Pith. sign in

REVIEW 2 major objections 8 minor 49 references

Efficient Event-Based Semantic Segmentation via Exploiting Frame-Event Fusion: A Hybrid Neural Network Approach

T0 review · 2 major / 8 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A hybrid ANN-SNN network with event-guided fusion achieves state-of-the-art accuracy on three event-based segmentation benchmarks while reducing energy by 65%.

desk verdict A genuinely effective hybrid ANN-SNN segmentation architecture with credible accuracy gains, but the headline 65% energy saving rests on an unverifiable FLOPs estimate and needs a much more transparent accounting. read the letter →

arxiv 2507.03765 v1 pith:SDBXL23L submitted 2025-07-04 cs.CV

Hebei Li , Yansong Peng , Jiahui Yuan , Peixi Wu , Jin Wang , Yueyi Zhang , Xiaoyan Sun This is my paper
classification cs.CV
keywords event-basedsemanticsegmentationspikingneuralnetworkshybridANN-SNNframe-eventfusiondeformableattentionenergy-efficientvisionneuromorphicadaptivetemporalweighting
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 proposes a hybrid network for event-based semantic segmentation that runs frames through an artificial neural network and events through a spiking neural network, fusing the two streams at intermediate layers. The authors claim that three interaction modules—adaptive temporal weighting, event-guided sparse injection, and channel selection fusion—let the network beat prior methods on three benchmarks while using far less energy. Their reported results reach 67.31, 66.57, and 50.91 mIoU on DDD17-Seg, DSEC-Semantic, and a newly introduced M3ED-Semantic dataset, with a 65% energy reduction over the previous most efficient method on DSEC-Semantic. The point of the work is to show that active, targeted fusion of event and frame features can improve both accuracy and efficiency.

What carries the argument

The load-bearing machinery is the event-guided fusion loop between an ANN branch and an SNN branch. The ATW Injector squeezes SNN features across space, passes them through an adapter to produce temporal weights, and applies softmax-normalized weighting before a sparse cross-attention layer with frame features. The EDS Injector downsamples the event voxel grid to obtain non-zero positions as reference points, computes offsets from SNN features via a shared MLP, and runs deformable attention that samples projected frame features at those event-guided locations. The CSF module applies a squeeze-and-excitation-style channel selection to each branch's features and sums the weighted maps. A Mask2Former decoder consumes the multi-scale fused features. The design idea is that event sparsity tells the network where motion matters, so attention can be sparse rather than dense.

What would settle it

Run the released model on a neuromorphic chip or a detailed cycle-accurate energy simulator that charges full cost for every non-spike operation (softmax, adapter MLP, deformable attention, channel selection), measuring total energy per inference on DSEC-Semantic; if the result exceeds 33.45 mJ by a substantial margin, the energy-efficiency claim fails. A simpler check: count the FLOPs of the ATW and EDS injectors including all float operations, and verify whether the 0.110 GFLOP SNN budget accounts for them.

Watch

Extended reading notes

Core claim

The paper's central claim is that a hybrid ANN-SNN architecture with three specialized fusion modules achieves state-of-the-art accuracy in event-based semantic segmentation while consuming substantially less energy than existing methods. The ANN branch extracts spatial and color features from frames; the SNN branch processes event voxel grids with spiking neurons; the ATW Injector weights SNN temporal features and injects them into the frame stream through cross-attention; the EDS Injector uses event locations as sparse reference points for deformable attention, pulling frame color features into the SNN stream; and the CSF module selects and weights the most informative channels from both branches before the Mask2Former decoder. The authors support the claim with experiments on DDD17-Seg, DSEC-Semantic, and a new M3ED-Semantic subset, reporting the highest mIoU on all three and a 65% lower inference energy on DSEC-Semantic than HALSIE, previously the most energy-efficient method.

Load-bearing premise

The load-bearing premise is that the reported energy figures (33.45 mJ on DSEC-Semantic) come from a fair accounting that counts the real-valued operations inside the hybrid modules—softmax attention, adapter MLPs, and deformable sampling—at realistic cost; if those operations are omitted or underestimated, the 65% energy reduction is overstated.

