pub struct CountMask { /* private fields */ }Expand description
Count-mask image (GEPT, Sec. 3.2 “Event Accumulation”, Eq. 2): a three-channel RGB encoding
where red and blue hold the per-pixel positive and negative event counts and green is a binary
activity mask — full scale wherever an event of either polarity landed. Timestamps are not
used at all, which is what separates it from Tencode (latest polarity plus
event age) and from EventCount (one plane of raw totals).
Both count planes are clipped and rescaled by a single alpha: the pct-th percentile of
the non-zero counts of the two planes pooled together. A per-channel percentile would let a
quiet polarity saturate against a busy one, so the joint scale is what keeps red and blue
comparable.
white_frame inverts the image to a white background. The black-background default is the form
downstream descriptor models are trained on; the inverted one is offered for parity with the
reference renderer, not for feeding a model.