logger.info('max_epochs : {}'.format(max_epochs))
logger.info('cuda available : {}'.format(torch.cuda.is_available()))
# Data
logger.info('Axial AID with nn.conv2d masked')
if not os.path.exists(data_directory + '/aid' + str(resize_size[0]) + '_hexa.h5'):
logger.info('Create hexagonal AID dataset')
img, _ = datasets.ImageFolder(data_directory + '/AID',
transform=transforms.Compose([transforms.Resize(resize_size),
transforms.ToTensor()]))[0]
index_matrix = utils.square_to_hexagonal_index_matrix(img)
aid = datasets.ImageFolder(data_directory + '/AID',
transform=transforms.Compose([transforms.Resize(resize_size),
transforms.ToTensor(),
utils.SquareToHexa()]))
with h5py.File(data_directory + '/aid' + str(resize_size[0]) + '_hexa.h5', 'w') as f:
images = []
labels = []
for i in range(len(aid)):
image, label = aid[i]
images.append(image.numpy())
labels.append(label)
f.create_dataset('images', data=np.array(images))
f.create_dataset('labels', data=np.array(labels))
f.attrs['index_matrix'] = index_matrix
f.attrs['class_names'] = np.array(aid.classes, dtype=h5py.special_dtype(vlen=str))
# load hexagonal cifar
f = h5py.File(data_directory + '/aid' + str(resize_size[0]) + '_hexa.h5', 'r')
data = f['images'][()]
logger.addHandler(console_handler)
formatter_file = logging.Formatter('%(asctime)s [%(levelname)s] - %(message)s')
file_handler = logging.FileHandler('{}/{}/{}.log'.format(main_directory,
experiment_name,
experiment_name))
file_handler.setFormatter(formatter_file)
logger.addHandler(file_handler)
batch_size = 64
test_batch_size = 1000
train_set = datasets.MNIST(main_directory + '/../ext_data', train=True, download=True,
transform=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.1307,), (0.3081,)),
utils.SquareToHexa()
]))
train_loader = torch.utils.data.DataLoader(train_set, batch_size=batch_size, shuffle=True)
test_loader = torch.utils.data.DataLoader(datasets.MNIST(main_directory + '/../ext_data', train=False,
transform=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.1307,), (0.3081,)),
utils.SquareToHexa()
])),
batch_size=test_batch_size, shuffle=True)
device = torch.device("cuda")
# Plot a resampled image to check
img, _ = datasets.MNIST(main_directory + '/../ext_data', train=True, download=True,
transform=transforms.Compose([
# Experiment parameters
logger.info('batch_size : {}'.format(batch_size))
logger.info('max_epochs : {}'.format(max_epochs))
logger.info('cuda available : {}'.format(torch.cuda.is_available()))
# Data
if hexa:
camera_layout = 'Hex'
logger.info('Hexagonal CIFAR')
img, _ = datasets.CIFAR10(data_directory, train=True, download=True, transform=transforms.ToTensor())[0]
index_matrix = utils.square_to_hexagonal_index_matrix(img)
if not os.path.exists(data_directory + '/cifar10.hdf5'):
train_set = datasets.CIFAR10(data_directory, train=True, download=True,
transform=transforms.Compose([transforms.ToTensor(), utils.SquareToHexa()]))
with h5py.File(data_directory + '/cifar10.hdf5', 'w') as f:
images = []
labels = []
for i in range(len(train_set)):
image, label = train_set[i]
images.append(image.numpy())
labels.append(label)
f.create_dataset('images', data=np.array(images))
f.create_dataset('labels', data=np.array(labels))
f.attrs['index_matrix'] = index_matrix
if not os.path.exists(data_directory + '/cifar10_test.hdf5'):
test_set = datasets.CIFAR10(data_directory, train=False,
transform=transforms.Compose([transforms.ToTensor(), utils.SquareToHexa()]))
with h5py.File(data_directory + '/cifar10_test.hdf5', 'w') as f:
images = []