Editorial extensions

If this is right

  • If the reported numbers hold, hybrid ANN-SNN fusion is a practical recipe for event-based dense prediction, not just classification.
  • Event-guided sparse attention reduces the cost of cross-modal fusion by sampling only where events occur, so the approach should scale better to higher resolutions than dense fusion.
  • The three modules each provide a measurable gain in the ablation, implying that temporal weighting, sparse event-guided sampling, and channel selection are complementary rather than redundant.
  • The new M3ED-Semantic benchmark gives the community a multi-platform outdoor dataset (drone and vehicle) for evaluating event segmentation.

Reading between the lines

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

  • The authors' efficiency claim presumably assumes spike-based or low-power execution for the SNN branch; on a conventional GPU the attention and softmax operations could dominate, so the 65% reduction may not transfer to standard hardware without re-measurement.
  • The same fusion pattern—use event sparsity to choose reference points and temporal weights—could be applied to other tasks that combine a fast asynchronous sensor with a slower frame sensor, such as depth estimation or object tracking.
  • Because the CSF module is a lightweight channel gating, the framework suggests a general design: fuse modalities at multiple scales by selection rather than concatenation, which keeps parameter counts low (1.79M).
  • The M3ED-Semantic introduction is a contribution in its own right; a larger version with more labels could become a standard benchmark for event segmentation in robotics.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 8 minor

Summary. The paper proposes a hybrid ANN-SNN framework for event-based semantic segmentation, with an ANN branch for frames and an SNN branch for events, connected by three fusion modules: the Adaptive Temporal Weighting (ATW) Injector, the Event-Driven Sparse (EDS) Injector, and the Channel Selection Fusion (CSF) module. The method is evaluated on DDD17-Seg, DSEC-Semantic, and a newly introduced M3ED-Semantic subset, reporting state-of-the-art mIoU (67.31, 66.57, and 50.91 respectively) and claiming a 65% energy reduction over the most efficient prior method on DSEC-Semantic. The ablation study shows positive contributions from each module, and the code is publicly available.

Significance. If the results are reproducible, the paper makes a strong empirical contribution by demonstrating that active intermediate fusion of frame and event features in a hybrid ANN-SNN architecture can improve segmentation accuracy while maintaining low parameter count and estimated energy consumption. The design of the three modules is well motivated, and the ablations provide clear evidence for their individual roles. The introduction of M3ED-Semantic adds a new benchmark within a multi-robot, multi-sensor dataset, which could benefit the community. The availability of code is a strength. However, the central efficiency claim depends on an energy-estimation protocol that is deferred to the supplementary material, and the new benchmark is evaluated against only three baselines; these gaps affect how strongly the results can be trusted.

