def apply(self):
extractor = self._extractor
images_dir = osp.join(self._save_dir, MotPath.IMAGE_DIR)
os.makedirs(images_dir, exist_ok=True)
self._images_dir = images_dir
anno_dir = osp.join(self._save_dir, 'gt')
os.makedirs(anno_dir, exist_ok=True)
anno_file = osp.join(anno_dir, MotPath.GT_FILENAME)
with open(anno_file, 'w', encoding="utf-8") as csv_file:
writer = csv.DictWriter(csv_file, fieldnames=MotPath.FIELDS)
track_id_mapping = {-1: -1}
for idx, item in enumerate(extractor):
log.debug("Converting item '%s'", item.id)
frame_id = cast(item.id, int, 1 + idx)
for anno in item.annotations:
if anno.type != AnnotationType.bbox:
continue
track_id = int(anno.attributes.get('track_id', -1))
if track_id not in track_id_mapping:
track_id_mapping[track_id] = len(track_id_mapping)
track_id = track_id_mapping[track_id]
writer.writerow({
'frame_id': frame_id,
'track_id': track_id,
'x': anno.x,
'y': anno.y,
'w': anno.w,
'h': anno.h,
'confidence': int(anno.attributes.get('ignored') != True),
'class_id': 1 + cast(anno.label, int, -2),
'visibility': float(
anno.attributes.get('visibility',
1 - float(
anno.attributes.get('occluded', False)
)
)
)
})
if self._save_images:
if item.has_image and item.image.has_data:
self._save_image(item, osp.join(self._images_dir,
'%06d%s' % (frame_id, self._find_image_ext(item))))
else:
log.debug("Item '%s' has no image", item.id)
labels_file = osp.join(anno_dir, MotPath.LABELS_FILE)
with open(labels_file, 'w', encoding='utf-8') as f:
f.write('\n'.join(l.name
for l in extractor.categories()[AnnotationType.label])
)
def _convert_label_categories(self, obj):
converted = {
'labels': [],
}
for label in obj.items:
converted['labels'].append({
'name': cast(label.name, str),
'parent': cast(label.parent, str),
})
return converted
def _convert_annotation(self, obj):
assert isinstance(obj, Annotation)
ann_json = {
'id': cast(obj.id, int),
'type': cast(obj.type.name, str),
'attributes': obj.attributes,
'group': cast(obj.group, int, 0),
}
return ann_json
def _parse_attr(cls, value):
if value == 'true':
return True
elif value == 'false':
return False
elif str(cast(value, int, 0)) == value:
return int(value)
elif str(cast(value, float, 0)) == value:
return float(value)
else:
return value
def save_categories(self, dataset):
label_categories = dataset.categories().get(AnnotationType.label)
if label_categories is None:
return
for idx, cat in enumerate(label_categories.items):
self.categories.append({
'id': 1 + idx,
'name': cast(cat.name, str, ''),
'supercategory': cast(cat.parent, str, ''),
})
def _is_float(value):
if isinstance(value, str):
casted = cast(value, float)
if casted is not None:
if cast(casted, str) == value:
return True
return False
elif isinstance(value, float):
cast(value, float)
return True
return False
def save_categories(self, dataset):
label_categories = dataset.categories().get(AnnotationType.label)
if label_categories is None:
return
for idx, cat in enumerate(label_categories.items):
self.categories.append({
'id': 1 + idx,
'name': cast(cat.name, str, ''),
'supercategory': cast(cat.parent, str, ''),
'isthing': 0, # TODO: can't represent this information yet
})
def _parse_attributes(attr_str):
parsed = []
if not attr_str:
return parsed
for attr in [a.strip() for a in cls._escape(attr_str).split(',')]:
if not attr:
continue
if '=' in attr:
name, value = attr.split('=', maxsplit=1)
if value.lower() in {'true', 'false'}:
value = value.lower() == 'true'
elif 1 < len(value) and value[0] == '"' and value[-1] == '"':
value = value[1:-1]
else:
for t in [int, float]:
casted = cast(value, t)
if casted is not None and str(casted) == value:
value = casted
break
if isinstance(value, str):
value = cls._unescape(value)
parsed.append((cls._unescape(name), value))
else:
parsed.append((cls._unescape(attr), True))
return parsed
def _write_item(self, item, index):
if not self._context._reindex:
index = cast(item.attributes.get('frame'), int, index)
image_info = OrderedDict([
("id", str(index)),
])
filename = item.id + CvatPath.IMAGE_EXT
image_info["name"] = filename
if item.has_image:
size = item.image.size
if size:
h, w = size
image_info["width"] = str(w)
image_info["height"] = str(h)
if self._context._save_images:
self._context._save_image(
item, osp.join(self._context._images_dir, filename))
else:
log.debug("Item '%s' has no image info", item.id)
self._writer.open_image(image_info)
for ann in item.annotations:
if ann.type in {
AnnotationType.points, AnnotationType.polyline,
AnnotationType.polygon, AnnotationType.bbox
}:
self._write_shape(ann)
elif ann.type == AnnotationType.label:
self._write_tag(ann)
else:
continue
