* add blazeface_nas1

* fix fruit config on master branch

* refine doc

* fix obj365 cascade rcnn config
* fix obj365 cascade rcnn doc

* update YOLOv3_ENHANCEMENT.md
* update IoUloss and DropBlock configs for YOLOv3 related configs

* fix all documentation

* fix docs conf.py

* fix some error

* fix readthedocs

add libra rcnn, including
  * libra loss
  * libra fpn(bfp)
  * libra sampling

* add iou loss to yolov3
* modify the comment and delete redundant yml file
* fix the low speed bug in dropblock module

* add dropblock module
* move dropblock module from yolo_head to ops
* fix obj365 num classes

* add blazeface nas demo

* fix format

* fix configs

* fix reduce_rate

* add gcnet model : https://arxiv.org/abs/1904.11492

* add diou nms.
* fix typo in diou nms and soft nms (lod_leval->lod_level).

* Add GIOU loss.
* Add DIOU/CIOU loss.
* Change loss box type in bbox_head.py.
* Remove fuse_elewise_add_act_ops flag in train.py (Error may happen for elem_divbased on the flag).

- 1.add cn docs
- 2.fix fddb evaluate
- 3.fix get_shrink
- 4.delete FLAGS.dataset_dir

* split yolov3 loss

* add step in prior_box

* refine README

* add res2net model

* modify yml files

* add res2net model zoo

* add hrnet model

* modify model zoo

* minor fix readme

* update readme in dirs

* minor fix softnms bug

* refine model zoo

* modify yml files

* modify pretrain names

* Polish Reader to simplify preprocessing logic.
#  sample_transforms-> make batch -> batch_transforms in Reader.
* Clean some code
* Imporve yolov3_r50vd_dcn_obj365_pretrained_coco 41.4 to 41.8.
* Update all configs.

* Add cascade_rcnn_cls_aware_r200_vd_fpn_dcnv2_nonlocal_softnms model on oidv5 and obj365.
* Add obj365 config annotation
* Update docs.

* update face_eval_data

* add softnms, nonlocal, resnet200_vd_backbone
* add CBNet
* update model zoo

* update code & doc

* change voc_loader

* add CACascadeRCNN doc
* add pretrain model download link

* add face doc
* update face detection readme and add demo jpg

* Error out if fp16 is enabled and affine channel is used

* Add more documentation on mixed precision training

* add multi-scale test