diff --git a/model_server/extensions/chaeo/zmask.py b/model_server/extensions/chaeo/zmask.py
index a457516b17cad744797af6305aa444f34d514e66..f77ac431ab9cd5fc67f6ef5a83809e5153df75da 100644
--- a/model_server/extensions/chaeo/zmask.py
+++ b/model_server/extensions/chaeo/zmask.py
@@ -32,9 +32,41 @@ class RoiSet(object):
acc_raw: GenericImageDataAccessor,
params: RoiSetMetaParams = RoiSetMetaParams(),
):
- self.zmask_meta, self.df, self.interm = build_zmask_from_object_mask(
+ # parse filters
+ filters = params.filters
+ query_str = 'label > 0' # always true
+ if filters is not None:
+ for k, val in filters.dict(exclude_unset=True).items():
+ assert k in ('area', 'solidity')
+ vmin = val['min']
+ vmax = val['max']
+ assert vmin >= 0
+ query_str = query_str + f' & {k} > {vmin} & {k} < {vmax}'
+
+ # build dataframe of objects, assign z index to each object
+ argmax = acc_raw.data.argmax(axis=3, keepdims=True)[:, :, 0, 0].astype('uint16')
+ df = (
+ pd.DataFrame(
+ regionprops_table(
+ acc_obj_ids,
+ intensity_image=argmax,
+ properties=('label', 'area', 'intensity_mean', 'solidity', 'bbox', 'centroid')
+ )
+ )
+ .rename(
+ columns={'bbox-0': 'y0', 'bbox-1': 'x0', 'bbox-2': 'y1', 'bbox-3': 'x1',}
+ )
+ )
+ df['zi'] = df['intensity_mean'].round().astype('int')
+ df['keeper'] = False
+ df.loc[df.query(query_str).index, 'keeper'] = True
+ self.df = df
+
+ # remaining zmask_meta write ops
+ self.zmask_meta, _, self.interm = build_zmask_from_object_mask(
acc_obj_ids,
acc_raw,
+ df,
params=params,
)
self.acc_obj_ids = acc_obj_ids
@@ -61,6 +93,19 @@ class RoiSet(object):
def get_object_mask_by_class(self, class_id):
return self.object_id_labels == class_id
+
+
+ # def loc_mask(self, i):
+ # # compute contours
+ # # obmask = (lamap == ob.label) # TODO: on-the-fly
+ # # contour = find_contours(obmask) # TODO: on-the-fly
+ # # mask = obmask[ob.y0: ob.y1, ob.x0: ob.x1]
+ # ob = self.df.loc[i]
+ # return
+ #
+ # def loc_contour(self, i):
+ # pass
+
def get_patch_masks(self, **kwargs) -> MonoPatchStack:
return get_patch_masks(self, **kwargs)
@@ -177,68 +222,40 @@ class RoiSet(object):
def build_zmask_from_object_mask(
obmask: GenericImageDataAccessor,
zstack: GenericImageDataAccessor,
+ df,
params: RoiSetMetaParams = RoiSetMetaParams(),
):
- """
- Given a 2D mask of objects, build a 3D mask, where each object's z-position is determined by the index of
- maximum intensity in z. Return this zmask and a list of each object's meta information.
- :param obmask: GenericImageDataAccessor 2D map of objects IDs
- :param zstack: GenericImageDataAccessor monochrome zstack of same Y, X dimension as obmask
- :param params: RoiSetMetaParams
- filters: dictionary of form {attribute: (min, max)}; valid attributes are 'area' and 'solidity'
- mask_type: if 'boxes', zmask is True in each object's complete bounding box; otherwise 'contours'
- expand_box_by: (xy, z) expands bounding box by (xy, z) pixels except where this hits a boundary
- :return: tuple (zmask, meta)
- np.ndarray:
- boolean mask of same size as stack
- List containing one Dict per object, with keys:
- info: object's properties from skimage.measure.regionprops_table, including bounding box (y0, y1, x0, x1)
- slice: named slice (np.s_) of (optionally) expanded bounding box
- relative_bounding_box: bounding box (y0, y1, x0, x1) in relative frame of (optionally) expanded bounding box
- contour: object's contour returned by skimage.measure.find_contours
- mask: mask of object in relative frame of (optionally) expanded bounding box
- pd.DataFrame: objects, including bounding, box information after filtering
- Dict of intermediate image products:
- label_map: np.ndarray (h x w) where each unique object has an integer label
- argmax: np.ndarray (h x w x 1 x 1) z-index of highest intensity in zstack
- """
+ # """
+ # Given a 2D mask of objects, build a 3D mask, where each object's z-position is determined by the index of
+ # maximum intensity in z. Return this zmask and a list of each object's meta information.
+ # :param obmask: GenericImageDataAccessor 2D map of objects IDs
+ # :param zstack: GenericImageDataAccessor monochrome zstack of same Y, X dimension as obmask
+ # :param params: RoiSetMetaParams
+ # filters: dictionary of form {attribute: (min, max)}; valid attributes are 'area' and 'solidity'
+ # mask_type: if 'boxes', zmask is True in each object's complete bounding box; otherwise 'contours'
+ # expand_box_by: (xy, z) expands bounding box by (xy, z) pixels except where this hits a boundary
+ # :return: tuple (zmask, meta)
+ # np.ndarray:
+ # boolean mask of same size as stack
+ # List containing one Dict per object, with keys:
+ # info: object's properties from skimage.measure.regionprops_table, including bounding box (y0, y1, x0, x1)
+ # slice: named slice (np.s_) of (optionally) expanded bounding box
+ # relative_bounding_box: bounding box (y0, y1, x0, x1) in relative frame of (optionally) expanded bounding box
+ # contour: object's contour returned by skimage.measure.find_contours
+ # mask: mask of object in relative frame of (optionally) expanded bounding box
+ # pd.DataFrame: objects, including bounding, box information after filtering
+ # Dict of intermediate image products:
+ # label_map: np.ndarray (h x w) where each unique object has an integer label
+ # argmax: np.ndarray (h x w x 1 x 1) z-index of highest intensity in zstack
+ # """
filters = params.filters
expand_box_by = params.expand_box_by
# validate inputs
assert zstack.hw == obmask.shape
lamap = obmask
- query_str = 'label > 0' # always true
- if filters is not None:
- for k, val in filters.dict(exclude_unset=True).items():
- assert k in ('area', 'solidity')
- vmin = val['min']
- vmax = val['max']
- assert vmin >= 0
- query_str = query_str + f' & {k} > {vmin} & {k} < {vmax}'
-
- # build dataframe of objects, assign z index to each object
+
argmax = zstack.data.argmax(axis=3, keepdims=True)[:, :, 0, 0].astype('uint16')
- df = (
- pd.DataFrame(
- regionprops_table(
- lamap,
- intensity_image=argmax,
- properties=('label', 'area', 'intensity_mean', 'solidity', 'bbox', 'centroid')
- )
- )
- .rename(
- columns={
- 'bbox-0': 'y0',
- 'bbox-1': 'x0',
- 'bbox-2': 'y1',
- 'bbox-3': 'x1',
- }
- )
- )
- df['zi'] = df['intensity_mean'].round().astype('int')
- df['keeper'] = False
- df.loc[df.query(query_str).index, 'keeper'] = True
# convert bounding boxes to numpy slice objects
ebxy, ebz = expand_box_by