* support different norm_type for different module when convert to sync batch norm

* remove BatchNorm in esnet

* refine sync bn

* fix bugs of batch_norm

* fix bugs while deploying and modify BatchNorm to BatchNorm2D

* param_attr -> weight_attr in BatchNorm2D

* modify BatchNorm to BatchNorm2D

* add multi-scale test

* add resnet34-fpn-mstest item in model zoo

* unstructured prune for picodet

* add mcfairmot train eval export code

* fix eval and export

* update modelzoo

* add hrnet18_dlafpn

* unify jdetracker

* fix deploy infer for single class

* fix multi class fairmot deploy

* fix mcmot data source and deploy

* fix num_identities

* fix num_identities_dict

* update PicoDet Architecture and config

* update picodet inference deploy

* [dev] add TOOD

* fix link and notes

* replace ConvNormLayer

* 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 --save_prediction_only support for TopDown KeyPoint Metric

* add a use case for save_prediction_only

* add python train time eval

* add mpii infer support

* add flops

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

* add deepsort export model, reid infer

* fix jde_yolov3 metric

* fix deepsort deploy

* fix doc, clean code

* add mot deploy infer image, add run_benchmark

add dark deploy supported, mpii deploy supported;

* support --eval for s2anet

* add spine dataset and config yml for s2anet

* add doc for s2anet

* update doc

* fix typo, test=document_fix

* add some comments, and update import

* add mot score to txt results

* fix mot install typo

* add epoch for deepsort

* fix paddle no_grad

* deploy infer save image txt

* remove mot metric

* lazy import motmetrics

* jde fairmot export and video python infer

* bs>1 for YOLO

* update slim models and benchmark