Table 2.

DECOLLE Neural network used for the DvsGesture dataset.

Layer type	#	Data type	Dimensions
DVS	2	AEDAT 3.1	128 × 128
Downsample (Sum)	2	Integer	32 × 32
7 × 7 Conv	64	Float	30 × 30
2 × 2 MaxPool	64	Float	15 × 15
Spiking Non-linearity		Binary
Dropout (p = 0.5)		Float
Dense	11	Float	11
7 × 7 Conv	128	Float	13 × 13
Spiking Non-linearity		Binary
Dropout (p = 0.5)		Binary
Dense	11	Float	11
7 × 7 Conv	128	Float	11 × 11
2 × 2 MaxPool	128	Float	5 × 5
Spiking Non-linearity		Binary
Dropout (p = 0.5)		Binary
Dense	11	Float	11

Note that dense layers are used for the local classifiers only and were not fed to the subsequent convolutional layers. AEDAT 3.1 is a data format used for event-based data. The spiking nonlinearity was always applied after the pooling layers. Dropout layers were left active during testing.