def get_mnist_parsed_args():
args = parsing.parse_args()
if not args.dataset in MNIST_DATASETS:
raise Exception(
NOT_MNIST_EXCEPTION.format(args.dataset, MNIST_DATASETS))
args.image_transformers = parsing.parse_transformers(
['scale_2d', 'grayscale'])
args.tensor_transformers = parsing.parse_transformers(['normalize_2d'])
args.test_image_transformers = parsing.parse_transformers(
['scale_2d', 'grayscale'])
args.test_tensor_transformers = parsing.parse_transformers(
['normalize_2d'])
args.epochs = 1
args.max_batches_per_epoch = 100
args.num_classes = 10
args.wrap_model = True
args.num_images = 2
return args
from onconet.learn import train
import onconet.transformers.factory as transformer_factory
import onconet.visualize as visualize
import onconet.utils.parsing as parsing
import warnings
import onconet.learn.state_keeper as state
from onconet.utils.get_dataset_stats import get_dataset_stats
import onconet.utils.stats as stats
import pdb
import csv
#Constants
DATE_FORMAT_STR = "%Y-%m-%d:%H-%M-%S"
if __name__ == '__main__':
args = parsing.parse_args()
if args.ignore_warnings:
warnings.simplefilter('ignore')
repo = git.Repo(search_parent_directories=True)
commit = repo.head.object
args.commit = commit.hexsha
print("OncoNet main running from commit: \n\n{}\n{}author: {}, date: {}".format(
commit.hexsha, commit.message, commit.author, commit.committed_date))
if args.get_dataset_stats:
print("\nComputing image mean and std...")
args.img_mean, args.img_std = get_dataset_stats(args)
print('Mean: {}'.format(args.img_mean))
print('Std: {}'.format(args.img_std))
shuffle=False,
num_workers=args.num_workers,
drop_last=False,
pin_memory=True,
collate_fn=ignore_None_collate)
for batch in tqdm(data_loader):
img = batch['x']
paths = batch['path']
if args.cuda:
img = img.cuda()
B, C, H, W = img.size()
left_half = img[:, :, :, :W//2].contiguous().view(B,-1)
right_half = img[:, :, :, W//2:].contiguous().view(B,-1)
is_right_aligned = right_half.sum(dim=-1) > left_half.sum(dim=-1)
for indx, path in enumerate(paths):
image_to_side[path] = bool(is_right_aligned[indx])
return image_to_side
if __name__ == "__main__":
args = parse_args()
image_is_right_side = get_image_to_right_side(args)
json.dump(image_is_right_side, open(IMAGE_RIGHT_ALIGNED_PATH,'w'))