First off, the "Interactive Facial Recognition Overlays" example in https://plot.ly/python/images/#interactive-facial-recognition-overlays is awesome. It would be even more awesome if the background image (of the people) were in data coordinates (rather than paper coordinates) so that the image and rectangles stayed in sync during zoom/pan.

We should be able to use the same approach as in the previous example: https://plot.ly/python/images/#zoom-on-static-images