* add cascade mask

* add cascade rcnn

* update code

* remove comments

* refine resume process and modify code to support eval while batch_size is larger than 1

* modify code according to review

* split yolov3 loss using paddle op

* add sync_bn and modify list to LayerList to use sync_bn, add missing op, support cutmix op in dataset, modify reader to use new ops

* modify code according to review

* modify code to run eval.py successfully

* modify code to run eval and train successfully

* modify code to use mixup

* rebase code on lastest dygraph

* modify code to run in sync_bn mode

* modify code according to review

* modify target size of ResizeOp

* update reader in dygraph

* update dataloader some detail

* fix num_classes with background

* split config into models, optimizers, dataset, readers，runtime

* modify code to be suited for launch in 2.0rc

* modify code according to review

* migrate detection ops from Paddle to PaddleDetection

* add unittest for detection ops

* update api & unittest

* add copyright & api name to migrate

* refine PaddleDetection benchmard print, test=develop

* refine PaddleDetection benchmard print, test=develop

* refine PaddleDetection benchmard print, test=develop

* refine PaddleDetection benchmard print, test=develop

* refactor dygraph & add mask rcnn fpn

* fix initialization & minor update

* refine architecture of mask_rcnn

add train&eval 
add reader padding 
add new config 

clear directories & files for dygraph branch 

Add image for visualdl board.

* Update docs
* Update README.md

* optimize python path

* add yolov4

* update check config

* fix slim/xx

* Revert back to `fluid.clip` for gradient clipping

current API is in flux, wait for it to stabilize

* Please CI

still buggy

* add devices_num on reader

* refine capacity value

* refine bufsize

* 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).

* split yolov3 loss

* 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.

* update code & doc

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

* Add more documentation on mixed precision training