About raw event streams

[caomq123]

Hello,

I would like to inquire about the "raw event streams" mentioned in the paper, where it states, "we propose a Spiking Neural Network (SNN) that takes the raw event streams as input directly for energy-efficient perception." Could you please elaborate on how these raw event streams are loaded and utilized? Additionally, how are they represented in the code? Does the section "Analysis on Different Event Fragments and CLIP" in the paper suggest that the raw event streams are segmented? Could you provide a general overview of the process from loading the raw event streams from a .aedat4 file to inputting them into the SNN for efficient perception?

Thank you for your time and assistance.

[WuZongzhen]

The raw event stream is formatted as {x, y, t, p}. It is segmented into multiple segments based on the time dimension (for instance, n segments), and the time dimension of each segment is scaled to a specific value, such as 16. Consequently, the format of the segmented event stream fragments is {x, y, 16}, which does not include the representation of event polarity. The input network's event stream is represented as n sets of {x, y, 16}.