def __init__(self, training):
BaseOptions.__init__(self)
# dataset arguments
if training: # assumes unpaired data loader is used during training
self.parser.add_argument('--dirA',
type=str,
required=True,
help='Path to training shadow dataset')
self.parser.add_argument(
'--dirB',
type=str,
required=True,
help='Path to training shadow free dataset')
else: # assumes single data loader is used during testing
self.parser.add_argument('--dir',
type=str,
required=True,
help='Path to test shadow dataset')
# model arguments
self.parser.add_argument(
'--lamA',
type=float,
default=10.0,
help='weight for forward cycle loss (A->B->A)')
self.parser.add_argument(
'--lamB',
type=float,
default=10.0,
help='weight for backward cycle loss (B->A->B)')
self.parser.add_argument('--lambda_ident',
type=float,
default=0.0,
help='weight for identity loss')
self.parser.add_argument(
'--ngf',
type=int,
default=64,
help='# of filters in first conv. layer of generator')
self.parser.add_argument(
'--ndf',
type=int,
default=64,
help='# of filters in first conv. layer of discriminator')
self.parser.add_argument(
'--pool_size',
type=int,
default=50,
help=
'the size of image buffer that stores previously generated images')
self.parser.add_argument(
'--queue_size',
type=int,
default=100,
help=
'the size of mask queue that stores previously generated shadow masks'
)
def __init__(self, training):
BaseOptions.__init__(self)
# dataset arguments
self.parser.add_argument('--dir',
type=str,
required=True,
help='Path to test dataset')
# generator arguments
self.parser.add_argument(
'--ngf',
type=int,
default=64,
help='# of filters in first conv. layer of generator')
# discriminator arguments
self.parser.add_argument(
'--ndf',
type=int,
default=64,
help='# of filters in first conv. layer of discriminator')
# training arguments
self.parser.add_argument(
'--num_epochs_init',
type=int,
default=100,
help='Number of epochs to train generator for initialization')
self.parser.add_argument('--num_epochs',
type=int,
default=2000,
help='Number of epochs to train for')
self.parser.add_argument('--lr_decay',
type=float,
default=0.1,
help='Decay learning rate')
self.parser.add_argument('--decay_every',
type=int,
default=1000,
help='When to decay learning rate')
self.parser.add_argument('--vgg_choose',
type=str,
default='block5_conv4',
help='Choose layer of VGG')
def __init__(self):
BaseOptions.__init__(self)
def __init__(self, training):
BaseOptions.__init__(self)
# dataset arguments
if training: # assumes unpaired data loader is used during training
self.parser.add_argument('--dirA',
type=str,
required=True,
help='Path to training dataset A')
self.parser.add_argument('--dirB',
type=str,
required=True,
help='Path to training dataset B')
else: # assumes single data loader is used during testing
self.parser.add_argument('--dir',
type=str,
required=True,
help='Path to test dataset')
# generator arguments
self.parser.add_argument(
'--ngf',
type=int,
default=32,
help='# of filters in first conv. layer of generator')
self.parser.add_argument(
'--netG',
type=str,
default='sid_unet_resize',
help=
'Specify generator architecture [resnet_9blocks | resnet_6blocks | sid_unet_resize]'
)
self.parser.add_argument(
'--self_attention',
action='store_true',
help='Adding attention on the input of generator')
self.parser.add_argument('--times_residual',
action='store_true',
help='output = input + residual*attention')
self.parser.add_argument('--skip',
type=float,
default=1.0,
help='B = G(A) + skip*A')
# discriminator arguments
self.parser.add_argument(
'--ndf',
type=int,
default=64,
help='# of filters in first conv. layer of discriminator')
self.parser.add_argument(
'--netD',
type=str,
default='no_norm_n_layers',
help=
'Specify discriminator architecture [basic | n_layers | no_norm_n_layers | pixel].'
)
self.parser.add_argument(
'--n_layers',
type=int,
default=5,
help=
'# of layers for discriminator. Only used if netD==[n_layers | no_norm_n_layers]'
)
self.parser.add_argument(
'--n_layers_patch',
type=int,
default=4,
help=
'# of layers for patch discriminator. Only used if netD==[n_layers | no_norm_n_layers]'
)
self.parser.add_argument('--patchD',
action='store_true',
help='Use patch discriminator')
self.parser.add_argument(
'--patchD_3',
type=int,
default=0,
help='Choose number of crops for patch discriminator')
self.parser.add_argument('--patch_size',
type=int,
default=32,
help='Size to crop patches to')
# other arguments
self.parser.add_argument('--vgg',
action='store_true',
help='Use perceptual loss')
self.parser.add_argument('--vgg_choose',
type=str,
default='block5_conv1',
help='Choose layer of VGG')
self.parser.add_argument(
'--no_vgg_instance',
action='store_true',
help='Whether to apply instance normalization on extracted features'
)
self.parser.add_argument('--patch_vgg',
action='store_true',
help='use vgg loss between each patch')
self.parser.add_argument(
'--pool_size',
type=int,
default=50,
help=
'the size of image buffer that stores previously generated images')
self.parser.add_argument(
'--gan_mode',
type=str,
default='lsgan',
help='Use least square GAN or vanilla GAN. Default is LSGAN.')
self.parser.add_argument('--use_ragan',
action='store_true',
help='Use ragan')
self.parser.add_argument('--hybrid_loss',
action='store_true',
help='Use lsgan and ragan separately')
def __init__(self, training):
BaseOptions.__init__(self)
def __init__(self, training):
BaseOptions.__init__(self)
# dataset arguments
if training: # assumes unpaired data loader is used during training
self.parser.add_argument('--dirA',
type=str,
required=True,
help='Path to training dataset A')
self.parser.add_argument('--dirB',
type=str,
required=True,
help='Path to training dataset B')
else: # assumes single data loader is used during testing
self.parser.add_argument('--dir',
type=str,
required=True,
help='Path to test dataset')
# model arguments
self.parser.add_argument(
'--lamA',
type=float,
default=10.0,
help='weight for forward cycle loss (A->B->A)')
self.parser.add_argument(
'--lamB',
type=float,
default=10.0,
help='weight for backward cycle loss (B->A->B)')
self.parser.add_argument('--lambda_ident',
type=float,
default=0.0,
help='weight for identity loss')
self.parser.add_argument(
'--ngf',
type=int,
default=64,
help='# of filters in first conv. layer of generator')
self.parser.add_argument(
'--ndf',
type=int,
default=64,
help='# of filters in first conv. layer of discriminator')
self.parser.add_argument(
'--netG',
type=str,
default='resnet_9blocks',
help=
'Specify generator architecture [resnet_9blocks | resnet_6blocks | unet_256 | unet_128]'
)
self.parser.add_argument(
'--netD',
type=str,
default='basic',
help=
'Specify discriminator architecture [basic | n_layers | pixel].')
self.parser.add_argument(
'--n_layers',
type=int,
default=3,
help='# of layers for discriminator. Only used if netD==n_layers')
self.parser.add_argument(
'--pool_size',
type=int,
default=50,
help=
'the size of image buffer that stores previously generated images')
self.parser.add_argument(
'--gan_mode',
type=str,
default='lsgan',
help='Use least square GAN or vanilla GAN. Default is LSGAN.')
