Uploaded August 2026 | Updated November 2026, 35 minutes ago
Event cameras provide sequential visual data with spatial sparsity and high temporal resolution, making them attractive for low-latency object detection. Existing asynchronous event-based neural networks exploit this low-latency advantage by updating predictions event by event, but still suffer from two bottlenecks: recurrent architectures are difficult to train efficiently on long sequences, and improving accuracy often increases per-event computation and latency. Linear attention is appealing because it enables parallel training and recurrent inference. However, its dense state updates make per-event computation scale with the state size, yielding a poor accuracy-efficiency trade-off for object detection, where accurate localization requires fine-grained spatial states. The key challenge is therefore to introduce sparse state activation that exploits the spatial sparsity of events while preserving efficient parallel training. We propose Spatially-Sparse Linear Attention (SSLA), which introduces a mixture-of-spaces state decomposition and a scatter-compute-gather training procedure, enabling state-level sparsity as well as training parallelism. Building on SSLA, we develop an end-to-end asynchronous linear attention model, SSLA-Det, for low-latency event-based object detection. On Gen1 and N-Caltech101, SSLA-Det achieves state-of-the-art accuracy among asynchronous methods, reaching 0.375 mAP and 0.515 mAP, respectively, while reducing per-event computation by over 20 times compared with the strongest prior asynchronous baseline, demonstrating the potential of linear attention for low-latency event-based vision.
Reference:
Haiqing Hao, Zhipeng Sui, Rong Zou, Zijia Dai, Nikola Zubić, Davide Scaramuzza, Wenhui Wang
"Low-Latency Event-Based Object Detection with Spatially-Sparse Linear Attention"
European Conference on Computer Vision (ECCV), 2026
PDF: https://rpg.ifi.uzh.ch/docs/hao_eccv2026.pdf
Code: github.com/haohq19/ssla
Our research page on event-based vision: http://rpg.ifi.uzh.ch/research_dvs.html
For event-camera datasets, see here:
1. https://dsec.ifi.uzh.ch/
2. https://rpg.ifi.uzh.ch/davis_data.html
3. github.com/uzh-rpg/event-based_vision_resources#datasets
For an event camera simulator: http://rpg.ifi.uzh.ch/esim
For a survey paper on event cameras, see here: https://rpg.ifi.uzh.ch/docs/EventVisionSurvey.pdf
Other resources on event cameras (publications, software, drivers, where to buy, etc.): github.com/uzh-rpg/event-based_vision_resources
Affiliation:
Haiqing Hao, Zhipeng Sui, and Wenhui Wang are with the State Key Laboratory of Precision Measurement Technology and Instruments, Department of Precision Instrument, Tsinghua University, China
Rong Zou, Nikola Zubić, and Davide Scaramuzza are with the Robotics and Perception Group, University of Zurich, Switzerland https://rpg.ifi.uzh.ch/
Zijia Dai is with the ShanghaiTech University, China
Event cameras provide sequential visual data with spatial sparsity and high temporal resolution, making them attractive for low-latency object detection. Existing asynchronous event-based neural networks exploit this low-latency advantage by updating predictions event by event, but still suffer from two bottlenecks: recurrent architectures are difficult to train efficiently on long sequences, and improving accuracy often increases per-event computation and latency. Linear attention is appealing because it enables parallel training and recurrent inference. However, its dense state updates make per-event computation scale with the state size, yielding a poor accuracy-efficiency trade-off for object detection, where accurate localization requires fine-grained spatial states. The key challenge is therefore to introduce sparse state activation that exploits the spatial sparsity of events while preserving efficient parallel training. We propose Spatially-Sparse Linear Attention (SSLA), which introduces a mixture-of-spaces state decomposition and a scatter-compute-gather training procedure, enabling state-level sparsity as well as training parallelism. Building on SSLA, we develop an end-to-end asynchronous linear attention model, SSLA-Det, for low-latency event-based object detection. On Gen1 and N-Caltech101, SSLA-Det achieves state-of-the-art accuracy among asynchronous methods, reaching 0.375 mAP and 0.515 mAP, respectively, while reducing per-event computation by over 20 times compared with the strongest prior asynchronous baseline, demonstrating the potential of linear attention for low-latency event-based vision.
Reference:
Haiqing Hao, Zhipeng Sui, Rong Zou, Zijia Dai, Nikola Zubić, Davide Scaramuzza, Wenhui Wang
"Low-Latency Event-Based Object Detection with Spatially-Sparse Linear Attention"
European Conference on Computer Vision (ECCV), 2026
PDF: https://rpg.ifi.uzh.ch/docs/hao_eccv2026.pdf
Code: github.com/haohq19/ssla
Our research page on event-based vision: http://rpg.ifi.uzh.ch/research_dvs.html
For event-camera datasets, see here:
1. https://dsec.ifi.uzh.ch/
2. https://rpg.ifi.uzh.ch/davis_data.html
3. github.com/uzh-rpg/event-based_vision_resources#datasets
For an event camera simulator: http://rpg.ifi.uzh.ch/esim
For a survey paper on event cameras, see here: https://rpg.ifi.uzh.ch/docs/EventVisionSurvey.pdf
Other resources on event cameras (publications, software, drivers, where to buy, etc.): github.com/uzh-rpg/event-based_vision_resources
Affiliation:
Haiqing Hao, Zhipeng Sui, and Wenhui Wang are with the State Key Laboratory of Precision Measurement Technology and Instruments, Department of Precision Instrument, Tsinghua University, China
Rong Zou, Nikola Zubić, and Davide Scaramuzza are with the Robotics and Perception Group, University of Zurich, Switzerland https://rpg.ifi.uzh.ch/
Zijia Dai is with the ShanghaiTech University, China










