def train_and_predict_with_config(test_name, config_path):
"""
Function that helps unit test whether a session with a given config
runs the train and predict functions
"""
tf.get_logger().setLevel(3)
tf.keras.backend.clear_session(
) # needed for pytest, otherwise model output name will change
gpu = ""
gpu_allow_growth = False
ckpt_path = ""
log_dir = "pytest_train_" + test_name
train(
gpu=gpu,
config_path=config_path,
gpu_allow_growth=gpu_allow_growth,
ckpt_path=ckpt_path,
log_dir=log_dir,
)
ckpt_path = os.path.join("logs", log_dir, "save", "weights-epoch2.ckpt")
log_dir = "pytest_predict_" + test_name
predict(
gpu=gpu,
gpu_allow_growth=gpu_allow_growth,
ckpt_path=ckpt_path,
mode="test",
batch_size=1,
log_dir=log_dir,
sample_label="all",
config_path="",
)
def test_train():
"""
Test train by checking it can run.
"""
gpu = ""
config_path = "deepreg/config/unpaired_labeled_ddf.yaml"
gpu_allow_growth = False
ckpt_path = ""
log_dir = "test_train"
train(
gpu=gpu,
config_path=config_path,
gpu_allow_growth=gpu_allow_growth,
ckpt_path=ckpt_path,
log_dir=log_dir,
)
dest="test",
action="store_false",
)
parser.set_defaults(test=True)
args = parser.parse_args()
print("\n\n\n\n\n"
"=======================================================\n"
"The training can also be launched using the following command.\n"
"deepreg_train --gpu '0' "
f"--config_path demos/{name}/{name}.yaml "
f"--log_dir demos/{name} "
"--exp_name logs_train\n"
"=======================================================\n"
"\n\n\n\n\n")
log_dir = f"demos/{name}"
exp_name = "logs_train/" + datetime.now().strftime("%Y%m%d-%H%M%S")
config_path = [f"demos/{name}/{name}.yaml"]
if args.test:
config_path.append("config/test/demo_unpaired_grouped.yaml")
train(
gpu="0",
config_path=config_path,
gpu_allow_growth=True,
ckpt_path="",
log_dir=log_dir,
exp_name=exp_name,
)
dest="test",
action="store_false",
)
parser.set_defaults(test=True)
args = parser.parse_args()
print("\n\n\n\n\n"
"=======================================================\n"
"The training can also be launched using the following command.\n"
"deepreg_train --gpu '0' "
f"--config_path demos/{name}/{name}.yaml "
f"--log_root demos/{name} "
"--log_dir logs_train\n"
"=======================================================\n"
"\n\n\n\n\n")
log_root = f"demos/{name}"
log_dir = "logs_train/" + datetime.now().strftime("%Y%m%d-%H%M%S")
config_path = [f"demos/{name}/{name}.yaml"]
if args.test:
config_path.append("config/test/demo_unpaired_grouped.yaml")
train(
gpu="0",
config_path=config_path,
gpu_allow_growth=True,
ckpt_path="",
log_root=log_root,
log_dir=log_dir,
)
from deepreg.train import train
######## NOW WE DO THE TRAINING ########
gpu = "0"
gpu_allow_growth = False
ckpt_path = ""
log_dir = "learn2reg_t2_unpaired_train_logs"
config_path = [
r"demos/unpaired_ct_lung/unpaired_ct_lung_train.yaml",
r"demos/unpaired_ct_lung/unpaired_ct_lung.yaml",
]
train(
gpu=gpu,
config_path=config_path,
gpu_allow_growth=gpu_allow_growth,
ckpt_path=ckpt_path,
log_dir=log_dir,
)
)
self.conv2 = tf.keras.layers.Conv3D(
filters=out_channels,
kernel_size=1,
kernel_initializer=out_kernel_initializer,
activation=out_activation,
padding="same",
)
def call(self, inputs: tf.Tensor, training=None, mask=None) -> tf.Tensor:
"""
Builds graph based on built layers.
:param inputs: shape = (batch, f_dim1, f_dim2, f_dim3, in_channels)
:param training:
:param mask:
:return: shape = (batch, f_dim1, f_dim2, f_dim3, out_channels)
"""
out = self.conv1(inputs)
out = self.conv2(out)
return out
config_path = "examples/config_custom_backbone.yaml"
train(
gpu="",
config_path=config_path,
gpu_allow_growth=True,
ckpt_path="",
)
