* Add dc_gan and fix readme.

* Fix readme style.

* Fix readme.

* Fix readme.

* Add conditionalGAN

* Fix readme.

* Fix image name in readme.

* Add DCGAN.

* Rename conditional_gan to c_gan.

* Move c_gan and cycle_gan to gan/

* Add trainer.py and utility.py

* Fix data reader

* Support variable image shape.
* Update padding mode.

* change epoch to iter
* add smooth loss for logging

* add pyreader

* add Deeplabv3+ model

return im_id in reader for evaluation 
add coco evaluation for Faster Rcnn model
visualize detection result when image_path is provided

* add HNAS models for image classification
* change name to HiNAS
* update license for HiNAS_models

* Add cycle gan.

* Fix bugs

* Fix program.

* Update cycleGAN

* Add pyreader, refine code and add profiling code.
* Update parameter initializer.

* Object detection model: faster-rcnn  by @jerrywgz @sefira @qingqing01

* ce fix

* refine cpm and interp_mode

* Accelerate reader in object_detection.

* fix images

The `get_feeder_data()` method needs the `need_label` parameter set to `False` in inference for it to work.

* fix data-anchor-sampling and use depthwise conv

* revert reader