010

  46. [54]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016

    2016

  47. [55]

    Design of an rgbw color vga rolling and global shutter dynamic and active-pixel vision sensor

    Chenghan Li, Christian Brandli, Raphael Berner, Hongjie Liu, Minhao Yang, Shih-Chii Liu, and Tobi Delbruck. Design of an rgbw color vga rolling and global shutter dynamic and active-pixel vision sensor. In 2015 IEEE International Symposium on Circuits and Systems (ISCAS), page...

    2015

  48. [56]

    Truenorth: Design and tool flow of a 65 mw 1 million neuron programmable neurosynaptic chip

    Filipp Akopyan, Jun Sawada, Andrew Cassidy, Rodrigo Alvarez-Icaza, John Arthur, Paul Merolla, Nabil Imam, Yutaka Nakamura, Pallab Datta, Gi-Joon Nam, et al. Truenorth: Design and tool flow of a 65 mw 1 million neuron programmable neurosynaptic chip. IEEE transactions on compute...

    2015

  49. [57]

    Loihi: A neuromorphic manycore processor with on-chip learning

    Mike Davies, Narayan Srinivasa, Tsung-Han Lin, Gautham Chinya, Yongqiang Cao, Sri Harsha Choday, Georgios Dimou, Prasad Joshi, Nabil Imam, Shweta Jain, et al. Loihi: A neuromorphic manycore processor with on-chip learning. IEEE Micro, 38(1):82–99, 2018

    2018

  50. [58]

    Automated vehicles and pedestrian safety: exploring the promise and limits of pedestrian detection

    Tabitha S Combs, Laura S Sandt, Michael P Clamann, and Noreen C McDonald. Automated vehicles and pedestrian safety: exploring the promise and limits of pedestrian detection. American journal of preventive medicine, 56(1):1–7, 2019

    2019

  51. [59]

    A literature review of iot energy platforms aimed at end users

    Miguel M Martín-Lopo, Jaime Boal, and Álvaro Sánchez-Miralles. A literature review of iot energy platforms aimed at end users. Computer Networks, page 107101, 2020

    2020

  52. [60]

    Low-power neuromorphic hardware for signal processing applications: A review of architectural and system-level design approaches

    Bipin Rajendran, Abu Sebastian, Michael Schmuker, Narayan Srinivasa, and Evangelos Eleftheriou. Low-power neuromorphic hardware for signal processing applications: A review of architectural and system-level design approaches. IEEE Signal Processing Magazine, 36(6):97–110, 2019

    2019

  53. [61]

    Scaling, low power needed for a neuromorphic future

    Chris Eliasmith and Terry Stewart. Scaling, low power needed for a neuromorphic future. EE Times. https://www.eetimes.com/scaling-low-power-needed-for-a-neuromorphic-future/, 2020

    2020

  54. [62]

    Mitigating unwanted biases with adversarial learning

    Brian Hu Zhang, Blake Lemoine, and Margaret Mitchell. Mitigating unwanted biases with adversarial learning. In Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society, pages 335–340, 2018

    2018

  55. [63]

    Certified robustness to adversarial examples with differential privacy

    Mathias Lecuyer, Vaggelis Atlidakis, Roxana Geambasu, Daniel Hsu, and Suman Jana. Certified robustness to adversarial examples with differential privacy. In 2019 IEEE Symposium on Security and Privacy (SP), pages 656–672. IEEE, 2019

    2019

  56. [64]

    Attribution-driven causal analysis for detection of adversarial examples

    Susmit Jha, Sunny Raj, Steven Lawrence Fernandes, Sumit Kumar Jha, Somesh Jha, Gunjan Verma, Brian Jalaian, and Ananthram Swami. Attribution-driven causal analysis for detection of adversarial examples. arXiv preprint arXiv:1903.05821, 2019

    1903 arXiv

  57. [65]

    Peernets: Exploiting peer wisdom against adversarial attacks

    Jan Svoboda, Jonathan Masci, Federico Monti, Michael M Bronstein, and Leonidas Guibas. Peernets: Exploiting peer wisdom against adversarial attacks. arXiv preprint arXiv:1806.00088, 2018

    2018 arXiv

  58. [66]

    Trends and issues in safe driver assistance systems: Driver acceptance and assistance for elderly drivers

    Sadayuki Tsugawa. Trends and issues in safe driver assistance systems: Driver acceptance and assistance for elderly drivers. IATSS research, 30(2):6–18, 2006

    2006

  59. [67]

    Self-driving car dilemmas reveal that moral choices are not universal

    Amy Maxmen. Self-driving car dilemmas reveal that moral choices are not universal. Nature, 562(7728): 469–469, 2018

    2018

  60. [68]

    Goldman sachs analysis of autonomous vehicle job loss, 2017

    Anita Balakrishnan. Goldman sachs analysis of autonomous vehicle job loss, 2017

    2017

  61. [69]

    The global expansion of ai surveillance

    Steven Feldstein. The global expansion of ai surveillance. Carnegie Endowment. https://carnegieendowment. org/2019/09/17/global-expansion-of-ai-surveillance-pub-79847, 2019

    2019

  62. [70]

    Artificial intelligence and the future of warfare

    Missy Cummings. Artificial intelligence and the future of warfare. Chatham House for the Royal Institute of International Affairs London, 2017

    2017

  63. [71]

    Military artificial intelligence can be easily and dangerously fooled

    Will Knight. Military artificial intelligence can be easily and dangerously fooled. MIT Techonology Re- view. https://www.technologyreview.com/2019/10/21/132277/military-artificial-intelligence-can-be-easily- and-dangerously-fooled/, 2019

    2019

  64. [72]

    Machine learning for high-speed corner detection

    Edward Rosten and Tom Drummond. Machine learning for high-speed corner detection. In European conference on computer vision, pages 430–443. Springer, 2006

    2006

  65. [73]

    Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography

    Martin A Fischler and Robert C Bolles. Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography. Communications of the ACM, 24(6):381–395, 1981

    1981

  66. [74]

    Asynchronous frameless event-based optical flow

    Ryad Benosman, Sio-Hoi Ieng, Charles Clercq, Chiara Bartolozzi, and Mandyam Srinivasan. Asynchronous frameless event-based optical flow. Neural Networks, 27:32–37, 2012. 20

    2012

  67. [2013]

    URL http://arxiv.org/abs/1301.3781

    arXiv

Pith tools

Reviewed August 27, 2026 · model on record in the stance chip above.