major comments (2)
  1. [Experiments and Evaluation (Tables 1-2)] The headline efficiency claim—a 65% energy reduction on DSEC-Semantic (33.45 mJ vs. 94.41 mJ for HALSIE)—rests entirely on an energy-estimation procedure that the main text describes only as 'calculating the average number of FLOPs' and defers to the supplementary. The architecture includes real-valued operations in the ATW Injector (softmax, attention), EDS Injector (shared MLP, deformable attention with learnable offsets), and CSF module (Conv, Sigmoid, AvgPool). The main text does not specify whether these operations are charged at full ANN-level MAC cost or at the lower spike-based energy rate used for SNN operations, nor does it provide the per-operation energy constants. Without this information, the reported E_total values and the 65% reduction cannot be verified from the manuscript. Please include the full energy-accounting equation and the assumptions for each operation type in the main text, or at least state how the hybrid modules are accounted for, so that a reader can audit the comparison against HALSIE and other baselines.
  2. [Evaluation on M3ED-Semantic Dataset (Table 2)] The paper introduces M3ED-Semantic as a new benchmark subset and claims state-of-the-art performance on it, but the comparison includes only E2VID, ESS-Sup (E), and ESS-Sup (E+F). In particular, HALSIE—the most relevant hybrid baseline, which is evaluated on DDD17-Seg and DSEC-Semantic—is missing from the M3ED-Semantic comparison, as are other recent methods such as EV-Segformer and OpenESS. Without these baselines, the claim of state-of-the-art accuracy on M3ED-Semantic is not well supported. Please evaluate at least HALSIE, and ideally other recent methods, under the same protocol on M3ED-Semantic, or present the benchmark as a new resource with a more limited comparative claim.
minor comments (8)
  1. [Ablation Study (Table 3)] The table header contains 'A WT' instead of 'ATW', and the text at one point uses 'AWT'; please use 'ATW' consistently throughout.
  2. [Ablation Study (Table 3)] The 'F-E' column label in Table 3 is not defined; please specify that it denotes the frame-event dual-branch baseline without the three proposed modules.
  3. [Methodology, Eq. (2)] The text says the adaptor output is averaged across the channel dimension C to produce a temporal weighting vector W_SNN ∈ R^{B×T×C}; if averaged over C, the result should be in R^{B×T}. Please clarify the intended shape and the corresponding notation.
  4. [Methodology, Eq. (3)] The variable α_{t,c} is introduced without explicit definition; state that it is the softmax output of W_SNN along the channel and temporal dimensions.
  5. [Methodology, Eq. (7)] The expression F_P^roj_ANN · F_SNN[r_k + Δr_k] is ambiguous; please clarify whether the dot denotes element-wise multiplication or sampling-based interpolation, and define the dimensions after sampling.
  6. [Introduction, Figure 1] Figure 1 is referenced in the abstract and introduction but not directly cited in the main text when discussing energy comparisons; please add an explicit reference at the appropriate location.
  7. [Methodology, Main Architecture] The Mask2Former decoder is mentioned but no citation is given for Mask2Former; add the appropriate reference.
  8. [Experiments and Evaluation] All experimental results are reported as single runs without error bars or standard deviations. Given the small differences between some ablation configurations (e.g., Table 3 rows 5-7), reporting mean and std over multiple seeds would strengthen the conclusions.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: accuracy claims are benchmark-driven empirical results, and no fitted parameter is relabeled as a prediction.

full rationale

This is an empirical architecture paper. The central claims are state-of-the-art segmentation accuracy on three external benchmarks and reduced estimated inference energy. The proposed ATW, EDS, and CSF modules are defined by explicit equations (Eqs. 2–8) and their contributions are tested by ablations (Table 3), not derived from a theory that assumes the outputs. No fitted parameter is presented as a prediction; the energy numbers are estimates whose methodology is deferred to the supplementary material, which is an auditability/correctness concern rather than a circularity concern. The paper cites prior work by the same authors (e.g., Li et al. 2024a, 2024b) only as related work and does not rest the architecture or conclusions on those citations. The M3ED-Semantic dataset is newly proposed, but it is a labeled subset of the existing M3ED dataset and is used as a benchmark, not as evidence for the method's validity. Overall, the derivation chain is independent: modules are architectural choices, performance is measured against external baselines, and ablations attribute gains to specific components. Therefore the circularity score is minimal, reflecting only the presence of self-citations that are not load-bearing and the internally referenced energy-estimation protocol that is not fully auditable from the main text.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The central empirical claims rest on standard deep learning components (Mask2Former, deformable attention, LIF surrogate training, voxel grids) plus the authors' new M3ED-Semantic benchmark and their own energy estimation protocol. The latter two are not externally validated in the main text.

