def point_creator(refer_dir:str, name:str, dst:str, point_label=True, direct_mask=False, limit_pkl:str=None) -> None:
"""the refer_dir is the directory where all refer are;
the name is the name of the created .pkl file;
the dst is where the .pkl file will be saved;
if the limit_pkl is provided, only the files in limit_pkl will be used,
this .pkl files should follow the format of metric_path_collector.py"""
containor = {}
refers = path_list_creator(refer_dir)
if limit_pkl:
target = load_pickle(limit_pkl)
refers = [refer for refer in refers if get_name(refer) in target['labels']]
if point_label:
for label_p in refers:
points = point_detect(cv2.cvtColor(cv2.imread(label_p), cv2.COLOR_BGR2HSV))
containor[get_name(label_p)] = points
elif direct_mask:
for mask_p in refers:
containor[get_name(mask_p)] = mask_p
else:
for mask_p in refers:
mask = cv2.imread(mask_p, 0)
cnts = mask_to_cnts_watershed(mask, min_distance=14, for_real_mask=True)
points = [middle(c) for c in cnts]
containor[get_name(mask_p)] = points
save_dict(containor, name, dst)
def _loc_updator(self):
if self.cur >= self.length:
self.cur = 0
elif self.cur < 0:
self.cur = self.length - 1
self.loc = self.path_list[self.cur]
self.ID = get_name(self.loc)
def _temp_init(self):
img_dir = filedialog.askdirectory(
title='Select the directory where your slides are ...',
mustexist=True)
self.dst = join(dirname(img_dir), 'temp')
os.makedirs(self.dst, exist_ok=True)
self.path_list = path_list_creator(img_dir)
self.cur = 0
self.length = len(self.path_list)
self.loc = self.path_list[self.cur]
self.ID = get_name(self.loc)
def metric(self,
img_p: str,
label_p: str = None,
show_info=False,
show_img=True,
dst='.',
name=None,
show_mask=False,
target=15,
label_points=None) -> None:
"""true_positives = Correct objects
false_positives = Missed objects
false_negatives = Extra objects
(PPV), Precision = Σ True positive / Σ Predicted condition positive
Sensitivity, probability of detection = Σ True positive / Σ Condition positive
If it is not form dir, we show the outcome directly"""
if not name:
name = get_name(img_p)
if show_info:
print(name)
img = cv2.imread(img_p)
pred_cnts, pred_mask_img = self._mask_creator(img)
mask_img = None
if isinstance(label_points, list):
label_points = label_points
elif label_p:
mask_img = cv2.imread(label_p, 0)
possible_right_cnts, label_num = overlap(pred_cnts,
mask_img=mask_img,
label_points=label_points)
if show_img:
fit_out = mask_visualization(img,
possible_right_cnts,
method='box',
target=target)
cv2.imwrite(os.path.join(dst, name + '_fit.jpg'), fit_out)
pred_out = mask_visualization(img, pred_cnts, target=target)
cv2.imwrite(os.path.join(dst, name + '_pred.jpg'), pred_out)
if show_mask:
cv2.imwrite(os.path.join(dst, name + '_mask.jpg'), pred_mask_img)
values = stats(label_num, pred_cnts, possible_right_cnts)
self.metric_record[name] = values
if show_info:
print_dict = {}
print_dict.update(dict(zip(self.metric_keys.split(', '), values)))
print(print_dict)
def make_gif_from_dir(metric_dir_path, dst):
maid = defaultdict(dict)
path_list = path_list_creator(metric_dir_path)
for fp in path_list:
name = get_name(fp)
if '_pred' in fp:
maid[name]['predicted'] = fp
elif '_label' in fp:
maid[name]['labeled'] = fp
# elif '_fit' in fp:
# maid[name]['fit'] = fp
for f_name, fp_dict in maid.items():
if len(fp_dict) == 2:
out_p = join(dst, f_name + '.gif')
make_gif(fp_dict, out_p)
def train_test_info_creator(label_dir: str,
tiles_dir: str,
dst: str,
percent: float = 0.2,
test_dir=None) -> None:
"""label_dir is the directory where point labelled images are;
tiles_dir is the directory where all tiles are, including 'raw_imgs' and 'labels';
dst is the directory where the 'test.pkl' and the 'others.pkl' will be saved"""
all_names = names_creator(label_dir)
if test_dir:
test_names = [get_name(fn) for fn in os.listdir(test_dir)]
else:
test_names = names_separator(all_names, percent=percent)
test_family, other_family = from_names_to_path_list(test_names, tiles_dir)
save_dict(test_family, 'test', dst)
save_dict(other_family, 'others', dst)
return len(other_family['labels'])
def predict_from_imgs(self,
img_p_list,
result_dst,
visualize_dst=None,
show_mask=False,
target=15,
assistance=False):
# self.result.clear()
for img_p in img_p_list:
img = cv2.imread(img_p)
ID = get_name(img_p)
if ID in self.result:
continue
self.predict_from_img(img,
ID,
visualize_dst=visualize_dst,
show_mask=show_mask,
target=target,
assistance=assistance)
show_result(self.result, ['Counts'], title='Count', dst=result_dst)
def test_dict(img_dir:str) -> list:
return {get_name(fp):fp for fp in path_list_creator(img_dir)}
def names_creator(dir_path):
return [get_name(fn) for fn in os.listdir(dir_path)]