johnnv commited on
Commit
b6cff75
1 Parent(s): a6aa8b1

Create new file

Browse files
Files changed (1) hide show
  1. CCAgT.py +355 -0
CCAgT.py ADDED
@@ -0,0 +1,355 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import json
2
+ import os
3
+ from collections import OrderedDict, defaultdict
4
+ from math import ceil
5
+
6
+ import numpy as np
7
+ import pandas as pd
8
+
9
+ import datasets
10
+
11
+
12
+ logger = datasets.logging.get_logger(__name__)
13
+
14
+ CCAGT_CLASSES = OrderedDict(
15
+ {
16
+ 1: "NUCLEUS",
17
+ 2: "CLUSTER",
18
+ 3: "SATELLITE",
19
+ 4: "NUCLEUS_OUT_OF_FOCUS",
20
+ 5: "OVERLAPPED_NUCLEI",
21
+ 6: "NON_VIABLE_NUCLEUS",
22
+ 7: "LEUKOCYTE_NUCLEUS",
23
+ }
24
+ )
25
+
26
+ _LICENSE = "CC BY NC 3.0 License"
27
+
28
+ _CITATION = """\
29
+ @misc{CCAgTDataset,
30
+ doi = {10.17632/WG4BPM33HJ.2},
31
+ url = {https://data.mendeley.com/datasets/wg4bpm33hj/2},
32
+ author = {Jo{\\~{a}}o Gustavo Atkinson Amorim and Andr{\'{e}} Vict{\'{o}}ria Matias and Tainee Bottamedi and Vinícius Sanches and Ane Francyne Costa and Fabiana Botelho De Miranda Onofre and Alexandre Sherlley Casimiro Onofre and Aldo von Wangenheim},
33
+ title = {CCAgT: Images of Cervical Cells with AgNOR Stain Technique},
34
+ publisher = {Mendeley},
35
+ year = {2022},
36
+ copyright = {Attribution-NonCommercial 3.0 Unported}
37
+ }
38
+ """
39
+
40
+ _HOMEPAGE = "https://data.mendeley.com/datasets/wg4bpm33hj"
41
+
42
+ _DESCRIPTION = """\
43
+ The CCAgT (Images of Cervical Cells with AgNOR Stain Technique) dataset contains 9339 images (1600x1200 resolution where each pixel is 0.111µmX0.111µm) from 15 different slides stained using the AgNOR technique.
44
+ Each image has at least one label. In total, this dataset has more than 63K instances of annotated object.
45
+ The images are from the patients of the Gynecology and Colonoscopy Outpatient Clinic of the Polydoro Ernani de São Thiago University Hospital of the Universidade Federal de Santa Catarina (HU-UFSC).
46
+ """
47
+
48
+ _DATA_URL = "https://md-datasets-cache-zipfiles-prod.s3.eu-west-1.amazonaws.com/wg4bpm33hj-2.zip"
49
+
50
+
51
+ def tvt(ids, tvt_size, seed=1609):
52
+ """From a list of indexes/ids (int) will generate the train-validation-test data.
53
+
54
+ Based on `github.com/scikit-learn/scikit-learn/blob/37ac6788c9504ee409b75e5e24ff7d86c90c2ffb/sklearn/model_selection/_split.py#L2321`
55
+ """
56
+ n_samples = len(ids)
57
+
58
+ qtd = {
59
+ "valid": ceil(n_samples * tvt_size[1]),
60
+ "test": ceil(n_samples * tvt_size[2]),
61
+ }
62
+ qtd["train"] = int(n_samples - qtd["valid"] - qtd["test"])
63
+
64
+ rng = np.random.RandomState(seed)
65
+ permutatation = rng.permutation(ids)
66
+
67
+ out = {
68
+ "train": set(permutatation[: qtd["train"]]),
69
+ "valid": set(permutatation[qtd["train"] : qtd["train"] + qtd["valid"]]),
70
+ "test": set(permutatation[qtd["train"] + qtd["valid"] :]),
71
+ }
72
+
73
+ return out["train"], out["valid"], out["test"]
74
+
75
+
76
+ def annotations_per_image(df):
77
+ """
78
+ based on: https://github.com/johnnv1/CCAgT-utils/blob/54ade78e4ddb2e2ed9507b8a1633940897767cac/CCAgT_utils/describe.py#L152
79
+ """
80
+ df_describe_images = df.groupby(["image_id", "category_id"]).size().reset_index().rename(columns={0: "count"})
81
+ df_describe_images = df_describe_images.pivot(columns=["category_id"], index="image_id")
82
+ df_describe_images = df_describe_images.rename(CCAGT_CLASSES, axis=1)
83
+ df_describe_images["qtd_annotations"] = df_describe_images.sum(axis=1)
84
+ df_describe_images = df_describe_images.fillna(0)
85
+ df_describe_images["NORs"] = (
86
+ df_describe_images[
87
+ "count",
88
+ CCAGT_CLASSES[2],
89
+ ]
90
+ + df_describe_images[
91
+ "count",
92
+ CCAGT_CLASSES[3],
93
+ ]
94
+ )
95
+
96
+ return df_describe_images
97
+
98
+
99
+ def tvt_by_nors(df, tvt_size=(0.7, 0.15, 0.15), **kwargs):
100
+ """This will split the CCAgT annotations based on the number of NORs
101
+ into each image. With a silly separation, first will split
102
+ between each fold images with one or less NORs, after will split
103
+ images with the amount of NORs is between 2 and 7, and at least will
104
+ split images that have more than 7 NORs.
105
+
106
+ based on `https://github.com/johnnv1/CCAgT-utils/blob/54ade78e4ddb2e2ed9507b8a1633940897767cac/CCAgT_utils/split.py#L64`
107
+ """
108
+ if sum(tvt_size) != 1:
109
+ raise ValueError("The sum of `tvt_size` need to be equal to 1!")
110
+
111
+ df_describe_imgs = annotations_per_image(df)
112
+
113
+ img_ids = {}
114
+ img_ids["low_nors"] = df_describe_imgs.loc[(df_describe_imgs["NORs"] < 2)].index
115
+ img_ids["medium_nors"] = df_describe_imgs[(df_describe_imgs["NORs"] >= 2) * (df_describe_imgs["NORs"] <= 7)].index
116
+ img_ids["high_nors"] = df_describe_imgs[(df_describe_imgs["NORs"] > 7)].index
117
+
118
+ train_ids = set({})
119
+ valid_ids = set({})
120
+ test_ids = set({})
121
+
122
+ for k, ids in img_ids.items():
123
+ logger.info(f"Splitting {len(ids)} images with {k} quantity...")
124
+ if len(ids) == 0:
125
+ continue
126
+ _train, _valid, _test = tvt(ids, tvt_size, **kwargs)
127
+ logger.info(f">T: {len(_train)} V: {len(_valid)} T: {len(_test)}")
128
+ train_ids = train_ids.union(_train)
129
+ valid_ids = valid_ids.union(_valid)
130
+ test_ids = test_ids.union(_test)
131
+
132
+ return train_ids, valid_ids, test_ids
133
+
134
+
135
+ def get_basename(path):
136
+ return os.path.splitext(os.path.basename(path))[0]
137
+
138
+
139
+ def get_slide_id(path):
140
+ bn = get_basename(path)
141
+ slide_id = bn.split("_")[0]
142
+ return slide_id
143
+
144
+
145
+ class CCAgTConfig(datasets.BuilderConfig):
146
+ """BuilderConfig for CCAgT."""
147
+
148
+ seed = 1609
149
+ tvt_size = (0.7, 0.15, 0.15)
150
+
151
+
152
+ class CCAgT(datasets.GeneratorBasedBuilder):
153
+ """Images of Cervical Cells with AgNOR Stain Technique (CCAgT) dataset"""
154
+
155
+ test_dummy_data = False
156
+
157
+ VERSION = datasets.Version("2.0.0")
158
+
159
+ BUILDER_CONFIG_CLASS = CCAgTConfig
160
+ BUILDER_CONFIGS = [
161
+ CCAgTConfig(name="semantic_segmentation", version=VERSION, description="The semantic segmentation variant."),
162
+ CCAgTConfig(name="object_detection", version=VERSION, description="The object detection variant."),
163
+ CCAgTConfig(name="instance_segmentation", version=VERSION, description="The instance segmentation variant."),
164
+ ]
165
+
166
+ DEFAULT_CONFIG_NAME = "semantic_segmentation"
167
+
168
+ def _info(self):
169
+ assert len(CCAGT_CLASSES) == 7
170
+
171
+ if self.config.name == "semantic_segmentation":
172
+ features = datasets.Features(
173
+ {
174
+ "image": datasets.Image(),
175
+ "annotation": datasets.Image(),
176
+ }
177
+ )
178
+ elif self.config.name == "object_detection":
179
+ features = datasets.Features(
180
+ {
181
+ "image": datasets.Image(),
182
+ "objects": datasets.Sequence(
183
+ {
184
+ "bbox": datasets.Sequence(datasets.Value("float32"), length=4),
185
+ "label": datasets.ClassLabel(names=list(CCAGT_CLASSES.values())),
186
+ }
187
+ ),
188
+ }
189
+ )
190
+ elif self.config.name == "instance_segmentation":
191
+ features = datasets.Features(
192
+ {
193
+ "image": datasets.Image(),
194
+ "objects": datasets.Sequence(
195
+ {
196
+ "bbox": datasets.Sequence(datasets.Value("float32"), length=4),
197
+ "segment": datasets.Sequence(datasets.Sequence(datasets.Value("float32"))),
198
+ "label": datasets.ClassLabel(names=list(CCAGT_CLASSES.values())),
199
+ }
200
+ ),
201
+ }
202
+ )
203
+ else:
204
+ raise NotImplementedError
205
+
206
+ return datasets.DatasetInfo(
207
+ description=_DESCRIPTION,
208
+ features=features,
209
+ homepage=_HOMEPAGE,
210
+ license=_LICENSE,
211
+ citation=_CITATION,
212
+ )
213
+
214
+ def _download_and_extract_all(self, dl_manager):
215
+ def extracted_by_slide(paths):
216
+ return {get_slide_id(path): dl_manager.extract(path) for path in paths}
217
+
218
+ data_dir = dl_manager.download_and_extract(_DATA_URL)
219
+ base_path = os.path.join(data_dir, "wg4bpm33hj-2")
220
+
221
+ logger.info("Extracting images...")
222
+ self.images_base_dir = os.path.join(base_path, "images")
223
+ images_to_extract = [
224
+ os.path.join(self.images_base_dir, fn) for fn in os.listdir(self.images_base_dir) if fn.endswith(".zip")
225
+ ]
226
+ self.images_extracted = extracted_by_slide(images_to_extract)
227
+
228
+ if self.config.name == "semantic_segmentation":
229
+ logger.info("Extracting masks...")
230
+ self.masks_base_dir = os.path.join(base_path, "masks")
231
+ masks_to_extract = [
232
+ os.path.join(self.masks_base_dir, fn) for fn in os.listdir(self.masks_base_dir) if fn.endswith(".zip")
233
+ ]
234
+ self.masks_extracted = extracted_by_slide(masks_to_extract)
235
+ elif self.config.name in {"object_detection", "instance_segmentation"}:
236
+ logger.info("Reading COCO OD file...")
237
+ ccagt_OD_COCO_path = os.path.join(base_path, "CCAgT_COCO_OD.json")
238
+ with open(ccagt_OD_COCO_path, "r", encoding="utf-8") as json_file:
239
+ coco_OD = json.load(json_file)
240
+
241
+ self._imageid_to_coco_OD_annotations = defaultdict(list)
242
+ for labels in coco_OD["annotations"]:
243
+ self._imageid_to_coco_OD_annotations[labels["image_id"]].append(labels)
244
+
245
+ logger.info("Loading dataset info...")
246
+ ccagt_raw_path = os.path.join(base_path, "CCAgT.parquet.gzip")
247
+ with open(ccagt_raw_path, "rb") as f:
248
+ self._ccagt_info = pd.read_parquet(f, columns=["image_name", "category_id", "image_id", "slide_id"])
249
+ self._bn_to_imageid = pd.Series(
250
+ self._ccagt_info["image_id"].values, index=self._ccagt_info["image_name"]
251
+ ).to_dict()
252
+
253
+ def _split_generators(self, dl_manager):
254
+ """Returns SplitGenerators."""
255
+
256
+ def build_path(basename, tp="images"):
257
+ slide = basename.split("_")[0]
258
+ if tp == "images":
259
+ dir_path = self.images_extracted[slide]
260
+ ext = ".jpg"
261
+ else:
262
+ dir_path = self.masks_extracted[slide]
263
+ ext = ".png"
264
+
265
+ return os.path.join(dir_path, slide, basename + ext)
266
+
267
+ def images_and_masks(basenames):
268
+ for bn in basenames:
269
+ yield build_path(bn), build_path(bn, "masks")
270
+
271
+ def images_and_boxes(basenames):
272
+ for bn in basenames:
273
+ image_id = self._bn_to_imageid[bn]
274
+ labels = [
275
+ {"bbox": annotation["bbox"], "label": annotation["category_id"] - 1}
276
+ for annotation in self._imageid_to_coco_OD_annotations[image_id]
277
+ ]
278
+
279
+ yield build_path(bn), labels
280
+
281
+ def images_and_instances(basenames):
282
+ for bn in basenames:
283
+ image_id = self._bn_to_imageid[bn]
284
+ instances = [
285
+ {
286
+ "bbox": annotation["bbox"],
287
+ "label": annotation["category_id"] - 1,
288
+ "segment": annotation["segmentation"],
289
+ }
290
+ for annotation in self._imageid_to_coco_OD_annotations[image_id]
291
+ ]
292
+
293
+ yield build_path(bn), instances
294
+
295
+ self._download_and_extract_all(dl_manager)
296
+
297
+ logger.info("Splitting dataset based on the NORs quantity by image...")
298
+ train_ids, valid_ids, test_ids = tvt_by_nors(
299
+ self._ccagt_info, tvt_size=self.config.tvt_size, seed=self.config.seed
300
+ )
301
+ train_bn_images = self._ccagt_info.loc[self._ccagt_info["image_id"].isin(train_ids), "image_name"].unique()
302
+ valid_bn_images = self._ccagt_info.loc[self._ccagt_info["image_id"].isin(valid_ids), "image_name"].unique()
303
+ test_bn_images = self._ccagt_info.loc[self._ccagt_info["image_id"].isin(test_ids), "image_name"].unique()
304
+
305
+ if self.config.name == "semantic_segmentation":
306
+ train_data = images_and_masks(train_bn_images)
307
+ valid_data = images_and_masks(valid_bn_images)
308
+ test_data = images_and_masks(test_bn_images)
309
+ elif self.config.name == "object_detection":
310
+ train_data = images_and_boxes(train_bn_images)
311
+ valid_data = images_and_boxes(valid_bn_images)
312
+ test_data = images_and_boxes(test_bn_images)
313
+ elif self.config.name == "instance_segmentation":
314
+ train_data = images_and_instances(train_bn_images)
315
+ valid_data = images_and_instances(valid_bn_images)
316
+ test_data = images_and_instances(test_bn_images)
317
+ else:
318
+ raise NotImplementedError
319
+
320
+ return [
321
+ datasets.SplitGenerator(
322
+ name=datasets.Split.TRAIN,
323
+ gen_kwargs={"data": train_data},
324
+ ),
325
+ datasets.SplitGenerator(
326
+ name=datasets.Split.TEST,
327
+ gen_kwargs={"data": test_data},
328
+ ),
329
+ datasets.SplitGenerator(
330
+ name=datasets.Split.VALIDATION,
331
+ gen_kwargs={"data": valid_data},
332
+ ),
333
+ ]
334
+
335
+ def _generate_examples(self, data):
336
+ if self.config.name == "semantic_segmentation":
337
+ for img_path, msk_path in data:
338
+ img_basename = get_basename(img_path)
339
+ image_id = self._bn_to_imageid[img_basename]
340
+ yield image_id, {
341
+ "image": img_path,
342
+ "annotation": msk_path,
343
+ }
344
+ elif self.config.name == "object_detection":
345
+ for img_path, labels in data:
346
+ img_basename = get_basename(img_path)
347
+ image_id = self._bn_to_imageid[img_basename]
348
+ yield image_id, {"image": img_path, "objects": labels}
349
+ elif self.config.name == "instance_segmentation":
350
+ for img_path, instances in data:
351
+ img_basename = get_basename(img_path)
352
+ image_id = self._bn_to_imageid[img_basename]
353
+ yield image_id, {"image": img_path, "objects": instances}
354
+ else:
355
+ raise NotImplementedError