Handling image classification for OOD images

So, my dataset has images which look like this:

And I need to sleeve length classification.

But I am observing that in production, there are a few images which look like this:

Some of the poses are a little more nuanced.

Two questions:

  1. Is there a way to detect that a particular image in the production/test set is out of distribution?
  2. Since such images would form a small fraction of the production set, is there a way to do sleeve classification using traditional CV. That will save me the effort of gathering such images and re-training.

Please advise!