* Add attention training model for ocr.

* Add beam search for infer.

* Fix data reader.

* Fix inference.

* Prune result of inference.

* Fix README

* update README

* Rsize figure.

* Resize image and fix format.