def define_F(opt, use_bn=False):
gpu_ids = opt['gpu_ids']
device = torch.device('cuda' if gpu_ids else 'cpu')
if 'unet' in opt and opt['unet'] == True:
# Use UNet.
netF = UNet(n_channels=3, n_classes=10)
unet_model_path = opt['unet_model']
netF.load_state_dict(torch.load(unet_model_path))
if gpu_ids:
netF = nn.DataParallel(netF)
elif 'pnasnet' in opt and opt['pnasnet'] == True:
netF = arch.PNasNetFeatureExtractor(use_input_norm=True, device=device)
else:
# pytorch pretrained VGG19-54, before ReLU.
if use_bn:
feature_layer = 49
else:
feature_layer = 34
netF = arch.VGGFeatureExtractor(feature_layer=feature_layer,
use_bn=use_bn,
use_input_norm=True,
device=device)
# netF = arch.ResNet101FeatureExtractor(use_input_norm=True, device=device)
if gpu_ids:
netF = nn.DataParallel(netF)
netF.eval() # No need to train
return netF
def define_F(opt, use_bn=False):
gpu_ids = opt['gpu_ids']
tensor = torch.cuda.FloatTensor if gpu_ids else torch.FloatTensor
# pytorch pretrained VGG19-54, before ReLU.
if use_bn:
feature_layer = 49
else:
feature_layer = 34
netF = arch.VGGFeatureExtractor(feature_layer=feature_layer, use_bn=use_bn, \
use_input_norm=True, tensor=tensor)
if gpu_ids:
netF = nn.DataParallel(netF).cuda()
netF.eval() # No need to train
return netF
def define_F(opt, use_bn=False):
gpu_ids = opt['gpu_ids']
device = torch.device('cuda' if gpu_ids else 'cpu')
# pytorch pretrained VGG19-54, before ReLU.
if use_bn:
feature_layer = 49
else:
feature_layer = 34
netF = arch.VGGFeatureExtractor(feature_layer=feature_layer, use_bn=use_bn, \
use_input_norm=True, device=device)
if gpu_ids:
netF = nn.DataParallel(netF)
netF.eval() # No need to train
return netF
def define_F(opt, use_bn=False):
gpu_ids = opt["gpu_ids"]
device = torch.device("cuda" if gpu_ids else "cpu")
# pytorch pretrained VGG19-54, before ReLU.
if use_bn:
feature_layer = 49
else:
feature_layer = 34
netF = arch.VGGFeatureExtractor(feature_layer=feature_layer,
use_bn=use_bn,
use_input_norm=True,
device=device)
# netF = arch.ResNet101FeatureExtractor(use_input_norm=True, device=device)
if gpu_ids:
netF = nn.DataParallel(netF)
netF.eval() # No need to train
return netF
def define_F(opt, use_bn=False):
gpu_ids = opt['gpu_ids']
device = torch.device('cuda' if gpu_ids else 'cpu')
# pytorch pretrained VGG19-54, before ReLU.
if use_bn:
feature_layer = 49
else:
feature_layer = 34
# ouyry
if 'network_F' not in opt:
mode = 'VGG19'
else:
mode = opt['network_F']['mode']
if mode == 'VGG19':
netF = arch.VGGFeatureExtractor(feature_layer=feature_layer, use_bn=use_bn, \
use_input_norm=True, device=device)
elif mode == 'VGG16':
feature_layer = 28
netF = arch.VGG16FeatureExtractor(feature_layer=feature_layer, use_bn=use_bn, \
use_input_norm=True, device=device)
elif mode == 'VGG16-MINC':
feature_layer = 28
model_path = opt['network_F']['path']
netF = arch.VGG16MINCFeatureExtractor(feature_layer=feature_layer, model_path=model_path,\
use_bn=use_bn, use_input_norm=True, device=device)
elif mode == 'Sphere20a':
feature_layer = 28
model_path = opt['network_F']['path']
if 'norm' in opt['network_F']:
norm = opt['network_F']['norm']
else:
norm = True
netF = arch.Sphere20aFeatureExtractor(feature_layer=feature_layer, model_path=model_path,\
use_bn=use_bn, use_input_norm=norm, device=device)
# netF = arch.ResNet101FeatureExtractor(use_input_norm=True, device=device)
# if 'distributed' in opt:
# assert torch.cuda.is_available()
# netF = nn.parallel.DistributedDataParallel(netF)
if gpu_ids:
assert torch.cuda.is_available()
netF = nn.DataParallel(netF)
netF.eval() # No need to train
return netF
def define_F(opt, use_bn=False,**kwargs):
gpu_ids = opt['gpu_ids']
device = torch.device('cuda' if gpu_ids else 'cpu')
# pytorch pretrained VGG19-54, before ReLU.
if 'arch' in kwargs.keys()and 'vgg11' in kwargs['arch']:
feature_layer = int(kwargs['arch'][len('vgg11_'):])
kwargs['arch'] = 'vgg11'
else:
if use_bn:
feature_layer = 49
else:
feature_layer = 34
netF = arch.VGGFeatureExtractor(feature_layer=feature_layer, use_bn=use_bn,use_input_norm=True, device=device,**kwargs)
# netF = arch.ResNet101FeatureExtractor(use_input_norm=True, device=device)
if gpu_ids:
netF = nn.DataParallel(netF)
netF.eval() # No need to train
return netF