free parameters (6)
  • SNN timestep T = 5
    Tuned via ablation (Table 5) as best among {1,3,5,7}; affects accuracy and energy.
  • Voxel temporal bins B = 5
    Set to match timestep; discretizes event windows into 5 bins.
  • LIF decay tau = 2.0
    Chosen for the LIF neuron; not ablated.
  • LIF threshold = 1.0
    Spike threshold; not ablated.
  • Deformable attention sample points K = not specified in paper
    Default from Zhu et al. 2020; not reported.
  • Event volume per voxel = 6400 (DDD17), 100000 (DSEC/M3ED)
    Data preprocessing choice; may affect results.
assumptions (6)
  • domain assumption Mask2Former decoder provides a suitable segmentation head for the fused features.
    Borrowed from prior work; assumed to work with hybrid features.
  • domain assumption Deformable attention (Zhu et al. 2020) can be used for cross-modal feature alignment with event-guided reference points.
    Core mechanism of the EDS injector.
  • domain assumption LIF neuron with surrogate gradient backpropagation is a valid SNN training approach for this architecture.
    Standard for direct SNN training.
  • domain assumption Voxel grid representation of events preserves task-relevant temporal information.
    Standard event representation from Gehrig et al. 2019.
  • domain assumption The M3ED-Semantic annotations are correct and the subset is representative.
    The dataset is introduced in this paper without external validation.
  • ad hoc to paper The energy estimation method gives a fair comparison across ANN, SNN, and hybrid models.
    The estimation protocol is only referenced to the supplementary material and is not auditable in the main text.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Efficient Event-Based Semantic Segmentation via Exploiting Frame-Event Fusion: A Hybrid Neural Network Approach." pith.science (2026). https://pith.science/paper/SDBXL23L

@misc{pith2026250703765,
  author       = {Pith},
  title        = {Pith review of: Efficient Event-Based Semantic Segmentation via Exploiting Frame-Event Fusion: A Hybrid Neural Network Approach},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SDBXL23L}},
  note         = {Machine review of arXiv:2507.03765}
}
read the original abstract

Event cameras have recently been introduced into image semantic segmentation, owing to their high temporal resolution and other advantageous properties. However, existing event-based semantic segmentation methods often fail to fully exploit the complementary information provided by frames and events, resulting in complex training strategies and increased computational costs. To address these challenges, we propose an efficient hybrid framework for image semantic segmentation, comprising a Spiking Neural Network branch for events and an Artificial Neural Network branch for frames. Specifically, we introduce three specialized modules to facilitate the interaction between these two branches: the Adaptive Temporal Weighting (ATW) Injector, the Event-Driven Sparse (EDS) Injector, and the Channel Selection Fusion (CSF) module. The ATW Injector dynamically integrates temporal features from event data into frame features, enhancing segmentation accuracy by leveraging critical dynamic temporal information. The EDS Injector effectively combines sparse event data with rich frame features, ensuring precise temporal and spatial information alignment. The CSF module selectively merges these features to optimize segmentation performance. Experimental results demonstrate that our framework not only achieves state-of-the-art accuracy across the DDD17-Seg, DSEC-Semantic, and M3ED-Semantic datasets but also significantly reduces energy consumption, achieving a 65\% reduction on the DSEC-Semantic dataset.

Figures

Figures reproduced from arXiv: 2507.03765 by the authors.

Figure 1
Figure 1. Comparison of various event-based semantic seg [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of the proposed framework.The architecture integrates ANN and SNN branches for efficient semantic seg [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Illustration of the proposed Adaptive Temporal [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (1 more)
Figure 5
Figure 5. Figure 5: Qualitative segmentation results from the DDD17-Seg, DSEC-Semantic, and M3ED-Semantic datasets. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

49 extracted references · 43 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Akopyan, F.; Sawada, J.; Cassidy, A.; Alvarez-Icaza, R.; Arthur, J.; Merolla, P.; Imam, N.; Nakamura, Y.; Datta, P.; Nam, G.-J.; et al. 2015. TrueNorth: Design and Tool Flow of a 65 mW 1 Million Neuron Programmable Neurosynaptic Chip. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 34(10): 1537--1557

    work page 2015

  4. [4]

    Alonso, I.; and Murillo, A. C. 2019. EV-SegNet: Semantic segmentation for event-based cameras. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

    work page 2019

  5. [5]

    D.; Kosta, A.; Liyanagedera, C.; Apolinario, M.; and Roy, K

    Biswas, S. D.; Kosta, A.; Liyanagedera, C.; Apolinario, M.; and Roy, K. 2024. Halsie: Hybrid approach to learning segmentation by simultaneously exploiting image and event modalities. In 2024 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 5952--5962. IEEE

    work page 2024

  6. [6]

    Cao, J.; Zheng, X.; Lyu, Y.; Wang, J.; Xu, R.; and Wang, L. 2024. Chasing day and night: Towards robust and efficient all-day object detection guided by an event camera. In 2024 IEEE International Conference on Robotics and Automation (ICRA), 9026--9032. IEEE

    work page 2024

  7. [7]

