Sequence-SOD: Sequence-aware Spiking Object Detection for Event Cameras
Keywords: Event Camera, Spiking Neural Network, Object Detection
Abstract: Due to the asynchronous sensing of changes in illumination by event cameras, they are highly energy-efficient and therefore exhibit great potential especially in mobile, low power scenarios. Moreover, they are able to acquire sparse data with a high temporal resolution in the order of milliseconds and achieve a large dynamic range. This enables the recording of reliable data with minimal motion blur even during rapid movements and in low light scenarios. SNNs are particularly suitable for the processing of event data due to their asynchronous and spike-based functionality while their low energy consumption enables their deployment in automotive embedded applications. However, recent spiking object detectors do not leverage the full temporal information and only consider a single, fixed-size sample of the event data. In this paper, we propose the first sequence-aware SNN, which processes long sequences of the event stream data and predicts bounding boxes with a frequency of 40 Hz. In combination with a SSD network design, we are able to reach 26.88 mAP on the Gen1 Automotive Detection Dataset.
Primary Area: applications to robotics, autonomy, planning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 3276
Loading