def prepare_datalist(args):
dimensions = args.dimensions
dataset_json = os.path.join(args.output, 'dataset.json')
if not os.path.exists(dataset_json):
logging.info('Processing dataset...')
with open(os.path.join(args.dataset_json)) as f:
datalist = json.load(f)
datalist = create_dataset(datalist=datalist[args.datalist_key],
base_dir=args.dataset_root,
output_dir=args.output,
dimension=dimensions,
pixdim=[1.0] * dimensions,
limit=args.limit,
relative_path=args.relative_path)
with open(dataset_json, 'w') as fp:
json.dump(datalist, fp, indent=2)
else:
logging.info('Pre-load existing dataset.json')
dataset_json = os.path.join(args.output, 'dataset.json')
with open(dataset_json) as f:
datalist = json.load(f)
logging.info('+++ Dataset File: {}'.format(dataset_json))
logging.info('+++ Total Records: {}'.format(len(datalist)))
logging.info('')
def test_create_dataset_3d(self):
with tempfile.TemporaryDirectory() as tempdir:
datalist = self._create_data(tempdir)
output_dir = os.path.join(tempdir, "3d")
deepgrow_datalist = create_dataset(datalist=datalist,
output_dir=output_dir,
dimension=3,
pixdim=(1, 1, 1))
self.assertEqual(len(deepgrow_datalist), 1)
self.assertEqual(deepgrow_datalist[0]["region"], 1)
def pre_process(self, request, datastore: Datastore):
self.cleanup(request)
cache_dir = self.get_cache_dir(request)
output_dir = os.path.join(cache_dir,
f"deepgrow_{self.dimension}D_train")
logger.info(
f"Preparing Dataset for Deepgrow-{self.dimension}D:: {output_dir}")
datalist = create_dataset(
datalist=datastore.datalist(),
base_dir=None,
output_dir=output_dir,
dimension=self.dimension,
pixdim=[1.0] * self.dimension,
)
logging.info(f"+++ Total Records: {len(datalist)}")
return datalist
def prepare_datalist(args):
dimensions = args.dimensions
dataset_json = os.path.join(args.output, 'dataset.json')
logging.info('Processing dataset...')
with open(os.path.join(args.dataset_json)) as f:
datalist = json.load(f)
datalist = create_dataset(datalist=datalist[args.datalist_key],
base_dir=args.dataset_root,
output_dir=args.output,
dimension=dimensions,
pixdim=[1.0] * dimensions,
limit=args.limit,
relative_path=args.relative_path)
with open(dataset_json, 'w') as fp:
json.dump(datalist, fp, indent=2)
dataset_json = os.path.join(args.output, 'dataset.json')
with open(dataset_json) as f:
datalist = json.load(f)
logging.info('+++ Dataset File: {}'.format(dataset_json))
logging.info('+++ Total Records: {}'.format(len(datalist)))
logging.info('')
train_ds, val_ds = partition_dataset(datalist,
ratios=[args.split, (1 - args.split)],
shuffle=True,
seed=args.seed)
dataset_json = os.path.join(args.output, 'dataset_0.json')
with open(dataset_json, 'w') as fp:
json.dump({'training': train_ds, 'validation': val_ds}, fp, indent=2)
logging.info('*** Dataset File: {}'.format(dataset_json))
logging.info('*** Total Records for Training: {}'.format(len(train_ds)))
logging.info('*** Total Records for Validation: {}'.format(len(val_ds)))
assert len(train_ds) > 0, "Train Dataset/Records is EMPTY"
assert len(val_ds) > 0, "Validation Dataset/Records is EMPTY"
def test_empty_datalist(self):
with self.assertRaises(ValueError):
create_dataset(datalist=[], output_dir=self.tempdir, dimension=3, pixdim=(1, 1, 1))
def test_invalid_dim(self):
with self.assertRaises(ValueError):
create_dataset(datalist=self._create_data(), output_dir=self.tempdir, dimension=4, pixdim=(1, 1, 1, 1))
def test_create_dataset(self, args, data_args, expected_length, expected_region):
datalist = self._create_data(**data_args)
deepgrow_datalist = create_dataset(datalist=datalist, output_dir=self.tempdir, **args)
self.assertEqual(len(deepgrow_datalist), expected_length)
if expected_region is not None:
self.assertEqual(deepgrow_datalist[0]["region"], expected_region)