    A.; Korpela, C.; Kumar, V.; Taylor, C

    Chaney, K.; Cladera, F.; Wang, Z.; Bisulco, A.; Hsieh, M. A.; Korpela, C.; Kumar, V.; Taylor, C. J.; and Daniilidis, K. 2023. M3ed: Multi-robot, multi-sensor, multi-environment event dataset. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 4016--4023

    work page 2023

  8. [8]

    H.; Dimou, G.; Joshi, P.; Imam, N.; Jain, S.; et al

    Davies, M.; Srinivasa, N.; Lin, T.-H.; Chinya, G.; Cao, Y.; Choday, S. H.; Dimou, G.; Joshi, P.; Imam, N.; Jain, S.; et al. 2018. Loihi: A Neuromorphic Manycore Processor with On-Chip Learning. IEEE Micro, 38(1): 82--99

    work page 2018

Show all 49 references
  1. [9]

    U.; Neil, D.; Binas, J.; Cook, M.; Liu, S.-C.; and Pfeiffer, M

    Diehl, P. U.; Neil, D.; Binas, J.; Cook, M.; Liu, S.-C.; and Pfeiffer, M. 2015. Fast-Classifying, High-Accuracy Spiking Deep Networks Through Weight and Threshold Balancing. In 2015 International Joint Conference on Neural Networks (IJCNN), 1--8. IEEE

    2015

  2. [10]

    B.; Galluppi, F.; Temple, S.; and Plana, L

    Furber, S. B.; Galluppi, F.; Temple, S.; and Plana, L. A. 2014. The SpiNNaker Project. Proceedings of the IEEE, 102(5): 652--665

    2014

  3. [11]

    J.; Conradt, J.; Daniilidis, K.; et al

    Gallego, G.; Delbr \"u ck, T.; Orchard, G.; Bartolozzi, C.; Taba, B.; Censi, A.; Leutenegger, S.; Davison, A. J.; Conradt, J.; Daniilidis, K.; et al. 2020. Event-based vision: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 44(1): 154--180

    2020

  4. [12]

    Gehrig, D.; Gehrig, M.; Hidalgo-Carri \'o , J.; and Scaramuzza, D. 2020. Video to events: Recycling video datasets for event cameras. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 3586--3595

    2020

  5. [13]

    G.; and Scaramuzza, D

    Gehrig, D.; Loquercio, A.; Derpanis, K. G.; and Scaramuzza, D. 2019. End-to-end learning of representations for asynchronous event-based data. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 5633--5643

    2019

  6. [14]

    Gehrig, D.; and Scaramuzza, D. 2022. Are high-resolution event cameras really needed? arXiv preprint arXiv:2203.14672

    2022 arXiv

  7. [15]

    Gerstner, W.; and Kistler, W. M. 2002. Spiking neuron models: Single neurons, populations, plasticity. Cambridge university press

    2002

  8. [16]

    Hamaguchi, R.; Furukawa, Y.; Onishi, M.; and Sakurada, K. 2023. Hierarchical neural memory network for low latency event processing. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 22867--22876

    2023

  9. [17]

    Houlsby, N.; Giurgiu, A.; Jastrzebski, S.; Morrone, B.; De Laroussilhe, Q.; Gesmundo, A.; Attariyan, M.; and Gelly, S. 2019. Parameter-efficient transfer learning for NLP. In International Conference on Machine Learning (ICML), 2790--2799. PMLR

    2019

  10. [18]

    Hu, J.; Shen, L.; and Sun, G. 2018. Squeeze-and-excitation networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 7132--7141

    2018

  11. [19]

    Huang, K.; Zhang, S.; Zhang, J.; and Tao, D. 2023. Event-based simultaneous localization and mapping: A comprehensive survey. arXiv preprint arXiv:2304.09793

    2023 arXiv

  12. [20]

    Jia, Z.; You, K.; He, W.; Tian, Y.; Feng, Y.; Wang, Y.; Jia, X.; Lou, Y.; Zhang, J.; Li, G.; et al. 2023. Event-based semantic segmentation with posterior attention. IEEE Transactions on Image Processing (TIP), 32: 1829--1842

