artificialhoney commited on
Commit
07a01fd
1 Parent(s): 4d447d5

feat: graffiti.org conditioning

Browse files
README.md CHANGED
@@ -3,14 +3,16 @@ dataset_info:
3
  features:
4
  - name: image
5
  dtype: image
 
 
6
  - name: text
7
  dtype: string
8
  splits:
9
  - name: train
10
- num_bytes: 690468770
11
- num_examples: 77487
12
- download_size: 11106922968
13
- dataset_size: 690468770
14
  ---
15
  # Dataset Card for Graffiti
16
 
 
3
  features:
4
  - name: image
5
  dtype: image
6
+ - name: conditioning_image
7
+ dtype: image
8
  - name: text
9
  dtype: string
10
  splits:
11
  - name: train
12
+ num_bytes: 10454347970
13
+ num_examples: 77478
14
+ download_size: 16111025975
15
+ dataset_size: 10454347970
16
  ---
17
  # Dataset Card for Graffiti
18
 
cli.py CHANGED
@@ -11,8 +11,7 @@ import os
11
  from controlnet_aux.processor import Processor
12
 
13
  from PIL import UnidentifiedImageError
14
- from PIL import Image, ImageFile
15
- ImageFile.LOAD_TRUNCATED_IMAGES = True
16
 
17
 
18
  class Scraper(object):
@@ -194,8 +193,8 @@ class CLI():
194
  path = Path("./images").rglob("*.jpg")
195
  for i, img_p in enumerate(path):
196
  try:
197
- Image.open(img_p)
198
- except UnidentifiedImageError:
199
  path_name = str(img_p)
200
  print(path_name + " is broken. Deleting!")
201
  os.remove(path_name)
@@ -233,7 +232,7 @@ class CLI():
233
  data = json.loads(row)
234
  with open('./images/' + data["file"] + '.json', 'w') as j:
235
  j.write(json.dumps(data, indent=2, ensure_ascii=False))
236
- elif args.command == 'condition':
237
  processor_id = 'softedge_hed'
238
  processor = Processor(processor_id)
239
 
 
11
  from controlnet_aux.processor import Processor
12
 
13
  from PIL import UnidentifiedImageError
14
+ from PIL import Image
 
15
 
16
 
17
  class Scraper(object):
 
193
  path = Path("./images").rglob("*.jpg")
194
  for i, img_p in enumerate(path):
195
  try:
196
+ Image.open(img_p).load()
197
+ except Exception:
198
  path_name = str(img_p)
199
  print(path_name + " is broken. Deleting!")
200
  os.remove(path_name)
 
232
  data = json.loads(row)
233
  with open('./images/' + data["file"] + '.json', 'w') as j:
234
  j.write(json.dumps(data, indent=2, ensure_ascii=False))
235
+ elif args.command == 'conditioning':
236
  processor_id = 'softedge_hed'
237
  processor = Processor(processor_id)
238
 
data/graffiti.org/conditioning.tar.gz ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:08b04aa1897e3a4c1e25b732cfb4356f91a8103ff6f3487d63acc4d991d6db8a
3
+ size 2587653112
data/graffiti.org/images.tar.gz CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:97cb1cd4fc428561c70dd620036fea594e3d38aeef2037b40bd253fa43aebd19
3
- size 3213661340
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3f7d02c10631c1c8964f6c4f824ad6df152abe46602073e2884341b37ecf8d23
3
+ size 3211879389
data/graffiti.org/metadata.jsonl CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b7b1c382ba1ea07f6107d8b3a813a15f96fe87a74773dea75ce74796314e8449
3
- size 6372995
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:353d4454f0d12c698a5509a8f3064e0a5ad2f934026853b900156eda3fe8ee7e
3
+ size 6370190
graffiti.py CHANGED
@@ -71,6 +71,7 @@ class Graffiti(datasets.GeneratorBasedBuilder):
71
  features=datasets.Features(
72
  {
73
  "image": datasets.Image(),
 
74
  "text": datasets.Value("string")
75
  }
76
  ),
@@ -85,8 +86,10 @@ class Graffiti(datasets.GeneratorBasedBuilder):
85
  """Returns SplitGenerators."""
86
  images = []
87
  metadata = []
 
