ideal number of regularisation images? double steps?

Hi

When enabling regularisation, a flat * 2 modifier is added to the step count, so we will run twice as many steps.
Can someone help me understand why the steps must be doubled, does this assume that we will have as many regularisation images as training images?

Does that mean ideal number of regularisation images is the same number as you have training images?
The code will only accept up to this many and no more.

If we run less regularisation images than training images, we still get double the steps, causing us to have more steps per training image than we configured with the folder name.
Is this advisable?

Thank you for the hard work.

Start by reviewing the regularisation step-count logic and the folder-name image limits described in the issue. Compare training-image and regularisation-image counts to determine whether doubling is intentional, what ratio is expected, and how fewer regularisation images should affect the configured steps.

Written by the indexing model from the issue text.