Vadzim Kashko commited on
Commit
f834331
1 Parent(s): 6fb443d

refactor: data and script

Browse files
Files changed (2) hide show
  1. data/images.zip +3 -0
  2. face_masks.py +97 -54
data/images.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:16db8eda5522a7451e803be4f636b44ab8efc5d5c7a2bcec3c479f65a1ec4712
3
+ size 100867838
face_masks.py CHANGED
@@ -1,21 +1,23 @@
1
  import datasets
 
2
  import pandas as pd
 
 
3
 
4
  _CITATION = """\
5
  @InProceedings{huggingface:dataset,
6
- title = {selfies_and_id},
7
  author = {TrainingDataPro},
8
  year = {2023}
9
  }
10
  """
11
 
12
  _DESCRIPTION = """\
13
- 4083 sets, which includes 2 photos of a person from his documents and
14
- 13 selfies. 571 sets of Hispanics and 3512 sets of Caucasians.
15
- Photo documents contains only a photo of a person.
16
- All personal information from the document is hidden.
17
  """
18
- _NAME = 'selfies_and_id'
19
 
20
  _HOMEPAGE = f"https://huggingface.co/datasets/TrainingDataPro/{_NAME}"
21
 
@@ -24,46 +26,89 @@ _LICENSE = "cc-by-nc-nd-4.0"
24
  _DATA = f"https://huggingface.co/datasets/TrainingDataPro/{_NAME}/resolve/main/data/"
25
 
26
 
27
- class SelfiesAndId(datasets.GeneratorBasedBuilder):
28
- """Small sample of image-text pairs"""
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
29
 
30
  def _info(self):
31
- return datasets.DatasetInfo(
32
- description=_DESCRIPTION,
33
- features=datasets.Features({
34
- 'id_1': datasets.Image(),
35
- 'id_2': datasets.Image(),
36
- 'selfie_1': datasets.Image(),
37
- 'selfie_2': datasets.Image(),
38
- 'selfie_3': datasets.Image(),
39
- 'selfie_4': datasets.Image(),
40
- 'selfie_5': datasets.Image(),
41
- 'selfie_6': datasets.Image(),
42
- 'selfie_7': datasets.Image(),
43
- 'selfie_8': datasets.Image(),
44
- 'selfie_9': datasets.Image(),
45
- 'selfie_10': datasets.Image(),
46
- 'selfie_11': datasets.Image(),
47
- 'selfie_12': datasets.Image(),
48
- 'selfie_13': datasets.Image(),
49
- 'user_id': datasets.Value('string'),
50
- 'set_id': datasets.Value('string'),
51
- 'user_race': datasets.Value('string'),
52
- 'name': datasets.Value('string'),
53
- 'age': datasets.Value('int8'),
54
- 'country': datasets.Value('string'),
55
- 'gender': datasets.Value('string')
56
- }),
57
- supervised_keys=None,
58
- homepage=_HOMEPAGE,
59
- citation=_CITATION,
60
- license=_LICENSE
61
- )
62
 
63
  def _split_generators(self, dl_manager):
64
- images = dl_manager.download(f"{_DATA}images.tar.gz")
65
  annotations = dl_manager.download(f"{_DATA}{_NAME}.csv")
66
- images = dl_manager.iter_archive(images)
67
  return [
68
  datasets.SplitGenerator(name=datasets.Split.TRAIN,
69
  gen_kwargs={
@@ -73,24 +118,22 @@ class SelfiesAndId(datasets.GeneratorBasedBuilder):
73
  ]
74
 
75
  def _generate_examples(self, images, annotations):
76
- annotations_df = pd.read_csv(annotations, sep=';')
77
- images_data = pd.DataFrame(columns=['URL', 'Bytes'])
78
- for idx, (image_path, image) in enumerate(images):
79
- images_data.loc[idx] = {'URL': image_path, 'Bytes': image.read()}
 
80
 
81
  annotations_df = pd.merge(annotations_df,
82
  images_data,
83
  how='left',
84
- on=['URL'])
85
- for idx, worker_id in enumerate(pd.unique(annotations_df['UserId'])):
86
- annotation = annotations_df.loc[annotations_df['UserId'] ==
87
- worker_id]
88
- annotation = annotation.sort_values(['FName'])
89
  data = {
90
- row[5].lower(): {
91
- 'path': row[6],
92
- 'bytes': row[10]
93
- } for row in annotation.itertuples()
94
  }
95
 
96
  age = annotation.loc[annotation['FName'] ==
 
1
  import datasets
2
+ import numpy as np
3
  import pandas as pd
4
+ import PIL.Image
5
+ import PIL.ImageOps
6
 
7
  _CITATION = """\
8
  @InProceedings{huggingface:dataset,
9
+ title = {face_masks},
10
  author = {TrainingDataPro},
11
  year = {2023}
12
  }
13
  """
14
 
15
  _DESCRIPTION = """\
16
+ Dataset includes 250 000 images, 4 types of mask worn on 28 000 unique faces.
17
+ All images were collected using the Toloka.ai crowdsourcing service and
18
+ validated by TrainingData.pro
 
