VictorSanh HF staff commited on
Commit
0f7c6fd
1 Parent(s): bc92237

aggregate captions

Browse files
Files changed (2) hide show
  1. LocalizedNarratives.py +81 -27
  2. README.md +3 -1
LocalizedNarratives.py CHANGED
@@ -60,33 +60,45 @@ _ANNOTATION_URLs = {
60
  }
61
 
62
 
63
- _FEATURES = datasets.Features(
64
- {
65
- "image": datasets.Image(),
66
- "image_url": datasets.Value("string"),
67
- "dataset_id": datasets.Value("string"),
68
- "image_id": datasets.Value("string"),
69
- "annotator_id": datasets.Value("int32"),
70
- "caption": datasets.Value("string"),
71
- "timed_caption": datasets.Sequence(
72
- {
73
- "utterance": datasets.Value("string"),
74
- "start_time": datasets.Value("float32"),
75
- "end_time": datasets.Value("float32"),
76
- }
77
- ),
78
- "traces": datasets.Sequence(
79
- datasets.Sequence(
80
  {
81
- "x": datasets.Value("float32"),
82
- "y": datasets.Value("float32"),
83
- "t": datasets.Value("float32"),
84
  }
85
- )
86
- ),
87
- "voice_recording": datasets.Value("string"),
88
- }
89
- )
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
90
 
91
 
92
  class LocalizedNarrativesOpenImages(datasets.GeneratorBasedBuilder):
@@ -95,7 +107,16 @@ class LocalizedNarrativesOpenImages(datasets.GeneratorBasedBuilder):
95
  VERSION = datasets.Version("1.0.0")
96
 
97
  BUILDER_CONFIGS = [
98
- datasets.BuilderConfig(name="OpenImages", version=VERSION, description="OpenImages subset of Localized Narratives"),
 
 
 
 
 
 
 
 
 
99
  ]
100
 
101
  DEFAULT_CONFIG_NAME = "OpenImages"
@@ -103,7 +124,7 @@ class LocalizedNarrativesOpenImages(datasets.GeneratorBasedBuilder):
103
  def _info(self):
104
  return datasets.DatasetInfo(
105
  description=_DESCRIPTION,
106
- features=_FEATURES,
107
  homepage=_HOMEPAGE,
108
  license=_LICENSE,
109
  citation=_CITATION,
@@ -120,6 +141,12 @@ class LocalizedNarrativesOpenImages(datasets.GeneratorBasedBuilder):
120
  ]
121
 
122
  def _generate_examples(self, annotation_list: str, split: str):
 
 
 
 
 
 
123
  counter = 0
124
  for annotation_file in annotation_list:
125
  with open(annotation_file, "r", encoding="utf-8") as fi:
@@ -138,3 +165,30 @@ class LocalizedNarrativesOpenImages(datasets.GeneratorBasedBuilder):
138
  "voice_recording": annotation["voice_recording"],
139
  }
140
  counter += 1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
60
  }
61
 
62
 
63
+ _FEATURES = {
64
+ "OpenImages": datasets.Features(
65
+ {
66
+ "image": datasets.Image(),
67
+ "image_url": datasets.Value("string"),
68
+ "dataset_id": datasets.Value("string"),
69
+ "image_id": datasets.Value("string"),
70
+ "annotator_id": datasets.Value("int32"),
71
+ "caption": datasets.Value("string"),
72
+ "timed_caption": datasets.Sequence(
 
 
 
 
 
 
 
73
  {
74
+ "utterance": datasets.Value("string"),
75
+ "start_time": datasets.Value("float32"),
76
+ "end_time": datasets.Value("float32"),
77
  }
78
+ ),
79
+ "traces": datasets.Sequence(
80
+ datasets.Sequence(
81
+ {
82
+ "x": datasets.Value("float32"),
83
+ "y": datasets.Value("float32"),
84
+ "t": datasets.Value("float32"),
85
+ }
86
+ )
87
+ ),
88
+ "voice_recording": datasets.Value("string"),
89
+ }
90
+ ),
91
+ "OpenImages_captions": datasets.Features(
92
+ {
93
+ "image": datasets.Image(),
94
+ "image_url": datasets.Value("string"),
95
+ "dataset_id": datasets.Value("string"),
96
+ "image_id": datasets.Value("string"),
97
+ "annotator_ids": [datasets.Value("int32")],
98
+ "captions": [datasets.Value("string")],
99
+ }
100
+ ),
101
+ }
102
 
