Table 2.

Characterization of event-based encoding for each of the generated datasets at different threshold values.

	Before time binning			After time binning
Threshold (ϑ)	$\overline{\mathrm{\text{Events}}}$	$\overline{\mathrm{\text{Comp. ratio}}}$ (γ)	$\overline{\mathrm{\text{MSE}}}$ (ε)	Bin size (ms)	$\overline{\mathrm{\text{Events}}}$	$\overline{\mathrm{\text{Comp. ratio}}}$ (γ)	$\overline{\mathrm{\text{MSE}}}$ (ε)
1	87.6	1	0	5	58.1	1.5	39.5
2	38.0	2.3	0.4	3	35.5	2.5	12.5
5	10.5	8.3	3.7	3	10.5	8.3	4.1
10	3.4	25.7	12.3	5	3.4	25.7	12.3

Compression ratio γ is defined as the number of events at perfect encoding (ϑ = 1) divided by the number of events at each higher threshold value. Values of the reconstruction error ε are calculated per reconstructed frame by Mean Squared Error (MSE). Mean events are calculated per sample. All time_bin_size, introduced to prepare the data for clock-driven computation, follow the results reported in Section 3.4 from the two-step HPO procedure.