def __send_data(self):
""" a single thread which send a data to the data queue
"""
priori_bboxes = config.priori_bboxes / config.img_size
while True:
r_n = random.uniform(0, 1)
if r_n >= 0.5:
img_name = random.sample(self.__inria_imgs_name, 1)[0]
else:
img_name = random.sample(self.__hazy_imgs_name, 1)[0]
filename = os.path.basename(img_name)
basefile = filename.split(".")[0]
if 'hazy_person' in img_name:
label_name = os.path.join(self.__hazy_label_dir, (basefile+".xml"))
# print(label_name)
elif 'inria_person' in img_name:
label_name = os.path.join(self.__inria_label_dir, (basefile + ".xml"))
# print(label_name)
else:
raise ValueError('not support now...')
img, bboxes = self.__read_one_sample(img_name,label_name)
## resize img and normalize img and bboxes##
if self.__for_what == "train":
if self.__whether_aug:
img, bboxes = train_tools.img_aug(img, bboxes)
if len(bboxes) == 0:
## sometimes aug func will corp no person##
#logger.warning("No person img, abandoned...")
continue
else:
x_ = (bboxes[:, 3] - bboxes[:, 1]) / img.shape[1]
y_ = (bboxes[:, 2] - bboxes[:, 0]) / img.shape[0]
if np.count_nonzero(x_ <= 0.08) > 0 or np.count_nonzero(y_ <= 0.08) > 0:
continue
img, bboxes = train_tools.normalize_data(img, bboxes, config.img_size)
labels, bboxes = \
train_tools.ground_truth_one_img(corner_bboxes=bboxes,
priori_boxes=priori_bboxes,
grid_cell_size=config.grid_cell_size,
surounding_size=config.surounding_size, top_k=config.top_k)
## put data into data queue ##
self.__data_queue.put([img, labels, bboxes])
else:
# raise ValueError('not support now...')
if self.__whether_aug:
img, bboxes = test_tools.img_aug(img, bboxes)
if len(bboxes) == 0:
## sometimes aug func will corp no person##
#logger.warning("No person img, abandoned...")
continue
img, bboxes = train_tools.normalize_data(img, bboxes, config.img_size)
self.__data_queue.put([img, bboxes])
def load_data_eval(self):
"""Traversing the test set in sequence
Return:
img: one img with shape (h, w, c), if end, None
bboxes: shape is (n, 4), if end, None
"""
try:
##get img name
img_name = self.__imgs_name_que.popleft()
##get corresponding anotation name
filename = os.path.basename(img_name)
basefile = filename.split(".")[0]
if 'hazy_person' in img_name:
label_name = os.path.join(self.__hazy_label_dir, (basefile+".xml"))
# print(label_name)
elif 'inria_person' in img_name:
label_name = os.path.join(self.__inria_label_dir, (basefile + ".xml"))
# print(label_name)
else:
raise ValueError('not support now...')
img, bboxes = self.__read_one_sample(img_name, label_name)
img, bboxes = train_tools.normalize_data(img, bboxes, config.img_size)
return img, bboxes, filename
except IndexError:
return None, None, None
def __send_data(self):
""" a single thread which send a data to the data queue
"""
priori_bboxes = config.priori_bboxes / config.img_size
while True:
img_name = random.sample(self.__imgs_name, 1)[0]
filename = os.path.basename(img_name)
basefile = filename.split(".")[0]
label_name = os.path.join(self.__label_dir, (basefile + ".xml"))
img, bboxes = self.__read_one_sample(img_name, label_name)
## resize img and normalize img and bboxes##
if self.__for_what == "train":
if self.__whether_aug:
img, bboxes = train_tools.img_aug(img, bboxes)
if len(bboxes) == 0:
## sometimes aug func will corp no person##
#logger.warning("No person img, abandoned...")
continue
img, bboxes = train_tools.normalize_data(
img, bboxes, config.img_size)
labels, bboxes = \
train_tools.ground_truth_one_img(corner_bboxes=bboxes,
priori_boxes=priori_bboxes,
grid_cell_size=config.grid_cell_size,
surounding_size=config.surounding_size, top_k=config.top_k)
## put data into data queue ##
self.__data_queue.put([img, labels, bboxes])
else:
if self.__whether_aug:
img, bboxes = test_tools.img_aug(img, bboxes)
if len(bboxes) == 0:
## sometimes aug func will corp no person##
#logger.warning("No person img, abandoned...")
continue
img, bboxes = train_tools.normalize_data(
img, bboxes, config.img_size)
self.__data_queue.put([img, bboxes])
def load_data_eval(self):
"""Traversing the test set in sequence
Return:
img: one img with shape (h, w, c), if end, None
bboxes: shape is (n, 4), if end, None
"""
try:
##get img name
img_name = self.__imgs_name_que.popleft()
##get corresponding anotation name
filename = os.path.basename(img_name)
basefile = filename.split(".")[0]
label_name = os.path.join(self.__label_dir, (basefile + '.xml'))
img, bboxes = self.__read_one_sample(img_name, label_name)
img, bboxes = train_tools.normalize_data(img, bboxes,
config.img_size)
return img, bboxes, filename
except IndexError:
return None, None, None