def _wrap_get_gd_image_multi_objs(mask_):
# [h, w, 1]
guide_image = array_kits.get_gd_image_multi_objs(
mask_,
obj_value=2,
partial=False,
with_fake_guides=False)[..., None]
return guide_image.astype(np.float32)
def after_run(self, run_context, run_values):
predictions = run_values.results
try:
self.features_val, self.labels_val = run_context.session.run(
[self.features, self.labels])
except errors_impl.OutOfRangeError:
self._save_guide()
self.guides.clear()
return run_context.request_stop()
else:
new_case = self.features_val["names"][0].decode("utf-8")
if self.cur_case != new_case:
# Finish a case
self._save_guide()
self.guides.clear()
# Update states with next case
self.cur_case = new_case
self.bbox = self.features_val["bboxes"][0]
# Reinitialize TumorManager
need_rev = self.cur_case.endswith("rev")
self.id = self.id - 1 if need_rev else 0
self.t_mgr.reset(direction=-1 if need_rev else 1)
self.t_mgr.name = self.cur_case
self.t_mgr.set_bbox(
self.bbox, shape=self.features_val["images"].shape[1:-1])
self.features_val["sp_guide"] = self.t_mgr.get_guide_image(
self.features_val["sp_guide"][0], new_id=self.id)
if self.t_mgr.direction == 1:
self.id += 1
else:
self.id -= 1
else:
# Update guide with last prediction
self.t_mgr.check_pred(predictions["TumorPred"][0, ..., 0])
if self.cbn:
# self.features_val["sp_guide"][0, ..., 0] = np.maximum(
# self.features_val["sp_guide"][0, ..., 0],
# get_gd_image_multi_objs(corrective_tumor,
# center_perturb=0., stddev_perturb=0.))
self.features_val["sp_guide"] = self.t_mgr.get_guide_image(
self.features_val["sp_guide"][0], self.id)
if self.t_mgr.direction == 1:
self.id += 1
else:
self.id -= 1
else:
self.features_val["images"][0, ..., -1] = np.maximum(
self.features_val["images"][0, ..., -1],
get_gd_image_multi_objs(self.t_mgr.pred,
center_perturb=0.,
stddev_perturb=0.))
if self.cbn:
self.guides.append(self.features_val["sp_guide"][0, ..., 0])
else:
self.guides.append(self.features_val["images"][0, ..., -1])
def _wrap_get_gd_image_multi_objs(mask):
guide_image = array_kits.get_gd_image_multi_objs(
mask,
obj_value=2,
center_perturb=0.,
stddev_perturb=0.,
partial=args.use_fewer_guide,
with_fake_guides=False,
partial_slice=args.guide)[..., None]
return guide_image.astype(np.float32)
def _wrap_get_gd_image_multi_objs(mask_):
guide_image = array_kits.get_gd_image_multi_objs(
mask_,
obj_value=2,
center_perturb=args.center_perturb,
stddev_perturb=args.stddev_perturb,
partial=False,
with_fake_guides=args.use_fake_guide,
fake_rate=args.fake_rate,
fake_range_value=1)[..., None]
return guide_image.astype(np.float32)
def test_get_gd_image_multi_objs(self):
for i in range(44, self.label.shape[0]):
guide = array_kits.get_gd_image_multi_objs(self.label[i],
2,
with_fake_guides=True)
plt.subplot(131)
plt.imshow(self.image[i], cmap="gray")
plt.subplot(132)
plt.imshow(self.label[i], cmap="gray")
plt.subplot(133)
plt.imshow(guide, cmap="gray")
plt.show()
def get_guide_image(self, guide, new_id):
self.set_guide_info(guide, new_id)
if len(self.centers_yx) > 0:
self.guides = arr_ops.create_gaussian_distribution(
self.shape, self.centers_yx[0, ::-1], self.stddevs_yx[0, ::-1])
for i in range(1, len(self.centers_yx)):
self.guides = np.maximum(
self.guides,
arr_ops.create_gaussian_distribution(
self.shape, self.centers_yx[i, ::-1],
self.stddevs_yx[i, ::-1]))
else:
self.guides = np.zeros(self.shape, dtype=np.float32)
if self.pred is None:
return self.guides[None, ..., None]
else:
self.guides = np.maximum(
self.guides,
arr_ops.get_gd_image_multi_objs(self.pred,
center_perturb=0.,
stddev_perturb=0.))
return self.guides[None, ..., None]