* refactor s2anet to support batch_size > 1

* fix problem of inference

* support batch_size > 1 for training

* fix empty results

* fix dota eval

* fix configs of s2anet_head

* modify s2anet_spine_1x to 73 mAP

* support no label training

* fix load_crowd

* update to fit rbox

* add empty_ratio

* update configs

* support mask rcnn

* clean drop_empty for mot & keypoint

* refine is_crowd

* Add S2ANet model for Oriented Object Detection.

* refine trainer

* refactor rcnn, test=dygraph

* add mask_rcnn, test=dygraph

* add Faster RCNN & Faster FPN, test=dygraph

* update according to the review, test=dygraph

Merge dygraph branch into master branch.
Co-authored-by: NFDInSky <48318485+FDInSky@users.noreply.github.com>
Co-authored-by: Nsunxl1988 <47514455+sunxl1988@users.noreply.github.com>
Co-authored-by: Nwangguanzhong <jerrywgz@126.com>
Co-authored-by: Nwangxinxin08 <69842442+wangxinxin08@users.noreply.github.com>
Co-authored-by: Nwanghuancoder <wanghuancoder@163.com>
Co-authored-by: Ncnn <liuhui29@baidu.com>
Co-authored-by: NGuanghua Yu <742925032@qq.com>
Co-authored-by: NKaipeng Deng <dengkaipeng@baidu.com>
Co-authored-by: Nroot <root@bjyz-sys-gpu-kongming0.bjyz.baidu.com>

add htc model 

* filter illegal data

* update warning message

* add yolov4

* Make logger less noisy

* Fix indent

* fix all documentation

* fix docs conf.py

* fix some error

* fix readthedocs

* refine cocoapi import

* fix infer when anno_path is None

* 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.