def __init__(self):
super(OrdNet, self).__init__()
# Load densenet and modify the forward function to match
# keras.applications' densenet169 when include_top=False (as in the DenseDepth paper)
self.encoder = densenet169(pretrained=True)
self.pools = {}
self.activation = {}
def get_activation(name, activation):
def hook(model, input, output):
activation[name] = output.clone()
return hook
self.encoder.features.relu0.register_forward_hook(get_activation("conv0", self.activation))
self.encoder.features.pool0.register_forward_hook(get_activation("pool0", self.activation))
self.encoder.features.transition1.pool.register_forward_hook(get_activation("pool1", self.activation))
self.encoder.features.transition2.pool.register_forward_hook(get_activation("pool2", self.activation))
# self.encoder.features.transition3.pool.register_forward_hook(get_activation("pool3", self.activation))
# Make Decoder
decoder_channels, conv0_channels, pool0_channels, pool1_channels, pool2_channels = \
self.get_num_encoder_channels()
self.conv2 = nn.Conv2d(in_channels=decoder_channels, out_channels=decoder_channels, kernel_size=1, stride=1, padding=0)
self.up1 = UpBlock(decoder_channels + pool2_channels, decoder_channels//2)
self.up2 = UpBlock(decoder_channels//2 + pool1_channels, decoder_channels//4)
self.up3 = UpBlock(decoder_channels//4 + pool0_channels , decoder_channels//8)
self.up4 = UpBlock(decoder_channels//8 + conv0_channels, decoder_channels//16)
self.conv3 = nn.Conv2d(in_channels=decoder_channels//16, out_channels=1, kernel_size=3, padding=1, stride=1)
def __init__(self,
subtype='densenet121',
out_stages=[2, 3, 4],
backbone_path=None):
super(Densenet, self).__init__()
self.out_stages = out_stages
self.backbone_path = backbone_path
self.out_channels = [64, 128, 256, 512, 1024]
if subtype == 'densenet121':
features = densenet121(pretrained=not backbone_path).features
elif subtype == 'densenet161':
features = densenet161(pretrained=not backbone_path).features
elif subtype == 'densenet169':
features = densenet169(pretrained=not backbone_path).features
elif subtype == 'densenet201':
features = densenet201(pretrained=not backbone_path).features
self.out_channels = self.out_channels[self.out_stages[0]:self.
out_stages[-1] + 1]
self.conv1 = nn.Sequential(features.conv0, features.norm0,
features.relu0, features.pool0)
self.layer1 = nn.Sequential(features.denseblock1, features.transition1)
self.layer2 = nn.Sequential(features.denseblock2, features.transition2)
self.layer3 = nn.Sequential(features.denseblock3, features.transition3)
self.layer4 = features.denseblock4
if self.backbone_path:
self.features.load_state_dict(torch.load(self.backbone_path))
else:
self.init_weights()
def recordDense(info):
global SKIP
import torchvision.models.densenet as denGen
if not (SKIP and 'densenet121' in info['name_list']):
INFO("proceeding for DenseNet121")
net = denGen.densenet121(pretrained=True).cuda()
sum = __summary(net, [3, 224, 224], verbose=True)
__writeInfoJSON(sum, 'densenet121')
else:
INFO("Skip DenseNet121")
if not (SKIP and 'densenet161' in info['name_list']):
INFO("proceeding for DenseNet161")
net = denGen.densenet161(pretrained=True).cuda()
sum = __summary(net, [3, 224, 224], verbose=True)
__writeInfoJSON(sum, 'densenet161')
else:
INFO("Skip DenseNet161")
if not (SKIP and 'densenet169' in info['name_list']):
INFO("proceeding for DenseNet169")
net = denGen.densenet169(pretrained=True).cuda()
sum = __summary(net, [3, 224, 224], verbose=True)
__writeInfoJSON(sum, 'densenet169')
else:
INFO("Skip DenseNet169")
if not (SKIP and 'densenet201' in info['name_list']):
INFO("proceeding for DenseNet201")
net = denGen.densenet201(pretrained=True).cuda()
