def detect(cls, context: FormatDetectionContext) -> None:
annot_paths = context.require_files('**/*.xml')
for annot_path in annot_paths:
with context.probe_text_file(
annot_path, "must be a LabelMe annotation file",
) as f:
elem_parents = []
for event, elem in ElementTree.iterparse(f, events=('start', 'end')):
if event == 'start':
if elem_parents == [] and elem.tag != 'annotation':
raise Exception
if elem_parents == ['annotation', 'object'] \
and elem.tag in {'polygon', 'segm'}:
return
elem_parents.append(elem.tag)
elif event == 'end':
elem_parents.pop()
if elem_parents == ['annotation'] and elem.tag == 'object':
# If we got here, then we found an object with no
# polygon and no mask, so it's probably the wrong
# format.
raise Exception
def detect(cls, context: FormatDetectionContext) -> None:
with context.require_any():
for prefix in (VggFace2Path.BBOXES_FILE,
VggFace2Path.LANDMARKS_FILE):
with context.alternative():
context.require_file(
f'{VggFace2Path.ANNOTATION_DIR}/{prefix}*.csv')
def detect(
cls,
context: FormatDetectionContext,
) -> Optional[FormatDetectionConfidence]:
if not cls.find_sources_with_params(context.root_path):
context.fail("specific requirement information unavailable")
return FormatDetectionConfidence.LOW
def detect(cls, context: FormatDetectionContext):
with context.require_any():
for image_dir in MarsPath.IMAGE_DIR_PATTERNS:
with context.alternative():
context.require_file('/'.join([
MarsPath.SUBSET_DIR_PATTERN, image_dir,
image_dir + MarsPath.IMAGE_NAME_POSTFIX
]))
def detect(cls, context: FormatDetectionContext) -> None:
try:
next(
find_files(context.root_path, VIDEO_EXTENSIONS,
recursive=True))
return FormatDetectionConfidence.LOW
except StopIteration:
context.fail("No video files found in '%s'. "
"Checked extensions: %s" %
(context.root_path, ', '.join(VIDEO_EXTENSIONS)))
def detect(cls, context: FormatDetectionContext) -> None:
annot_file = context.require_file('*.xml')
with context.probe_text_file(
annot_file,
"must be an XML file with an \"annotations\" root element",
) as f:
_, root_elem = next(ElementTree.iterparse(f, events=('start', )))
if root_elem.tag != 'annotations':
raise Exception
def detect(cls, context: FormatDetectionContext) -> None:
annot_file = context.require_file('annotations/*.json')
with context.probe_text_file(
annot_file,
"must be a JSON object with \"categories\" "
"and \"items\" keys",
) as f:
contents = parse_json(f.read())
if not {'categories', 'items'} <= contents.keys():
raise Exception
def detect(cls, context: FormatDetectionContext) -> None:
annot_path = context.require_file('*/**/*.json')
with context.probe_text_file(
annot_path,
"must be a JSON object with an \"annotation\" key",
) as f:
contents = parse_json(f.read())
if not isinstance(contents, dict):
raise Exception
if 'annotation' not in contents:
raise Exception
def detect(cls, context: FormatDetectionContext) -> None:
annot_path = context.require_file('*/gt.txt')
with context.probe_text_file(
annot_path, 'must be a ICDAR-like annotation file',
) as f:
reader = csv.reader(f,
doublequote=False, escapechar='\\', skipinitialspace=True)
fields = next(reader)
if len(fields) != 2: raise Exception
if osp.splitext(fields[0])[1] not in IMAGE_EXTENSIONS:
raise Exception
def detect(cls, context: FormatDetectionContext) -> None:
annot_file = context.require_file('*.xml')
with context.probe_text_file(
annot_file, "must be a KITTI-like annotation file",
) as f:
parser = ET.iterparse(f, events=('start',))
_, elem = next(parser)
if elem.tag != 'boost_serialization':
raise Exception
_, elem = next(parser)
if elem.tag != 'tracklets':
raise Exception
def detect(cls, context: FormatDetectionContext) -> None:
annot_path = context.require_file('*.txt',
exclude_fnames=ImagenetTxtPath.LABELS_FILE)
with context.probe_text_file(
annot_path,
"must be an ImageNet-like annotation file",
) as f:
for line in f:
_, _, label_ids = _parse_annotation_line(line)
if label_ids: break
else:
# If there are no labels in the entire file, it's probably
# not actually an ImageNet file.
raise Exception
def detect(cls, context: FormatDetectionContext) -> None:
annot_path = context.require_file(
'*.txt', exclude_fnames=CamvidPath.LABELMAP_FILE)
with context.probe_text_file(
annot_path,
'must be a CamVid-like annotation file',
) as f:
for line in f:
_, mask_path = _parse_annotation_line(line)
if mask_path is not None:
break
else:
# If there are no masks in the entire file, it's probably
# not actually a CamVid file.
raise Exception
context.require_file(glob.escape(mask_path))
def detect(cls, context: FormatDetectionContext) -> None:
# The `voc` format is inherently ambiguous with `voc_classification`,
# `voc_detection`, etc. To remove the ambiguity (and thus make it
# possible to use autodetection with the VOC datasets), disable
# autodetection for the single-task formats.
if len(cls._TASKS) == 1:
context.raise_unsupported()
with context.require_any():
task_dirs = {task_dir for _, task_dir in cls._TASKS.values()}
for task_dir in sorted(task_dirs):
with context.alternative():
context.require_file(
osp.join(VocPath.SUBSETS_DIR, task_dir, '*.txt'))
def detect(
cls,
context: FormatDetectionContext,
) -> FormatDetectionConfidence:
# The `coco` format is inherently ambiguous with `coco_instances`,
# `coco_stuff`, etc. To remove the ambiguity (and thus make it possible
# to use autodetection with the COCO dataset), disable autodetection
# for the single-task formats.
if len(cls._TASKS) == 1:
context.raise_unsupported()
with context.require_any():
for task in cls._TASKS.keys():
with context.alternative():
context.require_file(f'annotations/{task.name}_*.json')
def detect(cls, context: FormatDetectionContext) -> None:
context.require_file(MpiiJsonPath.ANNOTATION_FILE)
def detect(cls, context: FormatDetectionContext) -> None:
context.require_file('*' + VottJsonPath.ANNO_FILE_SUFFIX)
def detect(cls, context: FormatDetectionContext) -> None:
with context.require_any():
for prefix in (SynthiaPath.IMAGES_DIR, SynthiaPath.LABELS_SEGM_DIR,
SynthiaPath.SEMANTIC_SEGM_DIR):
with context.alternative():
context.require_file(f'{prefix}/**/*.png')
def detect(cls, context: FormatDetectionContext) -> None:
context.require_file(
f'*/{LfwPath.ANNOTATION_DIR}/{LfwPath.PAIRS_FILE}')
def detect(cls, context: FormatDetectionContext) -> None:
context.require_file('obj.data')
def detect(cls, context: FormatDetectionContext) -> None:
gt_txt_path = context.require_file('**/*_GT.txt')
gt_bmp_path = osp.splitext(gt_txt_path)[0] + '.bmp'
context.require_file(glob.escape(gt_bmp_path))
def detect(cls, context: FormatDetectionContext) -> None:
context.require_file('**/gt_*.txt')
def detect(cls, context: FormatDetectionContext) -> None:
context.require_file(f'{WiderFacePath.ANNOTATIONS_DIR}/*.txt')
def detect(cls, context: FormatDetectionContext) -> None:
with context.require_any():
for pattern in cls.POSSIBLE_ANNOTATION_PATTERNS:
with context.alternative():
context.require_file(
f'{OpenImagesPath.ANNOTATIONS_DIR}/{pattern}')
