Improving performance

[rsanchezgarc]

Nowadays, cryolo is, by default, picking each micrograph as a job, which is highly inefficient, as the deep model needs to be loaded to gpu in each gpu. One easy fixup will be to increase the default batch size (streaming tab) to a bigger number (eg. 32 for streaming and equal to the number of micrographs in static mode), otherwise, we are wasting gpu.

[delarosatrevin]

I agree with the idea of increasing the default batch size, but maybe a bit small (e.g 10), so that one can see how it is picking quickly, without waiting too mucho.

[rsanchezgarc]

Then 16, powers of 2 are always better when dealing with this kind of stuff

[pconesa]

this is done, but not released yet with 16 as default value.