Coverage for src/bob/bio/face/database/ijbc.py: 35%
86 statements
« prev ^ index » next coverage.py v7.6.0, created at 2024-07-13 00:04 +0200
1import copy
2import logging
3import os
5from functools import partial
7import pandas as pd
9from clapper.rc import UserDefaults
11import bob.io.base
13from bob.bio.base.pipelines.abstract_classes import Database
14from bob.pipelines import hash_string
15from bob.pipelines.sample import DelayedSample, SampleSet
17logger = logging.getLogger(__name__)
18rc = UserDefaults("bobrc.toml")
21def _make_sample_from_template_row(row, image_directory):
22 # Appending this key, so we can handle parallel writing correctly
23 # paying the penalty of having duplicate checkpoint files
24 key = os.path.splitext(row["FILENAME"])[0] + "-" + str(row["SUBJECT_ID"])
26 return DelayedSample(
27 load=partial(
28 bob.io.base.load, os.path.join(image_directory, row["FILENAME"])
29 ),
30 template_id=str(row["TEMPLATE_ID"]),
31 subject_id=str(row["SUBJECT_ID"]),
32 key=key,
33 # gender=row["GENDER"],
34 # indoor_outdoor=row["INDOOR_OUTDOOR"],
35 # skintone=row["SKINTONE"],
36 # yaw=row["YAW"],
37 # rool=row["ROLL"],
38 # occ1=row["OCC1"],
39 # occ2=row["OCC2"],
40 # occ3=row["OCC3"],
41 # occ4=row["OCC4"],
42 # occ5=row["OCC5"],
43 # occ6=row["OCC6"],
44 # occ7=row["OCC7"],
45 # occ8=row["OCC8"],
46 # occ9=row["OCC9"],
47 # occ10=row["OCC10"],
48 # occ11=row["OCC11"],
49 # occ12=row["OCC12"],
50 # occ13=row["OCC13"],
51 # occ14=row["OCC14"],
52 # occ15=row["OCC15"],
53 # occ16=row["OCC16"],
54 # occ17=row["OCC17"],
55 # occ18=row["OCC18"],
56 annotations={
57 "topleft": (float(row["FACE_Y"]), float(row["FACE_X"])),
58 "bottomright": (
59 float(row["FACE_Y"]) + float(row["FACE_HEIGHT"]),
60 float(row["FACE_X"]) + float(row["FACE_WIDTH"]),
61 ),
62 "size": (float(row["FACE_HEIGHT"]), float(row["FACE_WIDTH"])),
63 },
64 )
67def _make_sample_set_from_template_group(template_group, image_directory):
68 samples = list(
69 template_group.apply(
70 _make_sample_from_template_row,
71 axis=1,
72 image_directory=image_directory,
73 )
74 )
75 return SampleSet(
76 samples,
77 template_id=samples[0].template_id,
78 subject_id=samples[0].subject_id,
79 key=samples[0].template_id,
80 )
83class IJBCDatabase(Database): # TODO Make this a CSVDatabase?
84 """
86 This package contains the access API and descriptions for the IARPA Janus Benchmark C -- IJB-C database.
87 The actual raw data can be downloaded from the original web page: http://www.nist.gov/programs-projects/face-challenges (note that not everyone might be eligible for downloading the data).
89 Included in the database, there are list files defining verification as well as closed- and open-set identification protocols.
90 For verification, two different protocols are provided.
91 For the ``1:1`` protocol, gallery and probe templates are combined using several images and video frames for each subject.
92 Compared gallery and probe templates share the same gender and skin tone -- these have been matched to make the comparisons more realistic and difficult.
94 For closed-set identification, the gallery of the ``1:1`` protocol is used, while probes stem from either only images, mixed images and video frames, or plain videos.
95 For open-set identification, the same probes are evaluated, but the gallery is split into two parts, either of which is left out to provide unknown probe templates, i.e., probe templates with no matching subject in the gallery.
96 In any case, scores are computed between all (active) gallery templates and all probes.
98 The IJB-C dataset provides additional evaluation protocols for face detection and clustering, but these are (not yet) part of this interface.
101 .. warning::
103 To use this dataset protocol, you need to have the original files of the IJBC datasets.
104 Once you have it downloaded, please run the following command to set the path for Bob
106 .. code-block:: sh
108 bob config set bob.bio.face.ijbc.directory [IJBC PATH]
111 The code below allows you to fetch the gallery and probes of the "1:1" protocol.
113 .. code-block:: python
115 >>> from bob.bio.face.database import IJBCDatabase
116 >>> ijbc = IJBCDatabase(protocol="test1")
118 >>> # Fetching the gallery
119 >>> references = ijbc.references()
120 >>> # Fetching the probes
121 >>> probes = ijbc.probes()
123 """
125 name = "ijbc"
126 category = "face"
127 dataset_protocols_name = "ijbc.tar.gz"
128 dataset_protocols_urls = [
129 "https://www.idiap.ch/software/bob/databases/latest/face/ijbc-????.tar.gz",
130 "http://www.idiap.ch/software/bob/databases/latest/face/ijbc-????.tar.gz",
131 ]
132 dataset_protocols_hash = "????"
