def __init__(self, name, image, annos=None, meta=None):
for anno in annos:
assert isinstance(anno, Anno)
assert meta is None or isinstance(meta, SampleMeta)
assert is_file(image)
self.name = name
self.path = image
_, self.ext = os.path.splitext(os.path.split(self.path)[1])
self.image = imread(image)
self.meta = meta
self.annos = [] if annos is None else annos
def is_fully_within_image(self, image):
"""
Estimate whether the bounding box is fully inside the image area.
Parameters
----------
image : (H,W,...) ndarray or tuple of int
Image dimensions to use.
If an ndarray, its shape will be used.
If a tuple, it is assumed to represent the image shape
and must contain at least two integers.
Returns
-------
bool
True if the bounding box is fully inside the image area. False otherwise.
"""
shape = imread(image).shape
height, width = shape[0:2]
return self.xmin >= 0 and self.xmax < width and self.ymin >= 0 and self.ymax < height
def _parse_anno_from_attach(attach):
annos = []
if attach is not None and "file_name" in attach:
mask_path = os.path.join(self.id2attachpath[id],
attach["file_name"])
img_mask = imread(mask_path, 0)
bboxes = self._get_bboxes_from_mask(img_mask)
for bbox in bboxes:
in_others = False
for _bbox in bboxes:
if BBox.init_from(bbox).is_in(_bbox):
in_others = True
break
if in_others:
continue
xmin, ymin, xmax, ymax = bbox
if xmax - xmin <= 2 or ymax - ymin <= 2:
continue
anno = Anno(bbox=BBox(xmin=xmin,
ymin=ymin,
xmax=xmax,
ymax=ymax,
label=label),
label=label,
color=self.color_map[label],
meta=AnnoMeta())
bin_mask = np.zeros_like(img_mask, np.uint8)
bin_mask[ymin:ymax + 1,
xmin:xmax + 1] = img_mask[ymin:ymax + 1,
xmin:xmax + 1]
if np.where(bin_mask != 0)[0].size == 0:
continue
bin_mask[np.where(bin_mask != 0)] = 1
anno.mask = Mask(bin_mask)
annos.append(anno)
return annos
def is_partly_within_image(self, image):
"""
Estimate whether the bounding box is at least partially inside the image area.
Parameters
----------
image : (H,W,...) ndarray or tuple of int
Image dimensions to use.
If an ndarray, its shape will be used.
If a tuple, it is assumed to represent the image shape
and must contain at least two integers.
Returns
-------
bool
True if the bounding box is at least partially inside the image area. False otherwise.
"""
shape = imread(image).shape
height, width = shape[0:2]
eps = np.finfo(np.float32).eps
img_bb = BBox(xmin=0, xmax=width - eps, ymin=0, ymax=height - eps)
return self.intersection(img_bb) is not None
def _get_bboxes_from_mask(self, mask_or_path):
mask = imread(mask_or_path, 0)
contours = measure.find_contours(mask, 0.5)
reshaped_contour = []
for contour in contours:
contour = np.flip(contour, axis=1)
s = contour.ravel().tolist()
assert len(s) % 2 == 0
s = [(s[i], s[i + 1]) for i in range(len(s)) if i % 2 == 0]
reshaped_contour.append(s)
bboxes = []
for i in range(len(reshaped_contour)):
contour = np.array(reshaped_contour[i])
xmin, xmax = int(np.min(contour[:, 0])) + 1, int(
np.max(contour[:, 0])) + 1
ymin, ymax = int(np.min(contour[:, 1])) + 1, int(
np.max(contour[:, 1])) + 1
bboxes.append([xmin, ymin, xmax, ymax])
return bboxes