I would imagine the software would need to project series of images onto a cylinder, align them, and then recognize and extract the label.
Feels like I will need to write the solution myself, but figured I will ask before I start tackling the problem.
if the images are uniform enough in style you can then probably crop them all at once with mask coordinates.
opencv/imagemagick will handle all of it.
if the images aren't uniform enough to crop them all at once, consider the use of a 'lazy susan' style vertical rotisserie, take all the photos from the same plane, easy post-processing.