Training Parameters

Each of the following parameters can be passed to the LENS command-line interface to configure the training process.

LENS Training Parameters

Argument	Type	Description
--dataset	str	Name of the dataset to use for training and/or inference.
--camera	str	Type of camera used to capture the data (e.g., davis128).
--data_name	str	Identifier for the dataset instance, used for naming outputs.
--reference	str	Directory name of the reference dataset.
--query	str	Directory name of the query dataset.
--data_dir	str	Path to the base directory containing the data folders.
--reference_places	int	Number of locations (images) to use in the reference set.
--query_places	int	Number of locations (images) to use in the query set.
--feature_multiplier	float	Multiplier for expanding the feature layer size.
--filter	int	Skip every nth image during dataset iteration (used for downsampling).
--epoch_feat	int	Number of training epochs for the feature layer.
--epoch_out	int	Number of training epochs for the output layer.
--thr_l_feat	float	Lower threshold for membrane potential in the feature layer.
--thr_h_feat	float	Upper threshold for membrane potential in the feature layer.
--fire_l_feat	float	Lower bound for firing probability in the feature layer.
--fire_h_feat	float	Upper bound for firing probability in the feature layer.
--ip_rate_feat	float	Intrinsic plasticity learning rate for the feature layer.
--stdp_rate_feat	float	STDP learning rate for the feature layer.
--thr_l_out	float	Lower threshold for membrane potential in the output layer.
--thr_h_out	float	Upper threshold for membrane potential in the output layer.
--fire_l_out	float	Lower bound for firing probability in the output layer.
--fire_h_out	float	Upper bound for firing probability in the output layer.
--ip_rate_out	float	Intrinsic plasticity learning rate for the output layer.
--stdp_rate_out	float	STDP learning rate for the output layer.
--f_exc	float	Probability of excitatory connections in the feature layer.
--f_inh	float	Probability of inhibitory connections in the feature layer.
--o_exc	float	Probability of excitatory connections in the output layer.
--o_inh	float	Probability of inhibitory connections in the output layer.
--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.
--GT_tolerance	int	Ground truth tolerance for considering a match correct.
--train_model	flag	Enables training mode. If not set, the system will run in inference or demo mode.