def test_load_torch(self):
multi_slice_viewer_debug(self._t1, block=True) # only one image
multi_slice_viewer_debug(
[self._t1, self._t2],
block=True) # only two images on pair without given gts
multi_slice_viewer_debug(self._t1,
self._l1,
self._l2,
self._l3,
block=True) # only one image with 3 gts
multi_slice_viewer_debug([self._t1, self._t2],
self._l1,
self._l2,
self._l3,
block=True) # two images with 3 shared gt
def test_semi_dataloader(self):
iseg_handler = ISeg2017SemiInterface(
root_dir=self.dataset_root,
labeled_data_ratio=0.8,
unlabeled_data_ratio=0.2,
seed=0,
verbose=True,
)
iseg_handler.compile_dataloader_params(
batch_size=20, shuffle=True, drop_last=True
)
(
labeled_loader,
unlabeled_loader,
val_loader,
) = iseg_handler.SemiSupervisedDataLoaders(
labeled_transform=None,
unlabeled_transform=None,
val_transform=None,
group_labeled=True,
group_unlabeled=True,
group_val=True,
)
from deepclustering.viewer import multi_slice_viewer_debug
(T1, T2, Labels), filename = iter(labeled_loader).__next__()
multi_slice_viewer_debug(T1.squeeze(), Labels.squeeze(), block=False)
multi_slice_viewer_debug(T2.squeeze(), Labels.squeeze(), block=False)
(T1, T2, Labels), filename = iter(unlabeled_loader).__next__()
multi_slice_viewer_debug(
T1.squeeze(), T2.squeeze(), Labels.squeeze(), block=False
)
(T1, T2, Labels), filename = iter(val_loader).__next__()
multi_slice_viewer_debug(
T1.squeeze(), T2.squeeze(), Labels.squeeze(), block=False
)
def test_gt(self):
preds = self._network(self._img)
multi_slice_viewer_debug(
self._img.transpose(1, 3).transpose(1, 2),
preds.max(1)[1],
block=False,
no_contour=False,
)
from copy import deepcopy as dcp
preds2 = dcp(self._network)(self._img)
multi_slice_viewer_debug(
self._img.transpose(1, 3).transpose(1, 2),
preds2.max(1)[1],
block=False,
no_contour=False,
)
assert torch.allclose(preds, preds2)
fsgmGenerator = FSGMGenerator(net=self._network, eplision=0.01)
adv_img, _ = fsgmGenerator(self._img, gt=None)
assert adv_img.shape == self._img.shape
preds_adv = self._network(adv_img)
multi_slice_viewer_debug(
self._img.transpose(1, 3).transpose(1, 2),
preds_adv.max(1)[1],
block=False,
no_contour=False,
)
rand_pred = self._network(self._img + torch.randn_like(self._img).sign() * 0.01)
multi_slice_viewer_debug(
self._img.transpose(1, 3).transpose(1, 2),
rand_pred.max(1)[1],
block=True,
no_contour=False,
)
def test_semi_dataloader(self):
iseg_handler = WMHSemiInterface(
root_dir=self.dataset_root,
labeled_data_ratio=0.8,
unlabeled_data_ratio=0.2,
seed=0,
verbose=True,
)
iseg_handler.compile_dataloader_params(batch_size=20,
shuffle=True,
drop_last=True)
(
labeled_loader,
unlabeled_loader,
val_loader,
) = iseg_handler.SemiSupervisedDataLoaders(
group_labeled=False,
group_unlabeled=True,
group_val=True,
)
from deepclustering.viewer import multi_slice_viewer_debug
(t1, flair, gt), filename = iter(labeled_loader).__next__()
multi_slice_viewer_debug(t1.squeeze(),
flair.squeeze(),
gt.squeeze(),
block=False)
(t1, flair, gt), filename = iter(unlabeled_loader).__next__()
multi_slice_viewer_debug(t1.squeeze(),
flair.squeeze(),
gt.squeeze(),
block=False)
(t1, flair, gt), filename = iter(val_loader).__next__()
multi_slice_viewer_debug(t1.squeeze(),
flair.squeeze(),
gt.squeeze(),
block=True)
def test_load_numpy(self):
t1 = torch2numpy(self._t1)
t2 = torch2numpy(self._t2)
l1 = torch2numpy(self._l1)
l2 = torch2numpy(self._l2)
l3 = torch2numpy(self._l3)
multi_slice_viewer_debug(t1, block=True)
multi_slice_viewer_debug([t1, t2], block=True)
multi_slice_viewer_debug((t1, t2), block=True)
multi_slice_viewer_debug(t1, l1, l2, l3, block=True)
multi_slice_viewer_debug([t1, t2], l1, l2, l3, block=True)
multi_slice_viewer_debug((t1, t2), l1, l2, l3, block=True)
multi_slice_viewer_debug((t1, t2),
l1,
l2,
l3,
block=True,
no_contour=True)