Image resizing (and cropping?)

[tischi]

I would like to add a new module about image resizing. There is much to learn, e.g. in terms of which interpolation one should use.

I wonder: should one also include image cropping into this "resizing: module? I would say it is conceptually different but at the same time, technically, one obtains an image of a "different size", so it could fit. The main reason I am pondering about including cropping into resizing is that I don't think there is much to teach about cropping; I find it so easy that I would not think it deserves an own module, or am I missing something?

I would love to read your opinions!

ping @stemarcotti @AnniekStok @maulakhan @tibuch @manerotoni @grinic

[tischi]

In fact, unsure now, maybe "image resizing" should be part of an "image similarity transformation" module?!!

[manerotoni]

I can add only from the motivation side: to prepare data for Deep learning (or other algorithms) you do a lot of cropping (=tiling) and stitching. To make the patches of images of an identical dimension you use also quite a bit of padding and resizing

If you use cropping as tiling of a large image, this is not that trivial and you would like to do it in a clever way.

Don't forget resizing of labeled images that also need some different approach.

[tischi]

I see, good points, image cropping, tiling, chunking could be one module then...?

[manerotoni]

Yes, I think it something needed in particular if you have very large data sets. Try to pus 1 TB on your GPU. In a module you will obviously not work with such a large data set