* add dy_ghostnet_ssdlite

* precommit format

* unity weights

* update doc of tensorrt, test=document_fix

* mv tensorrt document, test=docunemt_fix

* update tools/eval, support output_eval/json_eval

* update tools/eval, support output_eval/json_eval

* update tools/eval, support output_eval/json_eval

* update tools/eval

* TensorRT support dynamic shape

* add TensorRT doc

* update model visual tool, test=fix_document

* python support trt6

* cpp support trt6

* fix code error

* clean code

* python trt dynamic shape and fix code style

* disable log and update doc style

* update opencv lib url

* solov2 trt support dynamic shape

* support tensorrt_int8

* add TensorRT int8 calib data for ppyolo

* set workspace_size=1 << 30 as default

* remove content

* remove some content, and fix code style

* remove images

* update doc style

* remove load_static_weights

* add scipy in requirements

* enable shared memory

* add VisualDL and save best_model

* add VisualDLWriter callback

* Update install doc and deploy doc.

* add yolov3 voc configs for dygraph

* simplify yolov3 darknet voc configs

* support num_classes shared from dataset config

* Fix the bug of export quantized inference model

* Fix the bug of export quantized inference model

* add blazeface & update rcnn modelzoo

* remove condition num_priors

* fix doc link, test=dygraph, test=document

* use scripts to build on windows flatform, test=dygraph

* Use new preproces OPs in deploy/infer.py, test=dygraph

* fix bug of image_shape can not set 2-dims, test=dygraph

* fix cpp build, test=dygraph

* set param target_size must be [h, w] in Resize function, test=dygraph

* fix bug, test=dygraph

* fix ssd output format (bbox: [N, 6], bbox_num: bbox_num), test=dygraph

* fix ssd prediction in cpp mode, test=dygraph