* Support yolov3 dy2st train

* Fix device_num

* Polish apply_to_static

* Modify config txt to support dy2st tipc

* pose3d train

* delete extra comments

* only pose3d parallel eval; delete pose3d config

* fix mistake edit

* add flag skipping postprocess to support edgeboard hardware

* add flag skipping postprocess to support edgeboard hardware

* add flag skipping postprocess to support edgeboard hardware

* add comment for the flag export_eb

* update picodet benchmark

* support rcnn onnx

* clean code

* update cascade rcnn

* add todo for rpn proposals

* support nms options when export model

* fix config

* add bytetrack yolov3 engine and deploy baseline

* fix bytetrack yolov3 configs

* fix cfgs tools

* fix tracker pred_dets

* fit pptracking jde sde infer

* rename cfgs, fix mot_sde_infer

* fix mot sde single class infer

* remove byte cfgs

* fix mtmct

* clean code

* add a new version code of PicoDet

* support export post_process in PicoDet

* add centernet

* add centernet

* add r50 for centernet

* support deploy for centernet

* add doc

* update pretrain

* clean code

* update r50 doc

* unify for_mot

* update code according to review

* update doc

* refine preprocess

* update doc

* update picodet inference deploy

* add keypoint top-down cpp infer; total:5ms,pre:0.7ms,infer:3.8ms,post:0.01ms

* add dark_pose support

Revert "add keypoint top-down cpp infer; total:5ms,pre:0.7ms,infer:3.8ms,post:0.01ms (#4058)" (#4216)

This reverts commit 08619b4e.

* set lr for 4 card as default, and update

* add gfl model and PicoDet

* mot video decode to images

* add infer_dir for mot

* fix export and unite pose infer

* remove task in mot source

* set image_dir

* update doc

* fix tracker trt, add tracker to infer_cfg yml

* fix doc and deploy tracker

* jde fairmot export and video python infer

* Automatically set use_dynamic_shape

* fix image_shape in export_model

* fix infer bug of s2anet

* add inference benchmark

* Add BottomUp HigherHRNet for pose.
* reverse xtcocotools to pycocotools

* Add S2ANet model for Oriented Object Detection.