def get_clf(self):
if self.flags.use_clf:
dir_clf = self.flags.dir_clf
if not dir_clf.exists():
download_zip_from_url(
url=
'https://www.dropbox.com/sh/lx8669lyok9ois6/AADM7Cs_QReijyo2kF8xzWqua/trained_classifiers/trained_clfs_mst?dl=1',
dest_folder=dir_clf)
model_clf = ClfImgSVHN()
model_clf.load_state_dict(
torch.load(os.path.join(self.flags.dir_clf, f"clf_m2"),
map_location=self.flags.device))
return model_clf.to(self.flags.device)
def get_clf(self):
if self.flags.use_clf:
clf_name_mapping = {'PA': 'pa', 'Lateral': 'lat'}
# finding the directory of the classifier
img_clf_path = Path(
__file__
).parent.parent / f'classifiers/state_dicts/{clf_name_mapping[self.name]}_clf_{self.flags.img_size}.pth'
if not img_clf_path.exists():
download_zip_from_url(
url=
'http://jimmy123.hopto.org:2095/nextcloud/index.php/s/GTc8pYiDKrq35ky/download',
dest_folder=img_clf_path.parent.parent,
verbose=True)
lightning_module = LM_(str_labels=self.labels,
transforms=self.clf_transforms)
lightning_module.model.load_state_dict(
torch.load(img_clf_path, map_location=self.flags.device))
return lightning_module.to(self.flags.device)
def __init__(self, args, alphabet, partition=0, transform=None):
self.dir_dataset_base = args.dir_data
if not self.dir_dataset_base.exists():
log.info(
f'Dataset not found under {self.dir_dataset_base}. Parent directory contains: '
f'{list(self.dir_dataset_base.parent)}')
download_zip_from_url(
url=
'https://www.dropbox.com/sh/lx8669lyok9ois6/AACCZqDiZuv0Q8RA3Qmwrwnca/celeba_data.zip?dl=1',
dest_folder=Path(self.dir_dataset_base).parent,
verbose=True)
filename_text = self.dir_dataset_base / (
'list_attr_text_' + str(args.len_sequence).zfill(3) + '_' +
str(args.random_text_ordering) + '_' +
str(args.random_text_startindex) + '_celeba.csv')
filename_partition = os.path.join(self.dir_dataset_base,
'list_eval_partition.csv')
filename_attributes = os.path.join(self.dir_dataset_base,
'list_attr_celeba.csv')
df_text = pd.read_csv(filename_text)
df_partition = pd.read_csv(filename_partition)
df_attributes = pd.read_csv(filename_attributes)
self.args = args
self.img_dir = os.path.join(self.dir_dataset_base, 'img_align_celeba')
self.txt_path = filename_text
self.attrributes_path = filename_attributes
self.partition_path = filename_partition
self.alphabet = alphabet
self.img_names = df_text.loc[df_partition['partition'] ==
partition]['image_id'].values
self.attributes = df_attributes.loc[df_partition['partition'] ==
partition]
self.labels = df_attributes.loc[
df_partition['partition'] ==
partition].values # atm, i am just using blond_hair as labels
self.y = df_text.loc[df_partition['partition'] ==
partition]['text'].values
self.transform = transform
def __init__(self, flags, alphabet, train=True, transform=None, target_transform=None):
super(SVHNMNIST, self).__init__(flags.dir_data)
self.flags = flags;
self.dataset = 'MNIST_SVHN';
self.dataset_mnist = 'MNIST';
self.dataset_svhn = 'SVHN';
self.len_sequence = flags.len_sequence
self.transform = transform
self.target_transform = target_transform
self.train = train # training set or test set
self.alphabet = alphabet;
self.dir_svhn = os.path.join(self.root, self.dataset_svhn)
self.dir_mnist = os.path.join(self.root, self.dataset_mnist)
print(self.dir_svhn)
if not self._check_exists_mnist():
download_zip_from_url(
url='https://www.dropbox.com/sh/lx8669lyok9ois6/AADMhr3EluBXJyZnV1_lYntTa/data_mnistsvhntext.zip?dl=1',
dest_folder=Path(self.dir_mnist).parent.parent)
assert self._check_exists_mnist(), 'Dataset MNIST not found.'
