Did you finally select it? Our team is using PaddleOCR with good results! We wrote an article (https://medium.com/adevinta-tech-blog/text-in-image-2-0-impr...) about it.