def __init__(self, path):
super().__init__(path)
subsets = {}
annotations = defaultdict(dict)
task_dir = osp.join(path, 'Segmentation')
if not osp.isdir(task_dir):
return
ann_files = os.listdir(task_dir)
for ann_dir in ann_files:
ann_parts = filter(None, ann_dir.strip().split('_'))
if len(ann_parts) != 4:
continue
_, subset_name, mark = ann_parts
if mark not in ['cls', 'inst']:
continue
item_dir = osp.join(task_dir, ann_dir)
items = dir_items(item_dir, '.png', truncate_ext=True)
items = { name: osp.join(item_dir, item + '.png') \
for name, item in items }
subset = VocResultsExtractor.Subset(subset_name, self)
subset.items = list(items)
subsets[subset_name] = subset
annotations[subset_name][mark] = items
self._subsets = subsets
self._annotations = dict(annotations)
def _parse_txt_ann(self, path, subsets, annotations, task):
task_desc = self._SUPPORTED_TASKS[task]
task_dir = osp.join(path, task_desc['dir'])
ann_ext = task_desc['ext']
if not osp.isdir(task_dir):
return
ann_files = dir_items(task_dir, ann_ext, truncate_ext=True)
for ann_file in ann_files:
ann_parts = filter(None, ann_file.strip().split('_'))
if len(ann_parts) != 4:
continue
_, mark, subset_name, label = ann_parts
if mark != task_desc['mark']:
continue
label_id = VocLabel[label].value
anns = defaultdict(list)
with open(osp.join(task_dir, ann_file + ann_ext), 'r') as f:
for line in f:
line_parts = line.split()
item = line_parts[0]
anns[item].append((label_id, *line_parts[1:]))
subset = VocResultsExtractor.Subset(subset_name, self)
subset.items = list(anns)
subsets[subset_name] = subset
annotations[subset_name] = dict(anns)
def _find_subsets(path):
anno_dir = osp.join(path, DatumaroPath.ANNOTATIONS_DIR)
if not osp.isdir(anno_dir):
raise Exception('Datumaro dataset not found at "%s"' % path)
return { name: osp.join(anno_dir, name + '.json')
for name in dir_items(anno_dir, '.json', truncate_ext=True)
}
def _load_det_annotations(self):
det_anno_dir = osp.join(self._path, VocPath.ANNOTATIONS_DIR)
det_anno_items = dir_items(det_anno_dir, '.xml', truncate_ext=True)
det_annotations = dict()
for ann_item in det_anno_items:
with open(osp.join(det_anno_dir, ann_item + '.xml'), 'r') as f:
ann_file_data = f.read()
det_annotations[ann_item] = ann_file_data
self._annotations[VocTask.detection] = det_annotations
def _load_det_annotations(self):
det_anno_dir = osp.join(self._path, VocPath.ANNOTATIONS_DIR)
det_anno_items = dir_items(det_anno_dir, '.xml', truncate_ext=True)
det_annotations = dict()
for ann_item in det_anno_items:
with open(osp.join(det_anno_dir, ann_item + '.xml'), 'r') as f:
ann_file_data = f.read()
ann_file_root = ET.fromstring(ann_file_data)
item = ann_file_root.find('filename').text
item = osp.splitext(item)[0]
det_annotations[item] = ann_file_data
self._annotations[VocTask.detection] = det_annotations
def _load_subsets(self, subsets_dir):
dir_files = dir_items(subsets_dir, '.txt', truncate_ext=True)
subset_names = [s for s in dir_files if '_' not in s]
subsets = {}
for subset_name in subset_names:
subset = __class__.Subset(subset_name, self)
with open(osp.join(subsets_dir, subset_name + '.txt'), 'r') as f:
subset.items = [line.split()[0] for line in f]
subsets[subset_name] = subset
return subsets
def __init__(self, path):
super().__init__(path)
subsets = {}
annotations = defaultdict(dict)
task = VocTask.person_layout
task_desc = self._SUPPORTED_TASKS[task]
task_dir = osp.join(path, task_desc['dir'])
if not osp.isdir(task_dir):
return
ann_ext = task_desc['ext']
