tf.app.flags.DEFINE_integer('class_num', 230, '')
tf.app.flags.DEFINE_string('test_data_path', '/path/to/your/testing images/',
'')
tf.app.flags.DEFINE_string('test_gt_path',
'/path/to/your/testing annotations/', '')
tf.app.flags.DEFINE_string('gpu_list', '0', '')
tf.app.flags.DEFINE_string('checkpoint_path', '/path/to/your/checkpoints/', '')
tf.app.flags.DEFINE_string('output_dir', 'outputs/', '')
tf.app.flags.DEFINE_bool('no_write_images', True, 'do not write images')
# tf.app.flags.DEFINE_bool('use_vacab', True, 'strong, normal or weak')
from module import Backbone_branch, RoI_rotate, classification_branch
from data_provider.data_utils import restore_rectangle
FLAGS = tf.app.flags.FLAGS
detect_part = Backbone_branch.Backbone(is_training=False)
roi_rotate_part = RoI_rotate.RoIRotate()
recognize_part = classification_branch.Recognition(is_training=False)
font = cv2.FONT_HERSHEY_SIMPLEX
def iou_cal(predict_box, gt_box):
union = np.bitwise_or(predict_box, gt_box)
inter = np.bitwise_and(predict_box, gt_box)
# union = cv2.cvtColor(union, cv2.COLOR_GRAY2BGR)
# inter = cv2.cvtColor(inter, cv2.COLOR_GRAY2BGR)
contours_union, hierarchy_union = cv2.findContours(union,
cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_NONE)
contours_inter, hierarchy_inter = cv2.findContours(inter,
cv2.RETR_EXTERNAL,
tf.app.flags.DEFINE_string('checkpoint_path', 'checkpoints/', '')
tf.app.flags.DEFINE_boolean('restore', False,
'whether to resotre from checkpoint')
tf.app.flags.DEFINE_integer('save_checkpoint_steps', 1000, '')
tf.app.flags.DEFINE_integer('save_summary_steps', 100, '')
tf.app.flags.DEFINE_string('pretrained_model_path', None, '')
import icdar
# import synth
from module import Backbone_branch, Recognition_branch, RoI_rotate
FLAGS = tf.app.flags.FLAGS
# gpus = list(range(len(FLAGS.gpu_list.split(','))))
detect_part = Backbone_branch.Backbone(is_training=True)
roi_rotate_part = RoI_rotate.RoIRotate()
recognize_part = Recognition_branch.Recognition(is_training=True)
def build_graph(input_images, input_transform_matrix, input_box_masks,
input_box_widths, input_seq_len):
shared_feature, f_score, f_geometry = detect_part.model(input_images)
pad_rois = roi_rotate_part.roi_rotate_tensor_pad(shared_feature,
input_transform_matrix,
input_box_masks,
input_box_widths)
recognition_logits = recognize_part.build_graph(pad_rois, input_box_widths)
# _, dense_decode = recognize_part.decode(recognition_logits, input_box_widths)
return f_score, f_geometry, recognition_logits