* 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

Besides removing unnecessary imports, this also fixes the missing logger module in test_architectures.py within the static directory.

* bs>1 for YOLO

* add script of dota_2_coco

* add args for dota_to_coco

* add script to generate test result of dota

* add loss weight for s2anet

* fix dy2st for s2anet

* fix code style

* Add S2ANet model for Oriented Object Detection.