def _predict(self, filenames):
images = read_images_for_torch(filenames).to(self.device)
labels = self.network.predict(images).astype('uint8')
infos = [
FilenameParser.get_imageinfo_from_filename(nm) for nm in filenames
]
return self.pack_into_annotated_images(infos, labels)
def _predict(self, filenames_and_boxes):
input_x = self._read_images_for_stage(filenames_and_boxes)
stage_probs = self.network.predict_proba(input_x)
infos = [
FilenameParser.get_imageinfo_from_filename(nm)
for nm, box in filenames_and_boxes]
return self.pack_into_annotated_images(infos, stage_probs)
def _grab_names_and_boxes_for_stages(
self, stage_smooth, valid_stages, names_and_boxes):
infos = [
FP.get_imageinfo_from_filename(nm) for nm, _ in names_and_boxes]
out = [
name_and_box
for name_and_box, info in zip(names_and_boxes, infos)
if stage_smooth[info].annotation in valid_stages]
return out
def test_create_from_filename_and_data(self):
filename = (
'/media/brian/Data/Dalit-Embryo-Images/training-data-folders/' +
'D2016_06_03_S1113_I313_pdb/WELL06/F0/168.jpg')
data = [1, 2, 4]
image = AnnotatedImage.create_from_filename_and_data(
filename, data)
true_info = FilenameParser.get_imageinfo_from_filename(filename)
self.assertEqual(image.info, true_info)
self.assertEqual(image.annotation, data)
def _predict(self, filenames_and_boxes):
boxes = [box for filename, box in filenames_and_boxes]
filenames = [filename for filename, box in filenames_and_boxes]
outputs_raw = [
self._predict_single_timepoint_raw(*name_and_box)
for name_and_box in filenames_and_boxes
]
output = self._transform_to_final_output(outputs_raw, boxes)
infos = [
FilenameParser.get_imageinfo_from_filename(nm) for nm in filenames
]
return self.pack_into_annotated_images(infos, output)
def test_grab_1cell_names_and_boxes(self):
stages = make_fake_smoothed_stages()
box = (1, 1, 100, 100)
names_and_boxes = [
(FP.get_partial_filename_from_imageinfo(i.info), box)
for i in stages.iterate_over_images()]
p = Pipeline('cpu')
onecell_names_boxes = p._grab_1cell_names_and_boxes(
stages, names_and_boxes)
images_in_onecell_stage = [
i for i in stages.iterate_over_images() if i.annotation == 1]
for name, _ in onecell_names_boxes:
info = FP.get_imageinfo_from_filename(name)
self.assertEqual(stages[info].annotation, 1)
self.assertEqual(
len(images_in_onecell_stage),
len(onecell_names_boxes))
def test_grab_cleavage_names_and_boxes(self):
stages = make_fake_smoothed_stages()
box = (1, 1, 100, 100)
names_and_boxes = [
(FP.get_partial_filename_from_imageinfo(i.info), box)
for i in stages.iterate_over_images()]
p = Pipeline('cpu')
cleavage_names_boxes = p._grab_cleavage_names_and_boxes(
stages, names_and_boxes)
cleavage_stages = {1, 2, 3, 4, 5, 6, 7, 8}
images_in_cleavage_stage = [
i for i in stages.iterate_over_images()
if i.annotation in cleavage_stages]
for name, _ in cleavage_names_boxes:
info = FP.get_imageinfo_from_filename(name)
self.assertIn(stages[info].annotation, cleavage_stages)
self.assertEqual(
len(images_in_cleavage_stage),
len(cleavage_names_boxes))
def _predict(self, f0_filenames_and_boxes):
filenames = [nm for nm, box in f0_filenames_and_boxes]
boxes = [box for nm, box in f0_filenames_and_boxes]
augmented_names = [augment_focus(nm) for nm in filenames]
fm_filenames_and_boxes = [
(names[0], box)
for names, box in zip(augmented_names, boxes)]
fp_filenames_and_boxes = [
(names[2], box)
for names, box in zip(augmented_names, boxes)]
frag_f0 = self._predict_for_focus_level(f0_filenames_and_boxes)
frag_fm = self._predict_for_focus_level(fm_filenames_and_boxes)
frag_fp = self._predict_for_focus_level(fp_filenames_and_boxes)
frag_values = np.mean([frag_f0, frag_fm, frag_fp], axis=0)
infos = [
FilenameParser.get_imageinfo_from_filename(nm)
for nm in filenames]
return self.pack_into_annotated_images(infos, frag_values)
def test_get_imageinfo_from_filename(self):
true_info = INFO_F0
parser = FilenameParser(EMBRYO_FOLDER)
filename = parser.get_filename_from_imageinfo(true_info)
parsed_info = parser.get_imageinfo_from_filename(filename)
self.assertEqual(parsed_info, true_info)