Support PIL resizse with PIL interpolation to train transformer.
Almost all vision transformer models need using PIL.Image.BICUBIC
as interpolation in resize.

添加slim功能

Add EfficientNet config

fix quick_start_recognition docs

update PaddleServing of clas

add metric_learning.md

add cpp whole_chain test

fix tnt inference bug when bs > 1

add metric_learning.md

fix issues when gallery == query dataset

replace vector_search with faiss