* update strategy

2. fix pre-set net spelling , pre-set net -> pre-defined net

3. remove version_compare usage

* add object detection dataset, reader, task and demo

* accelerate_mode default True

* surport ernie module v2

close the predictor log

* strategy support layer_params

* use dataloader

* add-inference-model

* optimize hook

* fix typo

* enhance predict dataset

* fix python2 bug

* Update README.md

Polish README
Co-authored-by: NZeyu Chen <chenzeyu01@baidu.com>
Co-authored-by: NSteffy-zxf <48793257+Steffy-zxf@users.noreply.github.com>

* optimize demo and cv reader

* remove needless import
Co-authored-by: Nwuzewu <wuzewu@baidu.com>

* finish coding predict

* optimize predict pr

* optimize _postprocessing

* optim namedtuple

* reconsitution reader and dataset

* add task hook

* colorlog

* optimize __repr__

* fix unpad

* optimize nltk