self._writer.close_image()
def _convert_label_object(self, obj):
converted = self._convert_annotation(obj)
converted.update({
'label_id': cast(obj.label, int),
})
return converted
def _write_item(self, item, index):
image_info = OrderedDict([
("id", str(cast(item.id, int, index))),
])
if item.has_image:
size = item.image.size
if size:
h, w = size
image_info["width"] = str(w)
image_info["height"] = str(h)
filename = item.image.filename
if self._context._save_images:
filename = self._save_image(item)
image_info["name"] = filename
else:
log.debug("Item '%s' has no image info" % item.id)
self._writer.open_image(image_info)
for ann in item.annotations:
if ann.type in {AnnotationType.points, AnnotationType.polyline,
AnnotationType.polygon, AnnotationType.bbox}:
self._write_shape(ann)
elif ann.type == AnnotationType.label:
self._write_tag(ann)
else:
continue
self._writer.close_image()
def _convert_caption_object(self, obj):
converted = self._convert_annotation(obj)
converted.update({
'caption': cast(obj.caption, str),
})
return converted
def _get_image_id(self, item):
image_id = self._image_ids.get(item.id)
if image_id is None:
image_id = cast(item.attributes.get('id'), int,
len(self._image_ids) + 1)
self._image_ids[item.id] = image_id
return image_id
def _write_item_annotations(self, item):
key_id_data = self._key_id_data
item_id = cast(item.attributes.get('frame'), int)
if item_id is None or self._context._reindex:
item_id = len(key_id_data['videos']) + 1
item_key = str(uuid.uuid4())
key_id_data['videos'][item_key] = item_id
item_user_info = {
k: item.attributes.get(k, default_v)
for k, default_v in self._default_user_info.items()
}
item_ann_data = {
'description': item.attributes.get('description', ''),
'key': item_key,
'tags': [],
'objects': [],
'figures': [],
}
self._export_item_attributes(item, item_ann_data, item_user_info)
self._export_item_annotations(item, item_ann_data, item_user_info)
ann_path = osp.join(self._ann_dir, item.id + '.pcd.json')
os.makedirs(osp.dirname(ann_path), exist_ok=True)
dump_json_file(ann_path, item_ann_data, indent=True)
def match_dm_item(item, task_data, root_hint=None):
is_video = task_data.meta['task']['mode'] == 'interpolation'
frame_number = None
if frame_number is None and item.has_image:
frame_number = task_data.match_frame(item.image.path, root_hint)
if frame_number is None:
frame_number = task_data.match_frame(item.id, root_hint)
if frame_number is None:
frame_number = cast(item.attributes.get('frame', item.id), int)
if frame_number is None and is_video:
frame_number = cast(osp.basename(item.id)[len('frame_'):], int)
if not frame_number in task_data.frame_info:
raise Exception("Could not match item id: '%s' with any task frame" %
item.id)
return frame_number
def _convert_bbox_object(self, obj):
converted = self._convert_annotation(obj)
converted.update({
'label_id': cast(obj.label, int),
'bbox': [float(p) for p in obj.get_bbox()],
})
return converted
def _convert_polygon_object(self, obj):
converted = self._convert_annotation(obj)
converted.update({
'label_id': cast(obj.label, int),
'points': [float(p) for p in obj.points],
})
return converted
def _convert_points_object(self, obj):
converted = self._convert_annotation(obj)
converted.update({
'label_id': cast(obj.label, int),
'points': [float(p) for p in obj.points],
'visibility': [int(v.value) for v in obj.visibility],
})
return converted
def _convert_cuboid_3d_object(self, obj):
converted = self._convert_annotation(obj)
converted.update({
'label_id': cast(obj.label, int),
'position': [float(p) for p in obj.position],
'rotation': [float(p) for p in obj.rotation],
'scale': [float(p) for p in obj.scale]
})
return converted
def _save_image(self, item, index):
if item.image.filename:
frame_id = osp.splitext(item.image.filename)[0]
else:
frame_id = item.id
frame_id = cast(frame_id, int, index)
image_filename = '%06d%s' % (frame_id, MotPath.IMAGE_EXT)
save_image(osp.join(self._images_dir, image_filename),
item.image.data)
def _convert_shape_object(self, obj):
assert isinstance(obj, _Shape)
converted = self._convert_annotation(obj)
converted.update({
'label_id': cast(obj.label, int),
'points': [float(p) for p in obj.points],
'z_order': obj.z_order,
})
return converted
def convert_instance(self, instance, item):
ann, polygons, mask, bbox = instance
is_crowd = mask is not None
if is_crowd:
segmentation = {
'counts': list(int(c) for c in mask['counts']),
'size': list(int(c) for c in mask['size'])
}
else:
segmentation = [list(map(float, p)) for p in polygons]
area = 0
if segmentation:
if item.has_image:
h, w = item.image.size
else:
