def __init__(self,
detr: Union[nn.DataParallel, CellDETR],
detr_optimizer: torch.optim.Optimizer,
detr_segmentation_optimizer: torch.optim.Optimizer,
training_dataset: DataLoader,
validation_dataset: DataLoader,
test_dataset: DataLoader,
loss_function: nn.Module,
learning_rate_schedule: torch.optim.lr_scheduler.
MultiStepLR = None,
device: str = "cuda",
save_data_path: str = "saved_data",
use_telegram: bool = True) -> None:
"""
Constructor method
:param detr: (Union[nn.DataParallel, DETR]) DETR model
:param detr_optimizer: (torch.optim.Optimizer) DETR model optimizer
:param detr_segmentation_optimizer: (torch.optim.Optimizer) DETR segmentation head optimizer
:param training_dataset: (DataLoader) Training dataset
:param validation_dataset: (DataLoader) Validation dataset
:param test_dataset: (DataLoader) Test dataset
:param loss_function: (nn.Module) Loss function
:param learning_rate_schedule: (torch.optim.lr_scheduler.MultiStepLR) Learning rate schedule
:param device: (str) Device to be utilized
:param save_data_path: (str) Path to store log data
:param use_telegram: (bool) If true telegram_send is used
"""
# Save parameters
self.detr = detr
self.detr_optimizer = detr_optimizer
self.detr_segmentation_optimizer = detr_segmentation_optimizer
self.training_dataset = training_dataset
self.validation_dataset = validation_dataset
self.test_dataset = test_dataset
self.loss_function = loss_function
self.learning_rate_schedule = learning_rate_schedule
self.device = device
self.save_data_path = save_data_path
self.use_telegram = use_telegram
# Init logger
self.logger = misc.Logger()
# Make directories to save logs, plots and models during training
time_and_date = str(datetime.now())
save_data_path = os.path.join(save_data_path, time_and_date)
os.makedirs(save_data_path, exist_ok=True)
self.path_save_models = os.path.join(save_data_path, "models")
os.makedirs(self.path_save_models, exist_ok=True)
self.path_save_plots = os.path.join(save_data_path, "plots")
os.makedirs(self.path_save_plots, exist_ok=True)
self.path_save_metrics = os.path.join(save_data_path, "metrics")
os.makedirs(self.path_save_metrics, exist_ok=True)
# Init variable to store best mIoU
self.best_miou = 0.0
from __future__ import print_function
if __name__ == '__main__':
from cifar_main import parse_argument
arguments={
'image-size':256,
'num_classes':4,
'batch-size':32,
'lr':0.002,
'gpu':1}
args = parse_argument(additional_arguments=arguments, description='Location classification for stomach images.')
import misc
misc.ensure_dir(args.logdir)
logger = misc.Logger(args.logdir, 'train_log')
print = logger.info
print("-----------------FLAGS-----------------")
for k, v in args.__dict__.items():
print('{}: {}'.format(k, v))
print("---------------------------------------\n")
image_path='Data/Normal'
splits_p=[0.9, 0.1]
import torchvision.transforms as transforms
train_transform = transforms.Compose([
transforms.Pad(args.image_size // 8),
transforms.RandomCrop(args.image_size),
transforms.RandomHorizontalFlip(),
])
trans = [train_transform, None]
from stomach_data import get_data_loaders
def __init__(self,
detr: Union[nn.DataParallel, CellDETR],
detr_optimizer: torch.optim.Optimizer,
detr_segmentation_optimizer: torch.optim.Optimizer,
training_dataset: DataLoader,
validation_dataset: DataLoader,
test_dataset: DataLoader,
class_labels: [
"11", "12", "13", "14", "15", "16", "17", "18", "21",
"22", "23", "24", "25", "26", "27", "28", "31", "32",
"33", "34", "35", "36", "37", "38", "41", "42", "43",
"44", "45", "46", "47", "48"
],
colors: list,
loss_function: nn.Module,
learning_rate_schedule: torch.optim.lr_scheduler.
MultiStepLR = None,
device: str = "cuda",
save_data_path: str = "model_zoo/Cell-DETR",
experiment: str = "first_run",
use_telegram: bool = True) -> None:
"""
Constructor method
:param detr: (Union[nn.DataParallel, DETR]) DETR model
:param detr_optimizer: (torch.optim.Optimizer) DETR model optimizer
:param detr_segmentation_optimizer: (torch.optim.Optimizer) DETR segmentation head optimizer
:param training_dataset: (DataLoader) Training dataset
:param validation_dataset: (DataLoader) Validation dataset
:param test_dataset: (DataLoader) Test dataset
:param loss_function: (nn.Module) Loss function
:param learning_rate_schedule: (torch.optim.lr_scheduler.MultiStepLR) Learning rate schedule
:param device: (str) Device to be utilized
:param save_data_path: (str) Path to store log data
:param use_telegram: (bool) If true telegram_send is used
"""
# Save parameters
self.detr = detr
self.detr_optimizer = detr_optimizer
self.detr_segmentation_optimizer = detr_segmentation_optimizer
self.training_dataset = training_dataset
self.validation_dataset = validation_dataset
self.test_dataset = test_dataset
self.class_labels = class_labels
self.colors = colors
self.experiment = experiment
self.loss_function = loss_function
self.learning_rate_schedule = learning_rate_schedule
self.device = device
self.save_data_path = save_data_path
self.use_telegram = use_telegram
# Init logger
self.logger = misc.Logger()
# Make directories to save logs, plots and models during training
save_data_path = os.path.join(save_data_path, experiment)
os.makedirs(save_data_path, exist_ok=True)
self.path_save_models = os.path.join(save_data_path, "models")
os.makedirs(self.path_save_models, exist_ok=True)
self.path_save_plots = os.path.join(save_data_path, "plots")
os.makedirs(self.path_save_plots, exist_ok=True)
self.path_save_metrics = os.path.join(save_data_path, "metrics")
os.makedirs(self.path_save_metrics, exist_ok=True)
# Init variable to store best mIoU
self.best_miou = 0.0