sum = __summary(net, [3, 224, 224], verbose=True)
__writeInfoJSON(sum, 'densenet201')
else:
INFO("Skip DenseNet201")
def __init__(self,
pretrained=True,
progress=True,
cfg_dropout2d=dict(p=0.1),
**kwargs):
super(densenet169_dropout2d, self).__init__()
backbone = densenet.densenet169(pretrained, progress, **kwargs)
self.backbone = backbone.features
self.dropout2d = nn.Dropout2d(**cfg_dropout2d)
def create_model(model_name, num_classes=1000, pretrained=False, **kwargs):
if 'test_time_pool' in kwargs:
test_time_pool = kwargs.pop('test_time_pool')
else:
test_time_pool = True
if 'extra' in kwargs:
extra = kwargs.pop('extra')
else:
extra = True
if model_name == 'dpn68':
model = dpn68(
num_classes=num_classes, pretrained=pretrained, test_time_pool=test_time_pool)
elif model_name == 'dpn68b':
model = dpn68b(
num_classes=num_classes, pretrained=pretrained, test_time_pool=test_time_pool)
elif model_name == 'dpn92':
model = dpn92(
num_classes=num_classes, pretrained=pretrained, test_time_pool=test_time_pool, extra=extra)
elif model_name == 'dpn98':
model = dpn98(
num_classes=num_classes, pretrained=pretrained, test_time_pool=test_time_pool)
elif model_name == 'dpn131':
model = dpn131(
num_classes=num_classes, pretrained=pretrained, test_time_pool=test_time_pool)
elif model_name == 'dpn107':
model = dpn107(
num_classes=num_classes, pretrained=pretrained, test_time_pool=test_time_pool)
elif model_name == 'resnet18':
model = resnet18(num_classes=num_classes, pretrained=pretrained, **kwargs)
elif model_name == 'resnet34':
model = resnet34(num_classes=num_classes, pretrained=pretrained, **kwargs)
elif model_name == 'resnet50':
model = resnet50(num_classes=num_classes, pretrained=pretrained, **kwargs)
elif model_name == 'resnet101':
model = resnet101(num_classes=num_classes, pretrained=pretrained, **kwargs)
elif model_name == 'resnet152':
model = resnet152(num_classes=num_classes, pretrained=pretrained, **kwargs)
elif model_name == 'densenet121':
model = densenet121(num_classes=num_classes, pretrained=pretrained, **kwargs)
elif model_name == 'densenet161':
model = densenet161(num_classes=num_classes, pretrained=pretrained, **kwargs)
elif model_name == 'densenet169':
model = densenet169(num_classes=num_classes, pretrained=pretrained, **kwargs)
elif model_name == 'densenet201':
model = densenet201(num_classes=num_classes, pretrained=pretrained, **kwargs)
elif model_name == 'inception_v3':
model = inception_v3(
num_classes=num_classes, pretrained=pretrained, transform_input=False, **kwargs)
else:
assert False, "Unknown model architecture (%s)" % model_name
return model
def create_model(model_name, num_classes=1000, pretrained=False, **kwargs):
if 'test_time_pool' in kwargs:
test_time_pool = kwargs.pop('test_time_pool')
else:
test_time_pool = True
if 'extra' in kwargs:
extra = kwargs.pop('extra')
else:
extra = True
if model_name == 'dpn68':
model = dpn68(
num_classes=num_classes, pretrained=pretrained, test_time_pool=test_time_pool)
elif model_name == 'dpn68b':
model = dpn68b(
num_classes=num_classes, pretrained=pretrained, test_time_pool=test_time_pool)
elif model_name == 'dpn92':
model = dpn92(
num_classes=num_classes, pretrained=pretrained, test_time_pool=test_time_pool, extra=extra)
elif model_name == 'dpn98':
model = dpn98(
num_classes=num_classes, pretrained=pretrained, test_time_pool=test_time_pool)
elif model_name == 'dpn131':
model = dpn131(
num_classes=num_classes, pretrained=pretrained, test_time_pool=test_time_pool)
elif model_name == 'dpn107':
model = dpn107(
num_classes=num_classes, pretrained=pretrained, test_time_pool=test_time_pool)
elif model_name == 'resnet18':
model = resnet18(num_classes=num_classes, pretrained=pretrained, **kwargs)