134 def __init__(
135 self,
136 protocol,
137 original_directory=rc.get("bob.bio.face.ijbc.directory"),
138 **kwargs,
139 ):
140 import warnings
142 warnings.warn(
143 f"The {self.name} database is not yet adapted to this version of bob. Please port it or ask for it to be ported (This one actually needs to be converted to a CSVDatabase).",
144 DeprecationWarning,
145 )
147 if original_directory is None or not os.path.exists(original_directory):
148 raise ValueError(
149 f"Invalid or non existent `original_directory`: {original_directory}"
150 )
152 self._check_protocol(protocol)
153 super().__init__(
154 protocol=protocol,
155 annotation_type="bounding-box",
156 fixed_positions=None,
157 memory_demanding=True,
158 )
160 self.image_directory = os.path.join(original_directory, "images")
161 self.protocol_directory = os.path.join(original_directory, "protocols")
162 self._cached_probes = None
163 self._cached_references = None
164 self.hash_fn = hash_string
166 self._load_metadata(protocol)
168 # For the test4 protocols
169 if "test4" in protocol:
170 self.score_all_vs_all = True
172 def _load_metadata(self, protocol):
173 # Load CSV files
174 if protocol == "test1" or protocol == "test2":
175 self.reference_templates = pd.read_csv(
176 os.path.join(
177 self.protocol_directory, protocol, "enroll_templates.csv"
178 )
179 )
181 self.probe_templates = pd.read_csv(
182 os.path.join(
183 self.protocol_directory, protocol, "verif_templates.csv"
184 )
185 )
187 self.matches = pd.read_csv(
188 os.path.join(self.protocol_directory, protocol, "match.csv"),
189 names=["ENROLL_TEMPLATE_ID", "VERIF_TEMPLATE_ID"],
190 ).astype("str")
192 # TODO: temporarily disabling the metadata
193 """
194 self.metadata = pd.read_csv(
195 os.path.join(self.protocol_directory, "ijbc_metadata_with_age.csv"),
196 usecols=[
197 "SUBJECT_ID",
198 "FILENAME",
199 "FACE_X",
200 "FACE_Y",
201 "FACE_WIDTH",
202 "FACE_HEIGHT",
203 "SIGHTING_ID",
204 "FACIAL_HAIR",
205 "AGE",
206 "INDOOR_OUTDOOR",
207 "SKINTONE",
208 "GENDER",
209 "YAW",
210 "ROLL",
211 ]
212 + [f"OCC{i}" for i in range(1, 19)],
213 )
215 # LEFT JOIN WITH METADATA
216 self.probe_templates = pd.merge(
217 self.probe_templates,
218 self.metadata,
219 on=[
220 "SUBJECT_ID",
221 "FILENAME",
222 "FACE_X",
223 "FACE_Y",
224 "FACE_WIDTH",
225 "FACE_HEIGHT",
226 ],
227 how="left",
228 )
230 # LEFT JOIN WITH METADATA
231 self.reference_templates = pd.merge(
232 self.reference_templates,
233 self.metadata,
234 on=[
235 "SUBJECT_ID",
236 "FILENAME",
237 "FACE_X",
238 "FACE_Y",
239 "FACE_WIDTH",
240 "FACE_HEIGHT",
241 ],
242 how="left",
243 )
244 """
246 elif "test4" in protocol:
247 gallery_file = (
248 "gallery_G1.csv" if "G1" in protocol else "gallery_G2.csv"
249 )
251 self.reference_templates = pd.read_csv(
252 os.path.join(self.protocol_directory, "test4", gallery_file)
253 )
255 self.probe_templates = pd.read_csv(
256 os.path.join(self.protocol_directory, "test4", "probes.csv")
257 )
259 self.matches = None
261 else:
262 raise ValueError(
263 f"Protocol `{protocol}` not supported. We do accept merge requests :-)"
264 )
266 def background_model_samples(self):
267 return None
269 def probes(self, group="dev"):
270 self._check_group(group)
271 if self._cached_probes is None:
272 logger.info(
273 "Loading probes. This operation might take some minutes"
274 )
276 self._cached_probes = list(
277 self.probe_templates.groupby("TEMPLATE_ID").apply(
278 _make_sample_set_from_template_group,
279 image_directory=self.image_directory,
280 )
281 )
283 # Wiring probes with references
284 if self.protocol == "test1" or self.protocol == "test2":
285 # Link probes to the references they have to be compared with
286 # We might make that faster if we manage to write it as a Panda instruction
287 grouped_matches = self.matches.groupby("VERIF_TEMPLATE_ID")
288 for probe_sampleset in self._cached_probes:
289 probe_sampleset.references = list(
290 grouped_matches.get_group(probe_sampleset.template_id)[
291 "ENROLL_TEMPLATE_ID"
292 ]
293 )
294 elif "test4" in self.protocol:
295 references = [s.template_id for s in self.references()]
296 # You compare with all biometric references
297 for probe_sampleset in self._cached_probes:
298 probe_sampleset.references = copy.deepcopy(references)
299 pass
301 else:
302 raise ValueError(f"Invalid protocol: {self.protocol}")
304 return self._cached_probes
306 def references(self, group="dev"):
307 self._check_group(group)
308 if self._cached_references is None:
309 logger.info(
310 "Loading templates. This operation might take some minutes"
311 )
313 self._cached_references = list(
314 self.reference_templates.groupby("TEMPLATE_ID").apply(
315 _make_sample_set_from_template_group,
316 image_directory=self.image_directory,
317 )
318 )
320 return self._cached_references
322 def all_samples(self, group="dev"):
323 self._check_group(group)
325 return self.references() + self.probes()
327 def groups(self):
328 return ["dev"]
330 def protocols(self):
331 return ["test1", "test2", "test4-G1", "test4-G2"]
333 def _check_protocol(self, protocol):
334 assert (
335 protocol in self.protocols()
336 ), "Invalid protocol `{}` not in {}".format(protocol, self.protocols())
338 def _check_group(self, group):
339 assert group in self.groups(), "Invalid group `{}` not in {}".format(
340 group, self.groups()
341 )