    2023

  13. [21]

    Jiang, J.; Zhou, X.; Wang, B.; Deng, X.; Xu, C.; and Shi, B. 2024. Complementing event streams and rgb frames for hand mesh reconstruction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 24944--24954

    2024

  14. [22]

    Jing, L.; Ding, Y.; Gao, Y.; Wang, Z.; Yan, X.; Wang, D.; Schaefer, G.; Fang, H.; Zhao, B.; and Li, X. 2024. HPL-ESS: Hybrid Pseudo-Labeling for Unsupervised Event-based Semantic Segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition...

    2024

  15. [23]

    Kim, Y.; Chough, J.; and Panda, P. 2022. Beyond classification: Directly training spiking neural networks for semantic segmentation. Neuromorphic Computing and Engineering, 2(4): 044015

    2022

  16. [24]

    X.; Cottereau, B

    Kong, L.; Liu, Y.; Ng, L. X.; Cottereau, B. R.; and Ooi, W. T. 2024. OpenESS: Event-based Semantic Scene Understanding with Open Vocabularies. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 15686--15698

    2024

  17. [25]

    Kugele, A.; Pfeil, T.; Pfeiffer, M.; and Chicca, E. 2021. Hybrid SNN-ANN: Energy-efficient classification and object detection for event-based vision. In DAGM German Conference on Pattern Recognition, 297--312. Springer

    2021

  18. [26]

    K.; and Roy, K

    Lee, C.; Kosta, A. K.; and Roy, K. 2022. Fusion-FlowNet: Energy-efficient optical flow estimation using sensor fusion and deep fused spiking-analog network architectures. In 2022 International Conference on Robotics and Automation (ICRA), 6504--6510. IEEE

    2022

  19. [27]

    Li, H.; Wang, J.; Yuan, J.; Li, Y.; Weng, W.; Peng, Y.; Zhang, Y.; Xiong, Z.; and Sun, X. 2024 a . Event-assisted Low-Light Video Object Segmentation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 3250--3259

    2024

  20. [28]

    Li, H.; Zhang, Y.; Xiong, Z.; and Sun, X. 2024 b . Deep multi-threshold spiking-UNet for image processing. Neurocomputing, 586: 127653

    2024

  21. [29]

    Liao, X.; Wu, Y.; Wang, Z.; Wang, D.; and Zhang, H. 2023. A Convolutional Spiking Neural Network with Adaptive Coding for Motor Imagery Classification. Neurocomputing, 126470

    2023

  22. [30]

    Lichtsteiner, P.; Posch, C.; and Delbruck, T. 2008. A 128 128 120 dB 15 s latency asynchronous temporal contrast vision sensor. IEEE Journal of Solid-state Circuits, 43(2): 566--576

    2008

  23. [31]

    Liu, Z.; Wu, J.; Shi, G.; Yang, W.; Dong, W.; and Zhao, Q. 2023. Motion-oriented hybrid spiking neural networks for event-based motion deblurring. IEEE Transactions on Circuits and Systems for Video Technology (TCSVT)

    2023

  24. [32]

    Messikommer, N.; Gehrig, D.; Gehrig, M.; and Scaramuzza, D. 2022. Bridging the gap between events and frames through unsupervised domain adaptation. IEEE Robotics and Automation Letters (RAL), 7(2): 3515--3522

    2022

  25. [33]

    Patel, K.; Hunsberger, E.; Batir, S.; and Eliasmith, C. 2021. A spiking neural network for image segmentation. arXiv preprint arXiv:2106.08921

    2021 arXiv

  26. [34]

    Pei, J.; Deng, L.; Song, S.; Zhao, M.; Zhang, Y.; Wu, S.; Wang, G.; Zou, Z.; Wu, Z.; He, W.; et al. 2019. Towards artificial general intelligence with hybrid Tianjic chip architecture. Nature, 572(7767): 106--111

    2019

  27. [35]

