I need to have a transformation that gets an image, and resizes/crops it, at the same time applies the appropriate transformation to some bounding boxes, this is basic enough so I thought I should find a standard/best-practice way.
In reading about the current state of things, at first I thought torchvision is the right place, but it seems there is also TNT. Further reading, many of TNT’s classes already have much better implementations in pytorch proper. So what is exactly the specific purpose of TNT? should we base on TNT, or torchvision?