# NOTE: here we can guess the image size as
# it is only needed for the area computation
w = bbox[0] + bbox[2]
h = bbox[1] + bbox[3]
rles = mask_utils.frPyObjects(segmentation, h, w)
if is_crowd:
rles = [rles]
else:
rles = mask_utils.merge(rles)
area = mask_utils.area(rles)
else:
_, _, w, h = bbox
segmentation = []
area = w * h
elem = {
'id': self._get_ann_id(ann),
'image_id': self._get_image_id(item),
'category_id': cast(ann.label, int, -1) + 1,
'segmentation': segmentation,
'area': float(area),
'bbox':
[round(float(n), _COORDINATE_ROUNDING_DIGITS) for n in bbox],
'iscrowd': int(is_crowd),
}
if 'score' in ann.attributes:
try:
elem['score'] = float(ann.attributes['score'])
except Exception as e:
log.warning("Item '%s': failed to convert attribute "
"'score': %e" % (item.id, e))
if self._context._allow_attributes:
attrs = self._convert_attributes(ann)
if attrs:
elem['attributes'] = attrs
return elem
def _parse_label_count(s: str) -> Tuple[str, int]:
label, count = s.split(':', maxsplit=1)
count = cast(count, int, default=None)
if not label:
raise argparse.ArgumentError(None, "Class name cannot be empty")
if count is None or count < 0:
raise argparse.ArgumentError(None,
f"Class '{label}' count is invalid")
return label, count
def _convert_points_categories(self, obj):
converted = {
'items': [],
}
for label_id, item in obj.items.items():
converted['items'].append({
'label_id': int(label_id),
'labels': [cast(label, str) for label in item.labels],
'joints': [list(map(int, j)) for j in item.joints],
})
return converted
def save_annotations(self, item):
if not item.has_image:
return
ann_filename = item.id + CocoPath.PANOPTIC_EXT
segments_info = list()
masks = []
next_id = self._min_ann_id
for ann in item.annotations:
if ann.type != AnnotationType.mask:
continue
if not ann.id:
ann.id = next_id
next_id += 1
segment_info = {}
segment_info['id'] = ann.id
segment_info['category_id'] = cast(ann.label, int, -1) + 1
segment_info['area'] = float(ann.get_area())
segment_info['bbox'] = [float(p) for p in ann.get_bbox()]
segment_info['iscrowd'] = cast(ann.attributes.get("is_crowd"), int,
0)
segments_info.append(segment_info)
masks.append(ann)
if not masks:
return
pan_format = mask_tools.merge_masks((m.image, m.id) for m in masks)
save_image(osp.join(self._context._segmentation_dir, ann_filename),
mask_tools.index2bgr(pan_format),
create_dir=True)
elem = {
'image_id': self._get_image_id(item),
'file_name': ann_filename,
'segments_info': segments_info
}
self.annotations.append(elem)
def _convert_points_categories(self, obj):
converted = {
'items': [],
}
for label_id, item in obj.items.items():
converted['items'].append({
'label_id':
int(label_id),
'labels': [cast(label, str) for label in item.labels],
'adjacent': [int(v) for v in item.adjacent],
})
return converted
def save_categories(self, dataset):
label_categories = dataset.categories().get(AnnotationType.label)
if label_categories is None:
return
point_categories = dataset.categories().get(AnnotationType.points)
for idx, label_cat in enumerate(label_categories.items):
cat = {
'id': 1 + idx,
'name': cast(label_cat.name, str, ''),
'supercategory': cast(label_cat.parent, str, ''),
'keypoints': [],
'skeleton': [],
}
if point_categories is not None:
kp_cat = point_categories.items.get(idx)
if kp_cat is not None:
cat.update({
'keypoints': [str(l) for l in kp_cat.labels],
'skeleton': [list(map(int, j)) for j in kp_cat.joints],
})
self.categories.append(cat)
def generate_next_name(names, basename, sep='.', suffix='', default=None):
pattern = re.compile(r'%s(?:%s(\d+))?%s' % \
tuple(map(re.escape, [basename, sep, suffix])))
matches = [match for match in (pattern.match(n) for n in names) if match]
max_idx = max([cast(match[1], int, 0) for match in matches], default=None)
if max_idx is None:
if default is not None:
idx = sep + str(default)
else:
idx = ''
else:
idx = sep + str(max_idx + 1)
return basename + idx + suffix
def save_image_info(self, item, filename):
h = 0
w = 0
if item.has_image and item.image.size:
h, w = item.image.size
self._data['images'].append({
'id': self._get_image_id(item),
'width': int(w),
'height': int(h),
'file_name': cast(filename, str, ''),
'license': 0,
'flickr_url': '',
'coco_url': '',
'date_captured': 0,
})
def _convert_mask_object(self, obj):
converted = self._convert_annotation(obj)
if isinstance(obj, RleMask):
rle = obj.rle
else:
rle = mask_utils.encode(
np.require(obj.image, dtype=np.uint8, requirements='F'))
converted.update({
'label_id': cast(obj.label, int),
'rle': {
# serialize as compressed COCO mask
'counts': rle['counts'].decode('ascii'),
'size': list(int(c) for c in rle['size']),
}
})
return converted
