How is it for image recognition/classification? OCR can be a huge chunk of the image classification pipeline. Presumably, it works just as well in this domain?