elif model_name == 'resnet34':
model = resnet34(num_classes=num_classes, pretrained=pretrained, **kwargs)
elif model_name == 'resnet50':
model = resnet50(num_classes=num_classes, pretrained=pretrained, **kwargs)
elif model_name == 'resnet101':
model = resnet101(num_classes=num_classes, pretrained=pretrained, **kwargs)
elif model_name == 'resnet152':
model = resnet152(num_classes=num_classes, pretrained=pretrained, **kwargs)
elif model_name == 'densenet121':
model = densenet121(num_classes=num_classes, pretrained=pretrained, **kwargs)
elif model_name == 'densenet161':
model = densenet161(num_classes=num_classes, pretrained=pretrained, **kwargs)
elif model_name == 'densenet169':
model = densenet169(num_classes=num_classes, pretrained=pretrained, **kwargs)
elif model_name == 'densenet201':
model = densenet201(num_classes=num_classes, pretrained=pretrained, **kwargs)
elif model_name == 'inception_v3':
model = inception_v3(
num_classes=num_classes, pretrained=pretrained, transform_input=False, **kwargs)
else:
assert False, "Unknown model architecture (%s)" % model_name
return model
def black_box_densenet169(cuda=True):
black_box_model = densenet169(pretrained=True)
black_box_model.train(False)
if cuda:
# black_box_model = black_box_model.cuda()
black_box_model = torch.nn.DataParallel(black_box_model).cuda()
freeze_model(black_box_model)
def black_box_fn(_images):
# with torch.no_grad():
# print(torch.min(_images[0]), torch.max(_images[0]))
return black_box_model(imagenet_normalize(0.5 * (_images + 1)))
return black_box_fn
def get_densenet_model(densenet, num_classes=-1):
if densenet == "densenet121":
model = densenet121(pretrained=True)
elif densenet == "densenet169":
model = densenet169(pretrained=True)
elif densenet == "densenet201":
model = densenet201(pretrained=True)
elif densenet == "densenet161":
model = densenet161(pretrained=True)
else:
raise RuntimeError("Unsupported model, %s" % densenet)
if num_classes > 0:
in_features = model.classifier.in_features
classifier = nn.Linear(in_features, num_classes, bias=True)
model.classifier = classifier
return model
def __init__(self, pretrained=True, progress=True, **kwargs):
super(densenet169, self).__init__()
backbone = densenet.densenet169(pretrained, progress, **kwargs)
self.backbone = backbone.features
def loss_selector(loss_net):
#base
if loss_net == "vgg16":
from torchvision.models.vgg import vgg16
net = vgg16(pretrained=True)
loss_network = nn.Sequential(*list(net.features)[:31]).eval()
return loss_network
elif loss_net == "vgg16_bn":
from torchvision.models.vgg import vgg16_bn
net = vgg16_bn(pretrained=True)
loss_network = nn.Sequential(*list(net.features)[:44]).eval()
return loss_network
elif loss_net == "resnet50":
from torchvision.models.resnet import resnet50
net=resnet50(pretrained=True)
loss_network=nn.Sequential(*[child_module for child_module in net.children()][:-2]).eval()
return loss_network
elif loss_net == "resnet101":
from torchvision.models.resnet import resnet101
net=resnet101(pretrained=True)
loss_network=nn.Sequential(*[child_module for child_module in net.children()][:-2]).eval()
return loss_network
elif loss_net == "resnet152":
from torchvision.models.resnet import resnet152
net=resnet152(pretrained=True)
loss_network=nn.Sequential(*[child_module for child_module in net.children()][:-2]).eval()
return loss_network
elif loss_net == "squeezenet1_1":
from torchvision.models.squeezenet import squeezenet1_1
net=squeezenet1_1(pretrained=True)
classifier=[item for item in net.classifier.modules()][1:-1]
loss_network=nn.Sequential(*[net.features,*classifier]).eval()
return loss_network
elif loss_net == "densenet121":