ann_files = dir_items(task_dir, ann_ext, truncate_ext=True)
for ann_file in ann_files:
ann_parts = filter(None, ann_file.strip().split('_'))
if len(ann_parts) != 4:
continue
_, mark, subset_name, _ = ann_parts
if mark != task_desc['mark']:
continue
layouts = {}
root = ET.parse(osp.join(task_dir, ann_file + ann_ext))
root_elem = root.getroot()
for layout_elem in root_elem.findall('layout'):
item = layout_elem.find('image').text
obj_id = int(layout_elem.find('object').text)
conf = float(layout_elem.find('confidence').text)
parts = []
for part_elem in layout_elem.findall('part'):
label_id = VocBodyPart[part_elem.find('class').text].value
bbox_elem = part_elem.find('bndbox')
xmin = float(bbox_elem.find('xmin').text)
xmax = float(bbox_elem.find('xmax').text)
ymin = float(bbox_elem.find('ymin').text)
ymax = float(bbox_elem.find('ymax').text)
bbox = [xmin, ymin, xmax - xmin, ymax - ymin]
parts.append((label_id, bbox))
layouts[item] = [obj_id, conf, parts]
subset = VocResultsExtractor.Subset(subset_name, self)
subset.items = list(layouts)
subsets[subset_name] = subset
annotations[subset_name] = layouts
self._subsets = subsets
self._annotations = dict(annotations)
def _load_cls_annotations(self, subsets_dir, subset_names):
dir_files = dir_items(subsets_dir, '.txt', truncate_ext=True)
label_annotations = defaultdict(list)
label_anno_files = [s for s in dir_files \
if '_' in s and s[s.rfind('_') + 1:] in subset_names]
for ann_file in label_anno_files:
with open(osp.join(subsets_dir, ann_file + '.txt'), 'r') as f:
label = ann_file[:ann_file.rfind('_')]
label_id = VocLabel[label].value
for line in f:
item, present = line.split()
if present == '1':
label_annotations[item].append(label_id)
self._annotations[VocTask.classification] = dict(label_annotations)
def _load_annotations(self):
annotations = defaultdict(list)
task_dir = osp.dirname(self._path)
anno_files = [
s for s in dir_items(task_dir, '.txt')
if s.endswith('_' + osp.basename(self._path))
]
for ann_filename in anno_files:
with open(osp.join(task_dir, ann_filename)) as f:
label = ann_filename[:ann_filename.rfind('_')]
label_id = self._get_label_id(label)
for line in f:
item, present = line.split()
if present == '1':
annotations[item].append(label_id)
return dict(annotations)
def _load_cls_annotations(self, subsets_dir, subset_names):
subset_file_names = [
n if n else DEFAULT_SUBSET_NAME for n in subset_names
]
dir_files = dir_items(subsets_dir, '.txt', truncate_ext=True)
label_annotations = defaultdict(list)
label_anno_files = [s for s in dir_files \
if '_' in s and s[s.rfind('_') + 1:] in subset_file_names]
for ann_filename in label_anno_files:
with open(osp.join(subsets_dir, ann_filename + '.txt'), 'r') as f:
label = ann_filename[:ann_filename.rfind('_')]
label_id = self._get_label_id(label)
for line in f:
item, present = line.split()
if present == '1':
label_annotations[item].append(label_id)
self._annotations[VocTask.classification] = dict(label_annotations)
def _load_subsets(self, subsets_dir):
dir_files = dir_items(subsets_dir, '.txt', truncate_ext=True)
subset_names = [s for s in dir_files if '_' not in s]
subsets = {}
for subset_name in subset_names:
subset_file_name = subset_name
if subset_name == DEFAULT_SUBSET_NAME:
subset_name = None
subset = __class__.Subset(subset_name, self)
subset.items = []
with open(osp.join(subsets_dir, subset_file_name + '.txt'),
'r') as f:
for line in f:
line = line.split()[0].strip()
if line:
subset.items.append(line)
subsets[subset_name] = subset
return subsets
