MrPotato commited on
Commit
bbff295
·
1 Parent(s): b866c7a

removed normalization of bboxes

Browse files
Files changed (1) hide show
  1. docbank.py +23 -25
docbank.py CHANGED
@@ -83,27 +83,27 @@ def load_image(image_path, size=None):
83
  return image, (w, h)
84
 
85
 
86
- def normalize_bbox(bbox, size):
87
- return [
88
- int(1000 * int(bbox[0]) / size[0]),
89
- int(1000 * int(bbox[1]) / size[1]),
90
- int(1000 * int(bbox[2]) / size[0]),
91
- int(1000 * int(bbox[3]) / size[1]),
92
- ]
93
-
94
-
95
- def simplify_bbox(bbox):
96
- return [
97
- min(bbox[0::2]),
98
- min(bbox[1::2]),
99
- max(bbox[2::2]),
100
- max(bbox[3::2]),
101
- ]
102
-
103
-
104
- def merge_bbox(bbox_list):
105
- x0, y0, x1, y1 = list(zip(*bbox_list))
106
- return [min(x0), min(y0), max(x1), max(y1)]
107
 
108
 
109
  # TODO: Name of the dataset usually matches the script name with CamelCase instead of snake_case
@@ -225,7 +225,8 @@ class Docbank(datasets.GeneratorBasedBuilder):
225
  with open(f_fp_txt, newline='', encoding='utf-8') as csvfile:
226
  reader = csv.reader(csvfile, delimiter='\t', quotechar=' ')
227
  for row in reader:
228
- normalized_bbox = normalize_bbox(row[1:5], size)
 
229
  tokenized_input = self.TOKENIZER(
230
  row[0],
231
  add_special_tokens=False,
@@ -268,6 +269,3 @@ class Docbank(datasets.GeneratorBasedBuilder):
268
  "labels": split_labels
269
  }
270
  key += 1
271
- break
272
-
273
-
 
83
  return image, (w, h)
84
 
85
 
86
+ # def normalize_bbox(bbox, size):
87
+ # return [
88
+ # int(1000 * int(bbox[0]) / size[0]),
89
+ # int(1000 * int(bbox[1]) / size[1]),
90
+ # int(1000 * int(bbox[2]) / size[0]),
91
+ # int(1000 * int(bbox[3]) / size[1]),
92
+ # ]
93
+ #
94
+ #
95
+ # def simplify_bbox(bbox):
96
+ # return [
97
+ # min(bbox[0::2]),
98
+ # min(bbox[1::2]),
99
+ # max(bbox[2::2]),
100
+ # max(bbox[3::2]),
101
+ # ]
102
+ #
103
+ #
104
+ # def merge_bbox(bbox_list):
105
+ # x0, y0, x1, y1 = list(zip(*bbox_list))
106
+ # return [min(x0), min(y0), max(x1), max(y1)]
107
 
108
 
109
  # TODO: Name of the dataset usually matches the script name with CamelCase instead of snake_case
 
225
  with open(f_fp_txt, newline='', encoding='utf-8') as csvfile:
226
  reader = csv.reader(csvfile, delimiter='\t', quotechar=' ')
227
  for row in reader:
228
+ # normalized_bbox = normalize_bbox(row[1:5], size)
229
+ normalized_bbox = [int(x) for x in row[1:5]]
230
  tokenized_input = self.TOKENIZER(
231
  row[0],
232
  add_special_tokens=False,
 
269
  "labels": split_labels
270
  }
271
  key += 1