19
  """
20
+ _NAME = 'face_masks'
21
 
22
  _HOMEPAGE = f"https://huggingface.co/datasets/TrainingDataPro/{_NAME}"
23
 
 
26
  _DATA = f"https://huggingface.co/datasets/TrainingDataPro/{_NAME}/resolve/main/data/"
27
 
28
 
29
+ def exif_transpose(img):
30
+ if not img:
31
+ return img
32
+
33
+ exif_orientation_tag = 274
34
+
35
+ # Check for EXIF data (only present on some files)
36
+ if hasattr(img, "_getexif") and isinstance(
37
+ img._getexif(), dict) and exif_orientation_tag in img._getexif():
38
+ exif_data = img._getexif()
39
+ orientation = exif_data[exif_orientation_tag]
40
+
41
+ # Handle EXIF Orientation
42
+ if orientation == 1:
43
+ # Normal image - nothing to do!
44
+ pass
45
+ elif orientation == 2:
46
+ # Mirrored left to right
47
+ img = img.transpose(PIL.Image.FLIP_LEFT_RIGHT)
48
+ elif orientation == 3:
49
+ # Rotated 180 degrees
50
+ img = img.rotate(180)
51
+ elif orientation == 4:
52
+ # Mirrored top to bottom
53
+ img = img.rotate(180).transpose(PIL.Image.FLIP_LEFT_RIGHT)
54
+ elif orientation == 5:
55
+ # Mirrored along top-left diagonal
56
+ img = img.rotate(-90,
57
+ expand=True).transpose(PIL.Image.FLIP_LEFT_RIGHT)
58
+ elif orientation == 6:
59
+ # Rotated 90 degrees
60
+ img = img.rotate(-90, expand=True)
61
+ elif orientation == 7:
62
+ # Mirrored along top-right diagonal
63
+ img = img.rotate(90,
64
+ expand=True).transpose(PIL.Image.FLIP_LEFT_RIGHT)
65
+ elif orientation == 8:
66
+ # Rotated 270 degrees
67
+ img = img.rotate(90, expand=True)
68
+
69
+ return img
70
+
71
+
72
+ def load_image_file(file, mode='RGB'):
73
+ # Load the image with PIL
74
+ img = PIL.Image.open(file)
75
+
76
+ if hasattr(PIL.ImageOps, 'exif_transpose'):
77
+ # Very recent versions of PIL can do exit transpose internally
78
+ img = PIL.ImageOps.exif_transpose(img)
79
+ else:
80
+ # Otherwise, do the exif transpose ourselves
81
+ img = exif_transpose(img)
82
+
83
+ img = img.convert(mode)
84
+
85
+ return np.array(img)
86
+
87
+
88
+ class FaceMasks(datasets.GeneratorBasedBuilder):
89
 
90
  def _info(self):
91
+ return datasets.DatasetInfo(description=_DESCRIPTION,
92
+ features=datasets.Features({
93
+ 'photo_1': datasets.Image(),
94
+ 'photo_2': datasets.Image(),
95
+ 'photo_3': datasets.Image(),
96
+ 'photo_4': datasets.Image(),
97
+ 'selfie_5': datasets.Image(),
98
+ 'worker_id': datasets.Value('string'),
99
+ 'age': datasets.Value('int8'),
100
+ 'country': datasets.Value('string'),
101
+ 'sex': datasets.Value('string')
102
+ }),
103
+ supervised_keys=None,
104
+ homepage=_HOMEPAGE,
105
+ citation=_CITATION,
106
+ license=_LICENSE)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
107
 
108
  def _split_generators(self, dl_manager):
109
+ images = dl_manager.download_and_extract(f"{_DATA}images.zip")
110
  annotations = dl_manager.download(f"{_DATA}{_NAME}.csv")
111
+ images = dl_manager.iter_files(images)
112
  return [
113
  datasets.SplitGenerator(name=datasets.Split.TRAIN,
114
  gen_kwargs={
 
118
  ]
119
 
120
  def _generate_examples(self, images, annotations):
121
+ annotations_df = pd.read_csv(annotations, sep=',')
122
+ images_data = pd.DataFrame(columns=['Link', 'Image'])
123
+ for idx, image_path in enumerate(images):
124
+ image = load_image_file(image_path)
125
+ images_data.loc[idx] = {'Link': image_path, 'Image': image}
126
 
127
  annotations_df = pd.merge(annotations_df,
128
  images_data,
129
  how='left',
130
+ on=['Link'])
131
+ for idx, worker_id in enumerate(pd.unique(annotations_df['WorkerId'])):
132
+ annotation: pd.DataFrame = annotations_df.loc[
133
+ annotations_df['WorkerId'] == worker_id]
134
+ annotation = annotation.sort_values(['Link'])
135
  data = {
136
+ f'photo_{row[0]}': row[6] for row in annotation.itertuples()
 
 
 
137
  }
138
 
139
  age = annotation.loc[annotation['FName'] ==