fix conflicts

datasets/xml_to_tfrecords.py

@@ -56,7 +56,7 @@ def _process_image(train_img_path, train_xml_path, name):
 
     for obj in root.findall('object'):
         label = obj.find('name').text
-        if label == '###' or label == 'none':
+        if label == 'none':
             label = 'none'
         else:
             label = 'text'

test.py

@@ -744,7 +744,7 @@ def nms_eff(boxes_list, score, overlap):
         out_dir,
         nms_th,
         score_th,
-        scales=[(768, 768)],
+        scales=[(384, 384)],
         save_res_path=save_res_path
         )
     instance.start_inference()

tools/gen_xml.py

@@ -46,6 +46,7 @@ def process_convert(txt_name, DIRECTORY_ANNOTATIONS, new_version=True):
     nodeSize.appendChild(nodeDepth)
     root.appendChild(nodeSize)    
     for l in lines:
+        l = l.encode('utf-8').decode('utf-8-sig')
         l = l.strip().split(',')
         difficult = 0
         label = 1
@@ -60,7 +61,7 @@ def process_convert(txt_name, DIRECTORY_ANNOTATIONS, new_version=True):
 
         if new_version is True:
             label_name = str(l[-1])
-            if label_name == 'none' or label_name == '###':
+            if label_name == 'none':
                 difficult = 1
             else:
                 difficult = 0
@@ -79,6 +80,7 @@ def process_convert(txt_name, DIRECTORY_ANNOTATIONS, new_version=True):
             y3_text = ymax_text
             y4_text = ymax_text
         else:
+            print(l)
             xs = [ int(l[i])  for i in (0, 2, 4, 6)]
             ys = [ int(l[i]) for i in (1, 3, 5, 7)]
             xmin_text = str(min(xs))
@@ -88,12 +90,12 @@ def process_convert(txt_name, DIRECTORY_ANNOTATIONS, new_version=True):
             x1_text = str(xs[0])
             x2_text = str(xs[1])
             x3_text = str(xs[2])
-            x4_text = str(x3[3])
+            x4_text = str(xs[3])
 
             y1_text = str(ys[0])
             y2_text = str(ys[1])
             y3_text = str(ys[2])
-            y4_text = str(y3[3])
+            y4_text = str(ys[3])
 
 
         nodeObject = annotation_xml.createElement('object')
@@ -178,4 +180,4 @@ def get_all_txt(directory, new_version=False):
     parser.add_argument('--in_dir','-i', default='/home/zsz/datasets/icdar15_anno/eval_img_2018.9.25/icdar15_eval_final', type=str)
     args = parser.parse_args()
     directory = args.in_dir
-    get_all_txt(directory, True)
+    get_all_txt(directory, False)

train.py

@@ -149,7 +149,7 @@
 tf.app.flags.DEFINE_integer('train_image_size', None, 'Train image size')
 tf.app.flags.DEFINE_string('training_image_crop_area', '0.1, 1.0',
                            'the area of image process for training')
-tf.app.flags.DEFINE_integer('max_number_of_steps', 60000,
+tf.app.flags.DEFINE_integer('max_number_of_steps', 120000,
                             'The maxim number of training steps.')
 # =========================================================================== #
 # Fine-Tuning Flags.