This add the setting to adjust rounding threshold. The 0.5 fraction is rounding down in standard rule, as int(v + 0.5) This allows to change the standard threshold and round as int(v + round_threshold).
The SPI community usually rounds the diffraction patterns to photons. But they used biased threshold, because the number of pixels decreases exponentially as the number of photons increases. This causes rounding errors to be unbalanced. Without this option they refuse to use our rounding and compression.