def init_directories(self):
"""
Creates the 'im_seg' directory in the data_path. Also creates the train/val/test/out
directories with the images/ and annotations/ directory for each.
If the directories already exist, then initialises the data_sizes based on existing
directories.
"""
Dataset._create_dirs(
self.im_seg_path,
self.train_path,
self.val_path,
self.test_path,
)
# Create train, val, test dirs, each with an images and annotations sub-directories
dirs = {
"train": self.train_path,
"val": self.val_path,
"test": self.test_path
}
for set_type, directory in dirs.items():
Dataset._create_dirs(os.path.join(directory, 'images'),
os.path.join(directory, 'annotations'))
# Size of each training, val and test directories
num_samples = len([name for name in os.listdir(os.path.join(directory, 'images'))\
if name.endswith('.jpg')])
self.data_sizes[set_type] = num_samples
def create_model_out_dir(self, model_name):
"""
Creates directories for metrics, ouput images and annotations for a
given model during training.
"""
try:
getattr(self, "model_path")
raise AttributeError("Attribute model_path already created")
except AttributeError as e:
pass
self.model_path = os.path.join(self.out_path, model_name)
self.checkpoint_path = os.path.join(self.model_path, 'checkpoints')
self.metrics_path = os.path.join(self.model_path, 'metrics')
self.preds_path = os.path.join(self.model_path, 'preds')
Dataset._create_dirs(self.out_path, self.model_path,
self.checkpoint_path, self.metrics_path,
self.preds_path)
def __init__(self, data_path, classes_path='classes.json'):
super().__init__(data_path, classes_path=classes_path)
self.meta_path = os.path.join(self.data_path, 'metadata')
Dataset._create_dirs(self.meta_path)
