def test_should_create_multimodal_train_valid_test_datasets(self):
train_dataset, valid_dataset, test_dataset, reconstruction_dataset, = iSEGSliceDatasetFactory.create_train_valid_test(
self.DATA_PATH, [Modality.T1, Modality.T2], 0, 0.2)
assert_that(train_dataset, instance_of(Dataset))
assert_that(valid_dataset, instance_of(Dataset))
assert_that(test_dataset, instance_of(Dataset))
assert_that(reconstruction_dataset, instance_of(Dataset))
def test_should_produce_a_single_modality_input_with_one_channel(self):
train_dataset, valid_dataset, test_dataset, reconstruction_dataset, = iSEGSliceDatasetFactory.create_train_valid_test(
self.DATA_PATH, Modality.T1, 0, 0.2)
sample = train_dataset[32000]
assert_that(sample.x.size(), is_(torch.Size([1, 32, 32, 32])))
plt.imshow(sample.x[0, 16, :, :], cmap="gray")
plt.show()
plt.imshow(sample.y[0, 16, :, :], cmap="gray")
plt.show()
model_trainer_factory = ModelTrainerFactory(
model_factory=CustomModelFactory(),
criterion_factory=CustomCriterionFactory())
model_trainers = model_trainer_factory.create(model_trainer_configs)
if not isinstance(model_trainers, list):
model_trainers = [model_trainers]
# Create datasets
if dataset_configs.get("iSEG", None) is not None:
iSEG_train, iSEG_valid, iSEG_test, iSEG_reconstruction = iSEGSliceDatasetFactory.create_train_valid_test(
source_dir=dataset_configs["iSEG"].path,
modalities=dataset_configs["iSEG"].modalities,
dataset_id=ISEG_ID,
test_size=dataset_configs["iSEG"].validation_split,
max_subjects=dataset_configs["iSEG"].max_subjects,
max_num_patches=dataset_configs["iSEG"].max_num_patches,
augment=dataset_configs["iSEG"].augment,
patch_size=dataset_configs["iSEG"].patch_size,
step=dataset_configs["iSEG"].step,
test_patch_size=dataset_configs["iSEG"].test_patch_size,
test_step=dataset_configs["iSEG"].test_step,
data_augmentation_config=data_augmentation_config)
train_datasets.append(iSEG_train)
valid_datasets.append(iSEG_valid)
test_datasets.append(iSEG_test)
reconstruction_datasets.append(iSEG_reconstruction)
if dataset_configs.get("MRBrainS", None) is not None:
MRBrainS_train, MRBrainS_valid, MRBrainS_test, MRBrainS_reconstruction = MRBrainSSliceDatasetFactory.create_train_valid_test(
source_dir=dataset_configs["MRBrainS"].path,
modalities=dataset_configs["MRBrainS"].modalities,