# if not self._check_exists_svhn():
# raise RuntimeError('Dataset SVHN not found.')
if self.train:
id_file_svhn = self.training_file_svhn_idx;
id_file_mnist = self.training_file_mnist_idx;
data_file_svhn = self.training_file_svhn;
data_file_mnist = self.training_file_mnist;
else:
id_file_svhn = self.test_file_svhn_idx;
id_file_mnist = self.test_file_mnist_idx;
data_file_svhn = self.test_file_svhn;
data_file_mnist = self.test_file_mnist;
# import here rather than at top of file because this is
# an optional dependency for torchvision
import scipy.io as sio
print(os.path.join(self.dir_svhn, data_file_svhn))
# reading(loading) mat file as array
loaded_mat = sio.loadmat(os.path.join(self.dir_svhn, data_file_svhn))
self.data_svhn = loaded_mat['X']
# loading from the .mat file gives an np array of type np.uint8
# converting to np.int64, so that we have a LongTensor after
# the conversion from the numpy array
# the squeeze is needed to obtain a 1D tensor
self.labels_svhn = loaded_mat['y'].astype(np.int64).squeeze()
# the svhn dataset assigns the class label "10" to the digit 0
# this makes it inconsistent with several loss functions
# which expect the class labels to be in the range [0, C-1]
np.place(self.labels_svhn, self.labels_svhn == 10, 0)
self.data_svhn = np.transpose(self.data_svhn, (3, 2, 0, 1))
samples_svhn = self.data_svhn.shape[0];
channels_svhn = self.data_svhn.shape[1];
width_svhn = self.data_svhn.shape[2];
height_svhn = self.data_svhn.shape[3];
self.data_mnist, self.labels_mnist = torch.load(os.path.join(self.processed_folder, data_file_mnist));
# # get transformed indices
self.labels_svhn = torch.LongTensor(self.labels_svhn);
mnist_l, mnist_li = self.labels_mnist.sort()
svhn_l, svhn_li = self.labels_svhn.sort()
self.mnist_idx, self.svhn_idx = rand_match_on_idx(mnist_l, mnist_li,
svhn_l, svhn_li,
max_d=10000,
dm=flags.data_multiplications)
return {_class: i for i, _class in enumerate(self.classes)}
def _check_exists_mnist(self):
return (Path(self.processed_folder) / self.training_file_mnist).exists() and (
Path(self.processed_folder) / self.test_file_mnist).exists()
def _check_exists_svhn(self):
return (os.path.exists(os.path.join(self.dir_svhn,
self.training_file_svhn)) and
os.path.exists(os.path.join(self.dir_svhn,
self.test_file_svhn)))
@staticmethod
def extract_gzip(gzip_path, remove_finished=False):
print('Extracting {}'.format(gzip_path))
with open(gzip_path.replace('.gz', ''), 'wb') as out_f, \
gzip.GzipFile(gzip_path) as zip_f:
out_f.write(zip_f.read())
if remove_finished:
os.unlink(gzip_path)
def extra_repr(self):
return "Split: {}".format("Train" if self.train is True else "Test")
if __name__ == '__main__':
config = json2dict(Path(get_config_path(dataset='mnistsvhntext')))
download_zip_from_url(
url='https://www.dropbox.com/sh/lx8669lyok9ois6/AADMhr3EluBXJyZnV1_lYntTa/data_mnistsvhntext.zip?dl=1',
dest_folder=Path(config['dir_data']).expanduser().parent, verbose=True)
data_size = torch.Size((1, flags.img_size, flags.img_size))
super().__init__(data_size=data_size,
flags=flags,
name='PA',
labels=labels,
rec_weight=rec_weight,
plot_img_size=plot_img_size)
class MimicLateral(MimicImg):
def __init__(self, flags, labels: typing.Iterable[str], rec_weight,
plot_img_size):
data_size = torch.Size((1, flags.img_size, flags.img_size))
super().__init__(data_size=data_size,
flags=flags,
name='Lateral',
labels=labels,
rec_weight=rec_weight,
plot_img_size=plot_img_size)
if __name__ == '__main__':
img_clf_path = Path(
__file__).parent.parent / f'classifiers/state_dicts/pa_clf_128.pth'
if not img_clf_path.exists():
download_zip_from_url(
url=
'http://jimmy123.hopto.org:2095/nextcloud/index.php/s/GTc8pYiDKrq35ky/download',
dest_folder=img_clf_path.parent.parent,
verbose=True)
label = torch.from_numpy(
(self.labels[index, 1:] > 0).astype(int)).float()
# img = torch.rand((self.args.image_channels, 64, 64))
# text_str = torch.ones((8, 71))
# img = torch.ones((1, 28, 28))
# text_str = torch.rand((256, 71))
sample = {'img': img, 'text': text_str}
# sample = {'img': img}
return sample, label
def __len__(self):
return self.y.shape[0]
def get_text_str(self, index):
return self.y[index]
if __name__ == '__main__':
config = json2dict(get_config_path(dataset='celeba'))
dir_data = Path(config['dir_data']).expanduser()
if not dir_data.exists():
download_zip_from_url(
url=
'https://www.dropbox.com/sh/lx8669lyok9ois6/AACCZqDiZuv0Q8RA3Qmwrwnca/celeba_data.zip?dl=1',
dest_folder=dir_data.parent,
verbose=True)
print("Done.")
"""Download trianed models and evaluation results from zenodo"""
from pathlib import Path
from modun.download_utils import download_zip_from_url
download_zip_from_url(
url="https://zenodo.org/record/5588294/files/data.zip?download=1",
dest_folder=Path(__file__).parent.parent,
verbose=True)