from single_stage_mr.models.feature_extracter import PoolVgg
from single_stage_mr.utils.vgg_features import get_features
import sys
import logging
import os
import pyicc
import json
logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s',
level=logging.INFO)
if __name__ == "__main__":
path = sys.argv[1]
os.environ["CUDA_VISIBLE_DEVICES"] = sys.argv[2]
logging.info("get data set")
inter_set, intra_set = get_data_set(path=path, resize_length=80)
# vgg特征提取器
vgg_feature_selector = PoolVgg().cuda()
inter_feat_df1, inter_feat_df2, intra_feat_df1, intra_feat_df2 = get_features(
vgg_model=vgg_feature_selector,
inter_set=inter_set,
intra_set=intra_set,
batchsize=16)
print(inter_feat_df1.shape)
print(inter_feat_df2.shape)
feat_ind1 = pyicc._icc([inter_feat_df1, inter_feat_df2], "icc3", 0.75)
print(len(feat_ind1))
feat_ind2 = pyicc._icc([intra_feat_df1, intra_feat_df2], "icc3", 0.75)
print(len(feat_ind2))
inds = list(set(feat_ind1).intersection(set(feat_ind2)))
print(len(inds))
FP = neg_num - TN
FN = pos_num - TP
acc = (TP + TN) / (FP + FN + TP + TN)
# ppv = TP / (TP + FP + 0.001)
# npv = TN / (TN + FN + 0.001)
# logging.info("cutoff:%f" % cutoff)
# logging.info("tp:%d tn:%d fp:%d fn:%d" % (TP, TN, FP, FN))
# logging.info('SEN:%f SPE:%f PPV:%f NPV:%f Acc:%f' % (sen, spe, ppv, npv, acc))
return acc, TP, TN, FP, FN
if __name__ == "__main__":
path = sys.argv[1]
os.environ["CUDA_VISIBLE_DEVICES"] = sys.argv[2]
logging.info("get data set")
train_neg_set, train_pos_set, test_pos_set, test_neg_set = get_data_set(
path=path, resize_length=80)
f = open(sys.argv[3])
d = json.load(f)
icc_inds = d["feature_index"]
f.close()
# 分数据集
total_train_set = train_neg_set + train_pos_set
total_test_set = test_neg_set + test_pos_set
# 负样本的数量
length = len(train_neg_set)
logging.info("get data set completed")
logging.info("get feature and cv")
# vgg特征提取器
vgg_feature_selector = PoolVgg().cuda()
# 混合mr
get_vgg_feature(vgg_model=vgg_feature_selector,