Anyone have an explanation on how an offset map works?

I guess that I have a pretty good idea how it works - at least it does exactly what I want it to do. So what I think happens is this (and I will explain this in pixels even though we learned that there are no pixels in FF):

FF loops through all the pixels in the image. At each pixel it looks at the greyscale value of the vertical and horizontal offset map(s). Now comes the important thing: it does not actually offset the pixel of the source map at that position, but it pulls another pixel to that position and the location of that other pixel depends on the greyscale value combination of the maps. And yes, you are right - if the offset map is <128 grey the pixels will be pulled from left/top, if it's >128 they will be pulled from right/bottom. A grey of 128 will cause no offset at all.

The formula is probably like this:

Where currentX/currentY are the inner loop

getPixelFromX = currentX + ( offsetMapHGreyAtCurrentXY / 128 ) * pixelSize * percent / 100;

getPixelFromY = currentY + ( offsetMapVGreyAtCurrentXY / 128 ) * pixelSize * percent / 100;

So for example to flip an image horizontally you use a horizontal gradient that runs from left 255 -> 0 right.

As you maybe noticed the offset component is currently my favorite because it is so powerful and flexible.

Quote
byRo wrote: the FF folks seem to have implemented this fully.

Yep -- that's because everything is calculated in floating-point

Quote
Quasimondo wrote: Of course it would also offer a whole range of unintended behaviours...

This is exactly why we dropped the negative values. Plus they were hard to visualize in component previews. However, the possibility of having values greater than 1 ("a partial HDR pipeline" as we call it internally) is still being discussed.