def detect(self, img):
img, scale = resize_img(img)
print('shape', img.shape)
print('scale', scale)
img = img_normailize(img)
h, w, c = img.shape
img_input = np.reshape(img, [1, h, w, c])
img_info = [h, w, 1]
s = time.time()
scores, pp_boxes = self._get_net_output(img_input, img_info)
print('net:', time.time() - s)
print(scores)
s = time.time()
text_connector = TextConnector()
# 得到是resize图像后的bbox
# print(img_info)
# print('boxes, scores[:, np.newaxis]',boxes.shape, scores[:, np.newaxis].shape,scores.shape)
text_proposals, scores, boxes = text_connector.detect(
pp_boxes, scores[:, np.newaxis], img_info[:2])
print('merge:', time.time() - s)
# 原图像的绝对bbox位置
original_bbox, scores = self._resize_bbox(boxes, scale)
return pp_boxes / scale, original_bbox, scores
def getbatch(self):
img = cv2.imread(os.path.join(self.img_dir, self.img_path_list[self.current_index]))
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
img, _ = resize_img(img)
img = img_normailize(img)
h, w, c = img.shape
# print(img.shape)
img_data = np.reshape(img, [1, h, w, c])
label_name = self.img_path_list[self.current_index].split('.')[0] + '.txt'
assert os.path.exists(os.path.join(self.label_dir, label_name)), \
"{} is not exist".format(os.path.join(self.label_dir, label_name))
labels_data = []
with open(os.path.join(self.label_dir, label_name)) as f:
lines = f.readlines()
for line in lines:
encode_data = line.split('\t')
label = []
label.append(float(encode_data[1]))
label.append(float(encode_data[2]))
label.append(float(encode_data[3]))
label.append(float(encode_data[4]))
label.append(int(self.class_dict[encode_data[0]]))
labels_data.append(label)
if self.current_index + 1 + 1 <= len(self.img_path_list):
self.current_index += 1
elif self.current_index + 1 + 1 > len(self.img_path_list):
self.epoch += 1
self.current_index = 0
random.shuffle(self.img_path_list)
return img_data, np.array(labels_data, np.int), np.array(img.shape).reshape([1, 3])
if basename.lower().split('.')[-1] not in [
'jpg', 'png', 'JPG', 'JPEG', 'jpeg', 'PNG'
]:
print(basename.lower().split('.')[-1])
continue
stem, ext = os.path.splitext(basename)
txt_file = os.path.join(xml_dir, stem + '.txt')
img_path = os.path.join(img_dir, file)
# print(img_path)
img = cv.imread(img_path)
img_size = img.shape
im_size_min = np.min(img_size[0:2])
im_size_max = np.max(img_size[0:2])
re_im, im_scale = resize_img(img)
re_size = re_im.shape
cv.imwrite(os.path.join(out_path, stem) + '.jpg', re_im)
class_list, bbox_list = parse_txt(txt_file)
assert len(class_list) == len(bbox_list), 'bbox和label不对应'
assert len(class_list) > 0, 'xml文件有问题{}'.format(txt_file)
for bbox_index in range(len(bbox_list)):
# if class_list[bbox_index] == 2:
# continue
if len(bbox_list[bbox_index]) == 8: