add annotations
Browse files- Caltech-101.py +80 -30
Caltech-101.py
CHANGED
@@ -14,10 +14,13 @@
|
|
14 |
"""Caltech 101 loading script"""
|
15 |
|
16 |
|
|
|
|
|
17 |
from pathlib import Path
|
18 |
|
19 |
import datasets
|
20 |
import numpy as np
|
|
|
21 |
from datasets.tasks import ImageClassification
|
22 |
|
23 |
_CITATION = """\
|
@@ -147,6 +150,15 @@ _NAMES = [
|
|
147 |
"wrench",
|
148 |
"yin_yang",
|
149 |
]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
150 |
|
151 |
_TRAIN_POINTS_PER_CLASS = 30
|
152 |
|
@@ -173,36 +185,56 @@ class Caltech101(datasets.GeneratorBasedBuilder):
|
|
173 |
]
|
174 |
|
175 |
def _info(self):
|
176 |
-
|
177 |
-
|
178 |
-
features=datasets.Features(
|
179 |
{
|
180 |
"image": datasets.Image(),
|
181 |
"label": datasets.features.ClassLabel(names=_NAMES),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
182 |
}
|
183 |
-
)
|
184 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
185 |
homepage=_HOMEPAGE,
|
186 |
license=_LICENSE,
|
187 |
citation=_CITATION,
|
188 |
-
task_templates=ImageClassification(
|
189 |
-
image_column="image", label_column="label"
|
190 |
-
),
|
191 |
)
|
192 |
|
193 |
def _split_generators(self, dl_manager):
|
194 |
data_root_dir = dl_manager.download_and_extract(_DATA_URL)
|
195 |
-
|
196 |
file
|
197 |
for file in dl_manager.iter_files(data_root_dir)
|
198 |
if Path(file).name == "101_ObjectCategories.tar.gz"
|
199 |
][0]
|
200 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
201 |
return [
|
202 |
datasets.SplitGenerator(
|
203 |
name=datasets.Split.TRAIN,
|
204 |
gen_kwargs={
|
205 |
-
"
|
|
|
206 |
"split": "train",
|
207 |
"config_name": self.config.name,
|
208 |
},
|
@@ -210,58 +242,76 @@ class Caltech101(datasets.GeneratorBasedBuilder):
|
|
210 |
datasets.SplitGenerator(
|
211 |
name=datasets.Split.TEST,
|
212 |
gen_kwargs={
|
213 |
-
"
|
|
|
214 |
"split": "test",
|
215 |
"config_name": self.config.name,
|
216 |
},
|
217 |
),
|
218 |
]
|
219 |
|
220 |
-
def _generate_examples(self,
|
221 |
# Same stratagy as the one proposed in TF datasets: 30 random examples from each class are added to the train
|
222 |
# split, and the remainder are added to the test split.
|
223 |
# Source: https://github.com/tensorflow/datasets/blob/1106d587f97c4fca68c5b593dc7dc48c790ffa8c/tensorflow_datasets/image_classification/caltech.py#L88-L140
|
224 |
|
225 |
is_train_split = split == "train"
|
226 |
-
|
227 |
# Sets random seed so the random partitioning of files is the same when
|
228 |
# called for the train and test splits.
|
229 |
numpy_original_state = np.random.get_state()
|
230 |
np.random.seed(1234)
|
231 |
|
232 |
-
for class_dir in
|
233 |
-
|
234 |
-
|
235 |
-
image_path
|
236 |
for image_path in class_dir.iterdir()
|
237 |
if image_path.name.endswith(".jpg")
|
238 |
]
|
239 |
# _TRAIN_POINTS_PER_CLASS datapoints are sampled for the train split,
|
240 |
# the others constitute the test split.
|
241 |
-
if _TRAIN_POINTS_PER_CLASS > len(
|
242 |
raise ValueError(
|
243 |
-
"Fewer than {} ({}) points in class {}"
|
244 |
-
_TRAIN_POINTS_PER_CLASS, len(fnames), class_dir.name
|
245 |
-
)
|
246 |
)
|
247 |
-
|
248 |
-
|
|
|
249 |
)
|
250 |
-
|
251 |
-
|
|
|
252 |
|
253 |
if (
|
254 |
-
|
255 |
and config_name == self._BUILDER_CONFIG_WITHOUT_BACKGROUND.name
|
256 |
):
|
257 |
print("skip BACKGROUND_Google")
|
258 |
continue
|
259 |
|
260 |
-
for
|
261 |
record = {
|
262 |
-
"image": str(
|
263 |
"label": class_dir.name.lower(),
|
264 |
}
|
265 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
266 |
# Resets the seeds to their previous states.
|
267 |
np.random.set_state(numpy_original_state)
|
|
|
14 |
"""Caltech 101 loading script"""
|
15 |
|
16 |
|
17 |
+
from __future__ import annotations
|
18 |
+
|
19 |
from pathlib import Path
|
20 |
|
21 |
import datasets
|
22 |
import numpy as np
|
23 |
+
import scipy.io
|
24 |
from datasets.tasks import ImageClassification
|
25 |
|
26 |
_CITATION = """\
|
|
|
150 |
"wrench",
|
151 |
"yin_yang",
|
152 |
]
|
153 |
+
# For some reason, the category names in "101_ObjectCategories" and
|
154 |
+
# "Annotations" do not always match. This is a manual map between the
|
155 |
+
# two. Defaults to using same name, since most names are fine.
|
156 |
+
_ANNOTATION_NAMES_MAP = {
|
157 |
+
"Faces": "Faces_2",
|
158 |
+
"Faces_easy": "Faces_3",
|
159 |
+
"Motorbikes": "Motorbikes_16",
|
160 |
+
"airplanes": "Airplanes_Side_2",
|
161 |
+
}
|
162 |
|
163 |
_TRAIN_POINTS_PER_CLASS = 30
|
164 |
|
|
|
185 |
]
|
186 |
|
187 |
def _info(self):
|
188 |
+
if self.config.name == self._BUILDER_CONFIG_WITHOUT_BACKGROUND.name:
|
189 |
+
features = datasets.Features(
|
|
|
190 |
{
|
191 |
"image": datasets.Image(),
|
192 |
"label": datasets.features.ClassLabel(names=_NAMES),
|
193 |
+
"annotation": {
|
194 |
+
"obj_contour": datasets.features.Array2D(
|
195 |
+
shape=(2, None), dtype="float64"
|
196 |
+
),
|
197 |
+
"box_coord": datasets.features.Array2D(
|
198 |
+
shape=(1, 4), dtype="int64"
|
199 |
+
),
|
200 |
+
},
|
201 |
}
|
202 |
+
)
|
203 |
+
else:
|
204 |
+
features = datasets.Features(
|
205 |
+
{
|
206 |
+
"image": datasets.Image(),
|
207 |
+
"label": datasets.features.ClassLabel(names=_NAMES),
|
208 |
+
}
|
209 |
+
)
|
210 |
+
return datasets.DatasetInfo(
|
211 |
+
description=_DESCRIPTION,
|
212 |
+
features=features,
|
213 |
homepage=_HOMEPAGE,
|
214 |
license=_LICENSE,
|
215 |
citation=_CITATION,
|
|
|
|
|
|
|
216 |
)
|
217 |
|
218 |
def _split_generators(self, dl_manager):
|
219 |
data_root_dir = dl_manager.download_and_extract(_DATA_URL)
|
220 |
+
img_folder_compress_path = [
|
221 |
file
|
222 |
for file in dl_manager.iter_files(data_root_dir)
|
223 |
if Path(file).name == "101_ObjectCategories.tar.gz"
|
224 |
][0]
|
225 |
+
annotations_folder_compress_path = [
|
226 |
+
file
|
227 |
+
for file in dl_manager.iter_files(data_root_dir)
|
228 |
+
if Path(file).name == "Annotations.tar"
|
229 |
+
][0]
|
230 |
+
img_dir = dl_manager.extract(img_folder_compress_path)
|
231 |
+
annotation_dir = dl_manager.extract(annotations_folder_compress_path)
|
232 |
return [
|
233 |
datasets.SplitGenerator(
|
234 |
name=datasets.Split.TRAIN,
|
235 |
gen_kwargs={
|
236 |
+
"img_dir": Path(img_dir) / "101_ObjectCategories",
|
237 |
+
"annotation_dir": Path(annotation_dir) / "Annotations",
|
238 |
"split": "train",
|
239 |
"config_name": self.config.name,
|
240 |
},
|
|
|
242 |
datasets.SplitGenerator(
|
243 |
name=datasets.Split.TEST,
|
244 |
gen_kwargs={
|
245 |
+
"img_dir": Path(img_dir) / "101_ObjectCategories",
|
246 |
+
"annotation_dir": Path(annotation_dir) / "Annotations",
|
247 |
"split": "test",
|
248 |
"config_name": self.config.name,
|
249 |
},
|
250 |
),
|
251 |
]
|
252 |
|
253 |
+
def _generate_examples(self, img_dir, annotation_dir, split, config_name):
|
254 |
# Same stratagy as the one proposed in TF datasets: 30 random examples from each class are added to the train
|
255 |
# split, and the remainder are added to the test split.
|
256 |
# Source: https://github.com/tensorflow/datasets/blob/1106d587f97c4fca68c5b593dc7dc48c790ffa8c/tensorflow_datasets/image_classification/caltech.py#L88-L140
|
257 |
|
258 |
is_train_split = split == "train"
|
259 |
+
|
260 |
# Sets random seed so the random partitioning of files is the same when
|
261 |
# called for the train and test splits.
|
262 |
numpy_original_state = np.random.get_state()
|
263 |
np.random.seed(1234)
|
264 |
|
265 |
+
for class_dir in img_dir.iterdir():
|
266 |
+
class_name = class_dir.name
|
267 |
+
index_codes = [
|
268 |
+
image_path.name.split("_")[1][: -len(".jpg")]
|
269 |
for image_path in class_dir.iterdir()
|
270 |
if image_path.name.endswith(".jpg")
|
271 |
]
|
272 |
# _TRAIN_POINTS_PER_CLASS datapoints are sampled for the train split,
|
273 |
# the others constitute the test split.
|
274 |
+
if _TRAIN_POINTS_PER_CLASS > len(index_codes):
|
275 |
raise ValueError(
|
276 |
+
f"Fewer than {_TRAIN_POINTS_PER_CLASS} ({len(index_codes)}) points in class {class_dir.name}"
|
|
|
|
|
277 |
)
|
278 |
+
|
279 |
+
train_indices = np.random.choice(
|
280 |
+
index_codes, _TRAIN_POINTS_PER_CLASS, replace=False
|
281 |
)
|
282 |
+
test_indices = set(index_codes).difference(train_indices)
|
283 |
+
|
284 |
+
indices_to_emit = train_indices if is_train_split else test_indices
|
285 |
|
286 |
if (
|
287 |
+
class_name == "BACKGROUND_Google"
|
288 |
and config_name == self._BUILDER_CONFIG_WITHOUT_BACKGROUND.name
|
289 |
):
|
290 |
print("skip BACKGROUND_Google")
|
291 |
continue
|
292 |
|
293 |
+
for indice in indices_to_emit:
|
294 |
record = {
|
295 |
+
"image": str(class_dir / f"image_{indice}.jpg"),
|
296 |
"label": class_dir.name.lower(),
|
297 |
}
|
298 |
+
if config_name == self._BUILDER_CONFIG_WITHOUT_BACKGROUND.name:
|
299 |
+
if class_name in _ANNOTATION_NAMES_MAP:
|
300 |
+
annotations_class_name = _ANNOTATION_NAMES_MAP[class_name]
|
301 |
+
else:
|
302 |
+
annotations_class_name = class_name
|
303 |
+
data = scipy.io.loadmat(
|
304 |
+
str(
|
305 |
+
annotation_dir
|
306 |
+
/ annotations_class_name
|
307 |
+
/ f"annotation_{indice}.mat"
|
308 |
+
)
|
309 |
+
)
|
310 |
+
# raise ValueError(data["obj_contour"].dtype, data["box_coord"])
|
311 |
+
record["annotation"] = {
|
312 |
+
"obj_contour": data["obj_contour"],
|
313 |
+
"box_coord": data["box_coord"],
|
314 |
+
}
|
315 |
+
yield f"{class_dir.name.lower()}/{f'image_{indice}.jpg'}", record
|
316 |
# Resets the seeds to their previous states.
|
317 |
np.random.set_state(numpy_original_state)
|