Speck Deployment Parameters

The following parameters are relevant when running LENS in event-driven mode on the Speck2fDevKit. These include options for data collection, on-chip deployment, simulation, and saving inputs.

LENS Speck2fDevKit Parameters

Argument	Type	Description
--event_driven	flag	Enables event-driven mode using the on-chip LENS model deployed to the Speck2fDevKit. Required to trigger live inferencing from event streams.
--simulated_speck	flag	Runs an offline, time-based simulation of the Speck inferencing model using saved spike data. Useful for debugging or replicating deployment behavior without hardware.
--collect_data	flag	Activates data collection mode from the Speck2fDevKit. Captured spikes are converted into images and saved for training new models.
--save_input	flag	If set, saves the raw input spike events as NumPy arrays for later analysis or use in simulated inference.
--headless	flag	Runs the system without a GUI preview. This is essential for deployment on headless devices or remote operation.
--dataset	str	Top-level dataset folder under lens/dataset/ where captured or processed data will be saved and referenced.
--camera	str	Subfolder indicating the camera device or identifier (e.g., speck, speckkit001).
--reference	str	Name of the subfolder containing reference images or pre-captured spikes used to initialize the model.
--reference_places	int	Number of reference places (images) to load into memory for place recognition.
--dims	list[int, int]	Size to resize input images to before feeding into the network.
--roi_dim	list[int, int]	Size of the input images for the network.
--timebin	int	Spike integration window in milliseconds (e.g., 1000 for 1 second). Defines the time frame for collecting input before inference.