103
 
104
  class LocalizedNarrativesOpenImages(datasets.GeneratorBasedBuilder):
 
107
  VERSION = datasets.Version("1.0.0")
108
 
109
  BUILDER_CONFIGS = [
110
+ datasets.BuilderConfig(
111
+ name="OpenImages",
112
+ version=VERSION,
113
+ description="OpenImages subset of Localized Narratives"
114
+ ),
115
+ datasets.BuilderConfig(
116
+ name="OpenImages_captions",
117
+ version=VERSION,
118
+ description="OpenImages subset of Localized Narratives where captions are groupped per image (images can have multiple captions). For this subset, `timed_caption`, `traces` and `voice_recording` are not available."
119
+ ),
120
  ]
121
 
122
  DEFAULT_CONFIG_NAME = "OpenImages"
 
124
  def _info(self):
125
  return datasets.DatasetInfo(
126
  description=_DESCRIPTION,
127
+ features=_FEATURES[self.config.name],
128
  homepage=_HOMEPAGE,
129
  license=_LICENSE,
130
  citation=_CITATION,
 
141
  ]
142
 
143
  def _generate_examples(self, annotation_list: str, split: str):
144
+ if self.config.name == "OpenImages":
145
+ return self._generate_examples_original_format(annotation_list, split)
146
+ elif self.config.name == "OpenImages_captions":
147
+ return self._generate_examples_aggregated_captions(annotation_list, split)
148
+
149
+ def _generate_examples_original_format(self, annotation_list: str, split: str):
150
  counter = 0
151
  for annotation_file in annotation_list:
152
  with open(annotation_file, "r", encoding="utf-8") as fi:
 
165
  "voice_recording": annotation["voice_recording"],
166
  }
167
  counter += 1
168
+
169
+ def _generate_examples_aggregated_captions(self, annotation_list: str, split: str):
170
+ result = {}
171
+ for annotation_file in annotation_list:
172
+ with open(annotation_file, "r", encoding="utf-8") as fi:
173
+ for line in fi:
174
+ annotation = json.loads(line)
175
+ image_url = f"https://s3.amazonaws.com/open-images-dataset/{split}/{annotation['image_id']}.jpg"
176
+ image_id = annotation["image_id"]
177
+ if image_id in result:
178
+ assert result[image_id]["dataset_id"] == annotation["dataset_id"]
179
+ assert result[image_id]["image_id"] == annotation["image_id"]
180
+ result[image_id]["annotator_ids"].append(annotation["annotator_id"])
181
+ result[image_id]["captions"].append(annotation["caption"])
182
+ else:
183
+ result[image_id] = {
184
+ "image": image_url,
185
+ "image_url": image_url,
186
+ "dataset_id": annotation["dataset_id"],
187
+ "image_id": image_id,
188
+ "annotator_ids": [annotation["annotator_id"]],
189
+ "captions": [annotation["caption"]],
190
+ }
191
+ counter = 0
192
+ for r in result.values():
193
+ yield counter, r
194
+ counter += 1
README.md CHANGED
@@ -45,7 +45,9 @@ Since the voice and the mouse pointer are synchronized, we can localize every si
45
  This dense visual grounding takes the form of a mouse trace segment per word and is unique to our data.
46
  We annotated 849k images with Localized Narratives: the whole COCO, Flickr30k, and ADE20K datasets, and 671k images of Open Images, all of which we make publicly available.
47
 
48
- As of now, there is only the OpenImages subset, but feel free to contribute the other subset of Localized Narratives!
 
 
49
 
50
  ### Supported Tasks and Leaderboards
51
 
 
45
  This dense visual grounding takes the form of a mouse trace segment per word and is unique to our data.
46
  We annotated 849k images with Localized Narratives: the whole COCO, Flickr30k, and ADE20K datasets, and 671k images of Open Images, all of which we make publicly available.
47
 
48
+ As of now, there is only the `OpenImages` subset, but feel free to contribute the other subset of Localized Narratives!
49
+
50
+ `OpenImages_captions` is similar to the `OpenImages` subset. The differences are that captions are groupped per image (images can have multiple captions). For this subset, `timed_caption`, `traces` and `voice_recording` are not available.
51
 
52
  ### Supported Tasks and Leaderboards
53