    Rebecq, H.; Ranftl, R.; Koltun, V.; and Scaramuzza, D. 2019. High speed and high dynamic range video with an event camera. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 43(6): 1964--1980

    2019

  28. [36]

    Rueckauer, B.; Lungu, I.-A.; Hu, Y.; Pfeiffer, M.; and Liu, S.-C. 2017. Conversion of Continuous-Valued Deep Networks to Efficient Event-Driven Networks for Image Classification. Frontiers in Neuroscience, 11: 682

    2017

  29. [37]

    Sengupta, A.; Ye, Y.; Wang, R.; Liu, C.; and Roy, K. 2019. Going Deeper in Spiking Neural Networks: VGG and Residual Architectures. Frontiers in Neuroscience, 13: 95

    2019

  30. [38]

    Sun, Z.; Messikommer, N.; Gehrig, D.; and Scaramuzza, D. 2022. Ess: Learning event-based semantic segmentation from still images. In European Conference on Computer Vision (ECCV), 341--357. Springer

    2022

  31. [39]

    Wang, L.; Chae, Y.; and Yoon, K.-J. 2021. Dual transfer learning for event-based end-task prediction via pluggable event to image translation. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2135--2145

    2021

  32. [40]

    Wang, L.; Chae, Y.; Yoon, S.-H.; Kim, T.-K.; and Yoon, K.-J. 2021. Evdistill: Asynchronous events to end-task learning via bidirectional reconstruction-guided cross-modal knowledge distillation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogniti...

    2021

  33. [41]

    Wu, Y.; Deng, L.; Li, G.; Zhu, J.; and Shi, L. 2018. Spatio-Temporal Backpropagation for Training High-Performance Spiking Neural Networks. Frontiers in Neuroscience, 12: 331

    2018

  34. [42]

    Xia, R.; Zhao, C.; Zheng, M.; Wu, Z.; Sun, Q.; and Tang, Y. 2023. Cmda: Cross-modality domain adaptation for nighttime semantic segmentation. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 21572--21581

    2023

  35. [43]

    Yang, Q.; Wu, J.; Zhang, M.; Chua, Y.; Wang, X.; and Li, H. 2022. Training Spiking Neural Networks with Local Tandem Learning. 36rd Conference on Neural Information Processing Systems (NIPS)

    2022

  36. [44]

    D.; Bihl, T.; and Liu, J

    Yue, Y.; Baltes, M.; Abujahar, N.; Sun, T.; Smith, C. D.; Bihl, T.; and Liu, J. 2023. Hybrid Spiking Neural Networks Fine-Tuning for Hippocampus Segmentation. In 2023 IEEE 20th International Symposium on Biomedical Imaging (ISBI), 1--5. IEEE

    2023

  37. [45]

    Zhang, L.; Zhou, S.; Zhi, T.; Du, Z.; and Chen, Y. 2019. TDSNN: From Deep Neural Networks to Deep Spike Neural Networks with Temporal-Coding. In Proceedings of the AAAI Conference on Artificial Intelligence, 1319--1326

    2019

  38. [46]

    Zhang, R.; Leng, L.; Che, K.; Zhang, H.; Cheng, J.; Guo, Q.; Liao, J.; and Cheng, R. 2024. Accurate and efficient event-based semantic segmentation using adaptive spiking encoder-decoder network. IEEE Transactions on Neural Networks and Learning Systems (TNNLS)

    2024

  39. [47]

    Zhang, X.; Liao, W.; Yu, L.; Yang, W.; and Xia, G.-S. 2021. Event-based synthetic aperture imaging with a hybrid network. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 14235--14244

    2021

  40. [48]

    Zhou, Z.; Wu, Z.; Boutteau, R.; Yang, F.; Demonceaux, C.; and Ginhac, D. 2023. Rgb-event fusion for moving object detection in autonomous driving. In 2023 IEEE International Conference on Robotics and Automation (ICRA), 7808--7815. IEEE

    2023

  41. [49]

    Zhu, X.; Su, W.; Lu, L.; Li, B.; Wang, X.; and Dai, J. 2020. Deformable DETR: Deformable Transformers for End-to-End Object Detection. In International Conference on Learning Representations (ICLR)

    2020

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.