* add small scale config

* add model zoo

* remove metric type

* 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.

* Update PicoDet adn GFl

* add faster rcnn with swin_transformer tiny

* fix use_gpu in deploy doc,  test=document_fix

* fix deploy doc, test=document_fix

* updata document, test=document_fix

* use 8 card for training as default, test=document_fix

* add LiteHRNet backbone and config .YML

* test lite18-network param

acc is same with ori-model 1. fix default darkpose=ON, 2. += is not inplace

add new keypoint model Lite-HRNet

* add new keypoint model Lite-HRNet

* 1. Add description of network type; 2. use channel_shuffle in ops.py

* use normal to init conv2d

* add network type description

* add frame_rate for mot infer video

* add doc of frame_rate for mot video infer, test=document_fix

* add --save_prediction_only support for TopDown KeyPoint Metric

* add a use case for save_prediction_only

* set lr for 4 card as default, and update

* s2anet update

* remove np anchor

* add gfl model and PicoDet

* add kitti metric

* clean kitti metric code

* fix kitticars doc

* remove kitti headtracking cfgs

* remove kitti headtracking cfgs, clean code

* fix datatype and ffmpeg, test=document_fix

* add soft link, test=document_fix

* add doc link, test=document_fix

* fix deploy image_list, test=document_fix

* clean code of s2anet

* fix for paddle==2.1.0

* remove debug info

* update comment

* fix train diff for fairmot

* set use_shared_memory to True for fairmot training

* add the config of fairmot_headtracking