Standardise the use of axes in cellfinder

[adamltyson]

Currently most of cellfinder uses height/width, but this is often in the wrong order for numpy axis ordering. The third dimension (z) is also often last (rather than first).

It would be helpful long term to standardise everything, e.g. around numpy axis ordering.

Ref #411

[matham]

In my pytorch branch, which I'm getting ready to PR, I had to go through all the detection parts to get it working with torch. And I removed all the flips or any assumptions about the order of the last two axis and it worked fine. So at least for the detection code it may be fixed in that PR.

I also started referring to these last two dimensions as dim1/dim2 for each plane, as it doesn't make any difference for the order as long as there's no flipping in the middle. But I can refer to them as height/width (in that order) if that's preferred.

[adamltyson]

In my pytorch branch, which I'm getting ready to PR, I had to go through all the detection parts to get it working with torch. And I removed all the flips or any assumptions about the order of the last two axis and it worked fine. So at least for the detection code it may be fixed in that PR.

Fantastic!

I also started referring to these last two dimensions as dim1/dim2 for each plane, as it doesn't make any difference for the order as long as there's no flipping in the middle. But I can refer to them as height/width (in that order) if that's preferred.

If it's not too much trouble, could you call them axis 0, axis 1 etc. to match numpy?