global is_need_init_find_k_FeatureHash

global falconn_query, mxnet_feature_extractor

global my_feature, my_class_name, my_file_path

def getFeature(image_filepath):

img = cv2.imread(image_filepath)

if img is not None:

image = imutils.opencv2matplotlib(img)

image = sess.run(image_preprocessing, feed_dict={'img:0': image})

result_feature = sess.run("resnet_v1_50/pool5:0", feed_dict={'input:0': [image]})

return np.ravel(result_feature)

else:

return None

if is_need_init_find_k_FeatureHash:

print '正在进行初始化...'

def init_falconn():

dim = 2048

# 获得数组

my_feature = np.load(os.path.join(model_path, 'tensorflow-feature.npy'))

print my_feature.shape

my_class_name = np.load(os.path.join(model_path, 'tensorflow-class_name.npy'))

print my_class_name.shape

my_file_path = np.load(os.path.join(model_path, 'tensorflow-file_path.npy'))

print my_file_path.shape

# 获取数组数量

trainNum=len(my_feature)

# 获得默认参数

p=falconn.get_default_parameters(trainNum, dim)

t=falconn.LSHIndex(p)

dataset = my_feature

# 生成hash

t.setup(dataset)

q = t.construct_query_pool()

return my_feature, my_class_name, my_file_path, q

my_feature, my_class_name, my_file_path, falconn_query = init_falconn()

is_need_init_find_k_FeatureHash = False

print '初始化结束...'

featrue = getFeature(image_file_path)

if featrue is not None:

find_list = falconn_query.find_k_nearest_neighbors(query=featrue, k=k)

class_name_list = my_class_name[find_list]

file_path_list = my_file_path[find_list]

return class_name_list, file_path_list

else:

if isinstance(k, int):

k = [k]

right_sum = np.zeros(len(k))

all_sum = np.zeros(len(k))

if isinstance(base_path, str):

base_path = [base_path]

def test_featureHash(base_path, model_path, k, image_format):

class_name_and_path_list = [[floder, os.path.join(base_path, floder)] for floder in os.listdir(base_path) if os.path.isdir(os.path.join(base_path, floder))]

all_bad_list = []

for t_idx, tk in enumerate(k):

num = 0

now_num = 0

bad_list = []

random.shuffle(class_name_and_path_list)

tic = 0

my_sum = 0

for class_name_and_path in class_name_and_path_list:

image_path_list = [os.path.join(class_name_and_path[1], image_file) for image_file in os.listdir(class_name_and_path[1]) if image_file.endswith(image_format)]

my_sum = my_sum + len(image_path_list)

print '一共要检测{}个图片'.format(my_sum)

all_sum[t_idx] += int(my_sum)

for class_name_and_path in class_name_and_path_list:

image_path_list = [os.path.join(class_name_and_path[1], image_file) for image_file in os.listdir(class_name_and_path[1]) if image_file.endswith(image_format)]

class_name = class_name_and_path[0]

for image_path in image_path_list:

c, f = find_k_FeatureHash(model_path, image_file_path=image_path, k=tk)

if c is not None:

is_find = False

now_num = now_num + 1

for tc in c:

#print tc, class_name

if str(tc) == str(class_name):

num = num + 1

right_sum[t_idx] += 1

is_find = True

break

if is_find is False:

bad_list.append(image_path)

tic += 1

if tic % 100 == 0:

print 'Rank=%d时的正确率为%.02f(%d/%d)' % (tk, float(num) / now_num, num, my_sum)

all_bad_list.append(bad_list)

print 'Rank=%d时的正确率为%.02f(%d/%d)' % (tk, float(num) / now_num, num, my_sum)

return all_bad_list

for idx, b_p in enumerate(base_path):

print '开始进行 {} 检测'.format(b_p)

all_bad_list = test_featureHash(b_p, model_path, k, image_format)

np.save('./all_bad_list_{}.npy'.format(idx), all_bad_list)

for idx in range(right_sum):

print '用法: -f [图片路径] [选项]... [选项]...'

print ''

print '将文件夹内的指定格式的图片文件进行FeatureHash测试'

print ''

print '必选参数.'

print '-f 指定图片所在文件夹路径，必须使用tidy_image进行格式处理'

print '-m 指定三个npy文件所在路径'

print ''

print '可选参数.'

print '-r 指定Rank 默认为[1, 5, 10]'

print '-t 指定图片格式, 默认为.webp'