def test2():
val_embed = read_npy(CIFARPROCESSED+'val_image.npy')
val_label = read_npy(CIFARPROCESSED+'val_label.npy')
cifar = NpairDatamanager(val_embed, val_label, CIFARNCLASS, nsclass=4)
_, _, anc_l, pos_l = cifar.next_batch(32)
print(anc_l)
print(pos_l)
def test1():
val_embed = read_npy(CIFARPROCESSED+'val_image.npy')
val_label = read_npy(CIFARPROCESSED+'val_label.npy')
cifar = TripletDatamanager(val_embed, val_label, CIFARNCLASS, nsclass=10)
count = np.zeros(cifar.nclass)
nbatch = cifar.ndata//50+1
for i in range(nbatch):
_, label = cifar.next_batch(50)
for index in range(len(label)):
count[label[index]]+=1
print(count)
def cifar_manager(dm_type='basic', nsclass=0):
'''
Args:
dm_type - string
defaults to be basic
nsclass - int
which required when dm_type='triplet'
Return:
dm_train, dm_val, dm_test
datamanager for each set
'''
assert dm_type in DATAMANAGER_DICT.keys(), "The type of data should be in {}".format(DATAMANAGER_DICT.keys())
dm = DATAMANAGER_DICT[dm_type]
train_input = read_npy(CIFARPROCESSED+'train_image.npy')
train_label = read_npy(CIFARPROCESSED+'train_label.npy')
val_input = read_npy(CIFARPROCESSED+'val_image.npy')
val_label = read_npy(CIFARPROCESSED+'val_label.npy')
test_input = read_npy(CIFARPROCESSED+'test_image.npy')
test_label = read_npy(CIFARPROCESSED+'test_label.npy')
if dm_type in ['triplet', 'npair']:
dm_train = dm(train_input, train_label, CIFARNCLASS, nsclass)
dm_val = dm(val_input, val_label, CIFARNCLASS, nsclass)
dm_test = dm(test_input, test_label, CIFARNCLASS, nsclass)
else:
dm_train = dm(train_input, train_label, CIFARNCLASS)
dm_val = dm(val_input, val_label, CIFARNCLASS)
dm_test = dm(test_input, test_label, CIFARNCLASS)
return dm_train, dm_val, dm_test
def imagenet32_manager(dm_type='basic', nsclass=0):
'''
Args:
dm_type - string
nsclass -int
'''
train_img = read_npy(IMAGENET32PROCESSED+'train_img.npy')
train_label = read_npy(IMAGENET32PROCESSED+'train_label.npy')
val_img = read_npy(IMAGENET32PROCESSED+'val_img.npy')
val_label = read_npy(IMAGENET32PROCESSED+'val_label.npy')
test_img = read_npy(IMAGENET32PROCESSED+'test_img.npy')
test_label = read_npy(IMAGENET32PROCESSED+'test_label.npy')
dm = DATAMANAGER_DICT[dm_type]
if dm_type in ['triplet', 'npair']:
dm_train = dm(train_img, train_label, IMAGENETCLASS, nsclass)
dm_val = dm(val_img, val_label, IMAGENETCLASS, nsclass)
dm_test = dm(test_img, test_label, IMAGENETCLASS, nsclass)
else:
dm_train = dm(train_img, train_label, IMAGENETCLASS)
dm_val = dm(val_img, val_label, IMAGENETCLASS)
dm_test = dm(test_img, test_label, IMAGENETCLASS)
return dm_train, dm_val, dm_test
'npair_d' : NpairDatamanagerDouble
}
if __name__=='__main__':
import sys
sys.path.append('../configs')
sys.path.append('../utils')
sys.path.append('../tfops')
# utils
from reader import read_npy
# config
from path import CIFARPROCESSED
from info import CIFARNCLASS
val_embed = read_npy(CIFARPROCESSED+'val_image.npy')
val_label = read_npy(CIFARPROCESSED+'val_label.npy')
cifar = TripletDatamanager(val_embed, val_label, CIFARNCLASS, nsclass=10)
count = np.zeros(cifar.nclass)
nbatch = cifar.ndata//50+1
for i in range(nbatch):
_, label = cifar.next_batch(50)
for index in range(len(label)):
count[label[index]]+=1
print(count)
cifar = NpairDatamanager(val_embed, val_label, CIFARNCLASS, nsclass=4)
_, _, anc_l, pos_l = cifar.next_batch(32)
print(anc_l)