* fix expand data aug

* fix expand data aug

* Enable ParallelExecutor in SSD-MobileNet and Refine code.

* Change the pre-trained model URL.

* Follow comments.

Avoid predicting <pad> by restricting the size of  fc_layer in Transformer

* fix code style problems

* fix bug when loading fluid model

* fix code style problem in brain.py

"add inference"

* Refine SE-ResNeXt model and use ParallelExecutor.

* Fix code style.

* Follow comments.

Decouple the program desc with batch_size in Transformer.

* ready to coco_reader

* complete coco_reader.py & coco_train.py

* complete coco reader

* rename file

* use argparse instead of explicit assignment

* fix

* fix reader bug for some gray image in coco data

* ready to train coco

* fix bug in test()

* fix bug in test()

* change coco dataset to coco2017 dataset

* change dataset from coco to coco2017

* change learning rate

* fix bug in gt label (category id 2 label)

* fix bug in background label

* save model when train finished

Add avg_cost and ppl for Transformer

Refine the inference to output special tokens optionally in Transformer

Adapt usage of reduce_mean to the latest fluid API.

Fix data feed in test & profile