1. data folder
 build_config                    - store the build configuration files
 case_config                     - store the case configuration files
 common.py                       - store common data like voc class_name
 dataset -> /swtest/NPU          - NFS dataset and model zoo folder


2. The dataset folder
It stores dataset and modelzoo which are all from NFS server: dataset -> /swtest/NPU.
About NFS mounting please refer: https://confluence.cixtech.com/display/SW/14+-+stream+server+user+guide
dataset
 coco
 ILSVRC2012
 model_zoo
 tar
 VOC2007
 VOC2012

3. case configure files
They are CSV files like:
Case_config/
 zhouyi_classify_fm.csv
 zhouyi_classify_sm.csv
 zhouyi_obj_sm.csv
 zhouyi_segment_sm.csv
fm - full-test mode. The accuracy, performance, profiling data will be calculated for fm test.
sm - simple-test mode. Just run few inference in board to verfiy model can run normally.
The format example(classification):
The first line:
case_name,model,framework,dataset,desc,build_cfg,cali_num,metric_num,input_num,topk,topk_index,top1,top5,input_way,show
The content exaple:
LBSP-AI-ZY-BuildSIM-Classify-SM-0001,tf_resnet50_v1,tensorflow,ILSVRC2012,Zhouyi Inference for tensorflow resnet50 simple mode,zhouyi/cls/tf_resnet_v1_50_build.cfg,50:1,200:1,50:1,5,0,0.7493,0.9238,caffe,False
Explantion for some fields:
    topk - 1 or 5, means calculate top1 or top5 accuracy
    topk_index - 0 or 1, if background is included in model inference result, it should be set as 1
    top1 - top1 accuracy reference
    top5 - top5 accuracy reference
    input_way - input pre-process method. The available methods include tf, torch, caffe and none
    show - True or False. If True, in accuracy calculation process, the picture after pre-process will be shown


