Support large images

[t-rad679]

Large images make tensorflow run OOM.

The biggest thing on the internet these days is to break the image down into smaller chunks, run the algorithm on those chunks, then glomp em back together. We can try this.

If anybody wants to look into reconfiguring tensorflow so that this doesn't happen, perhaps through changing memory settings or something about "streaming", be my guest. Also you can try just getting a really expensive GPU if you really want

It might be nice to have both of these options available, since, if the latter succeeds, some machines might not be able to run it