88
  for source in _SOURCES:
89
  images.append(dl_manager.iter_archive(dl_manager.download("./data/{0}/images.tar.gz".format(source))))
 
90
  metadata.append(dl_manager.download("./data/{0}/metadata.jsonl".format(source)))
91
  return [
92
  datasets.SplitGenerator(
@@ -95,27 +98,35 @@ class Graffiti(datasets.GeneratorBasedBuilder):
95
  gen_kwargs={
96
  "images": images,
97
  "metadata": metadata,
 
98
  },
99
  )
100
  ]
101
 
102
- def _generate_examples(self, metadata, images):
103
  idx = 0
104
  for index, meta in enumerate(metadata):
105
  m = []
106
  with open(meta, encoding="utf-8") as f:
107
  for row in f:
108
  m.append(json.loads(row))
 
109
  for file_path, file_obj in images[index]:
110
  data = [x for x in m if file_path.endswith(x["file"])][0]
 
 
 
 
 
111
  text = data["caption"]
112
  if data["artist"] != None:
113
- text += ", with text \"" + data["artist"] + "\""
114
- text += ", in the art of \"" + data["artist"] + "\""
115
  if data["city"] != None:
116
  text += ", located in " + data["city"]
117
  yield idx, {
118
  "image": {"path": file_path, "bytes": file_obj.read()},
119
- "text": text
 
120
  }
121
  idx+=1
 
71
  features=datasets.Features(
72
  {
73
  "image": datasets.Image(),
74
+ "conditioning_image": datasets.Image(),
75
  "text": datasets.Value("string")
76
  }
77
  ),
 
86
  """Returns SplitGenerators."""
87
  images = []
88
  metadata = []
89
+ conditioning = []
90
  for source in _SOURCES:
91
  images.append(dl_manager.iter_archive(dl_manager.download("./data/{0}/images.tar.gz".format(source))))
92
+ conditioning.append(dl_manager.iter_archive(dl_manager.download("./data/{0}/conditioning.tar.gz".format(source))))
93
  metadata.append(dl_manager.download("./data/{0}/metadata.jsonl".format(source)))
94
  return [
95
  datasets.SplitGenerator(
 
98
  gen_kwargs={
99
  "images": images,
100
  "metadata": metadata,
101
+ "conditioning": conditioning
102
  },
103
  )
104
  ]
105
 
106
+ def _generate_examples(self, metadata, images, conditioning):
107
  idx = 0
108
  for index, meta in enumerate(metadata):
109
  m = []
110
  with open(meta, encoding="utf-8") as f:
111
  for row in f:
112
  m.append(json.loads(row))
113
+ c = iter(conditioning[index])
114
  for file_path, file_obj in images[index]:
115
  data = [x for x in m if file_path.endswith(x["file"])][0]
116
+
117
+ conditioning_file = next(c)
118
+ conditioning_file_path = conditioning_file[0]
119
+ conditioning_file_obj = conditioning_file[1]
120
+
121
  text = data["caption"]
122
  if data["artist"] != None:
123
+ # text += ", with text " + data["artist"]
124
+ text += ", in the art of " + data["artist"]
125
  if data["city"] != None:
126
  text += ", located in " + data["city"]
127
  yield idx, {
128
  "image": {"path": file_path, "bytes": file_obj.read()},
129
+ "conditioning_image": {"path": conditioning_file_path, "bytes": conditioning_file_obj.read()},
130
+ "text": text,
131
  }
132
  idx+=1
prepare.sh CHANGED
@@ -28,5 +28,5 @@ test () {
28
  datasets-cli test graffiti.py --save_info --all_config
29
  }
30
 
31
- # list ${args[1]}
32
  tar ${args[0]} ${args[1]}
 
28
  datasets-cli test graffiti.py --save_info --all_config
29
  }
30
 
31
+ list ${args[1]}
32
  tar ${args[0]} ${args[1]}
requirements.txt CHANGED
@@ -1,4 +1,6 @@
1
  bs4
2
  requests
3
  transformers
4
- controlnet-aux==0.0.6
 
 
 
1
  bs4
2
  requests
3
  transformers
4
+ controlnet-aux==0.0.6
5
+ mediapipe
6
+ datasets