Fix ce threshold in Transformer

* 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.

Port text_classification model and object_detection to Python3

Fix ce of ocr.

* update ce

* refine images format
* Update README_cn.md

Fix ce of icnet.

disable kpi for models

fix ce of icnet

* add crop and reduction layer
* caffe2fluid: support crop and reduction layer convertion

Create a script to help with the documentation deployment

* refine eval and update pyramidbox parameter init

* refine doc

Speed up NTM reader

* Add MKL-DNN Benchmarking to CRNN-CTC

* Add MKL-DNN benchmarking for chinese_ner

* Make crnn-ctc scripts more portable

* Merge CRNN-CTC train & inference scripts

* Remove LD_LIBRARY_PATH from crnn-ctc scripts

* CRNN-CTC scripts: set parallel to true

Abort script if batch_size is lower than num of cores

* CRNN-CTC scripts: limit mode options in infer

* CRNN-CTC scripts: set mkldnn parallel to False

* CRNN-CTC scripts: remove mkldnn parallel warning

* Chinese-ner: Merge train & infer scripts, update readme

* Chinese_ner: add --parallel flag for train

fix bug after hide get_places and get_dims

* Fix ce of ocr.

* Remove debug print

Comment out the deploy script for now.