from torchvision.models.densenet import densenet121
net=densenet121(pretrained=True)
loss_network=nn.Sequential(*[net.features,nn.ReLU()]).eval()
return loss_network
elif loss_net == "densenet169":
from torchvision.models.densenet import densenet169
net=densenet169(pretrained=True)
loss_network=nn.Sequential(*[net.features,nn.ReLU()]).eval()
return loss_network
elif loss_net == "densenet201":
from torchvision.models.densenet import densenet201
net=densenet201(pretrained=True)
loss_network=nn.Sequential(*[net.features,nn.ReLU()]).eval()
return loss_network
elif loss_net == "mobilenet_v2":
from torchvision.models.mobilenet import mobilenet_v2
net=mobilenet_v2(pretrained=True)
loss_network=nn.Sequential(*[net.features]).eval()
return loss_network
elif loss_net == "resnext50_32x4d":
from torchvision.models.resnet import resnext50_32x4d
net=resnext50_32x4d(pretrained=True)
loss_network=nn.Sequential(*[child_module for child_module in net.children()][:-2]).eval()
return loss_network
elif loss_net == "resnext101_32x8d":
from torchvision.models.resnet import resnext101_32x8d
net=resnext101_32x8d(pretrained=True)
loss_network=nn.Sequential(*[child_module for child_module in net.children()][:-2]).eval()
return loss_network
elif loss_net == "wide_resnet50_2":
from torchvision.models.resnet import wide_resnet50_2
net=wide_resnet50_2(pretrained=True)
loss_network=nn.Sequential(*[child_module for child_module in net.children()][:-2]).eval()
return loss_network
elif loss_net == "wide_resnet101_2":
from torchvision.models.resnet import wide_resnet101_2
net=wide_resnet101_2(pretrained=True)
loss_network=nn.Sequential(*[child_module for child_module in net.children()][:-2]).eval()
return loss_network
elif loss_net == "inception_v3":
test_loss /= size
correct /= size
# Report it to the the console
print(
f"Test Error: \n Accuracy: {(100*correct):>0.1f}%, Avg loss: {test_loss:>8f} \n"
)
# Return a dictionary containing the results (for use in plotting etc.), list wrapped for dataframe usage
return {"loss": [test_loss], "accuracy": [correct]}
if __name__ == '__main__':
train_epochs = 50
device = "cpu"
model = densenet169()
training_data = CIFAR100Coarse(root="data",
train=True,
download=True,
transform=data_managers.train_transform)
testing_data = CIFAR100Coarse(root="data",
train=False,
download=True,
transform=data_managers.test_transform)
# Lock the random seed here to make the batches consistent between tests
torch.manual_seed(36246)
# Split the training data into sets of 5000 entries
training_sets = random_split(training_data, [5000] * 10)
def densenet_bc169(nb_classes=1000, pretrained=False):
return densenet169(num_classes=nb_classes, pretrained=pretrained)
from torchvision.models.densenet import _DenseBlock, _Transition
#import code
from copy import deepcopy
from collections import OrderedDict
# Project imports
from util.util import print # print with a header for this project
# Leveraged Torchvision implementation of DenseNet at:
# https://github.com/pytorch/vision/blob/master/torchvision/models/densenet.py
_use_pretrained = True
name_to_model_fns = {
"densenet121": lambda: densenet.densenet121(pretrained=_use_pretrained),
"densenet169": lambda: densenet.densenet169(pretrained=_use_pretrained),
"densenet201": lambda: densenet.densenet201(
pretrained=_use_pretrained) # We're just going to use this one
,
"densenet161": lambda: densenet.densenet161(pretrained=_use_pretrained)
}
def get_densenet(name):
return name_to_model_fns[name]() # Call the function
"""
old_model.modules is more accurate of all indiv. layers
old_model.features decomposes structure more, but doesn't include final FC classifier
(instead under old_model.classifier)