def load_vgg16(model_dir):
""" Use the model from https://github.com/abhiskk/fast-neural-style/blob/master/neural_style/utils.py """
if not os.path.exists(model_dir):
os.mkdir(model_dir)
if not os.path.exists(os.path.join(model_dir, 'vgg16.weight')):
if not os.path.exists(os.path.join(model_dir, 'vgg16.t7')):
os.system('wget https://www.dropbox.com/s/76l3rt4kyi3s8x7/vgg16.t7?dl=1 -O ' + os.path.join(model_dir, 'vgg16.t7'))
vgglua = torchfile.load(os.path.join(model_dir, 'vgg16.t7'))
vgg = Vgg16()
for (src, dst) in zip(vgglua.parameters()[0], vgg.parameters()):
dst.data[:] = src
torch.save(vgg.state_dict(), os.path.join(model_dir, 'vgg16.weight'))
vgg = Vgg16()
vgg.load_state_dict(torch.load(os.path.join(model_dir, 'vgg16.weight')))
return vgg
def init_vgg16(model_dir):
if not os.path.exists(os.path.join(model_dir, 'vgg16.weight')):
if not os.path.exists(os.path.join(model_dir, 'vgg16.t7')):
os.system(
'wget https://www.dropbox.com/s/76l3rt4kyi3s8x7/vgg16.t7?dl=1 -O ' + os.path.join(model_dir, 'vgg16.t7'))
vgglua = torchfile.load(os.path.join(model_dir, 'vgg16.t7'))
vgg = Vgg16()
for (src, dst) in zip(vgglua.parameters()[0], vgg.parameters()):
dst.data[:] = src
torch.save(vgg.state_dict(), os.path.join(model_dir, 'vgg16.weight'))
def load_vgg16(model_dir):
""" Use the model from https://github.com/abhiskk/fast-neural-style/blob/master/neural_style/utils.py """
if not os.path.exists(model_dir):
os.mkdir(model_dir)
if not os.path.exists(os.path.join(model_dir, 'vgg16.weight')):
if not os.path.exists(os.path.join(model_dir, 'vgg16.t7')):
os.system(
'wget https://www.dropbox.com/s/76l3rt4kyi3s8x7/vgg16.t7?dl=1 -O '
+ os.path.join(model_dir, 'vgg16.t7'))
# vgglua = load_lua(os.path.join(model_dir, 'vgg16.t7')) /media/HDD1/convert_torch_to_pytorch/vgg16.pth
vgglua = torch.load('/media/HDD1/convert_torch_to_pytorch/vgg16.pth')
# vgglua = torchfile.load(os.path.join(model_dir, 'vgg16.t7'))
vgg = Vgg16()
value_list = []
for name, v in vgglua.items():
value_list.append(v)
for (src, dst) in zip(value_list, vgg.parameters()):
dst.data[:] = src
torch.save(vgg.state_dict(), os.path.join(model_dir, 'vgg16.weight'))
vgg = Vgg16()
vgg.load_state_dict(torch.load(os.path.join(model_dir, 'vgg16.weight')))
return vgg
def inference(config):
hair_model, skin_model = load_model(config)
#train_loader, val_loader = get_loaders(hair_model, skin_model, config)
try:
your_pic = Image.open(config.your_pic)
celeb_pic = Image.open(config.celeb_pic)
except:
return
your_pic, your_pic_mask, celeb_pic, celeb_pic_mask = DataLoader(
transform(your_pic, celeb_pic, config.image_size, hair_model,
skin_model, config.device))
# Initialize
vgg = Vgg16().to(config.device)
resnet = ResNet18(requires_grad=True, pretrained=True).to(config.device)
generator = GeneratorUNet().to(config.device)
# discriminator = Discriminator().to(config.device)
try:
resnet.load_state_dict(
torch.load(
os.path.join(config.checkpoints,
'epoch_%d_%s.pth' % (20, 'resnet'))))
generator.load_state_dict(
torch.load(
os.path.join(config.checkpoints,
'epoch_%d_%s.pth' % (20, 'generator'))))
except OSError:
print('Check if your pretrained weight is in the right place.')
z1 = resnet(your_pic * your_pic_mask) #skin
z2 = resnet(celeb_pic * celeb_pic_mask) #hair
fake_im = generator(your_pic, z1, z2) # z1 is skin, z2 is hair
images = [your_pic[0], celeb_pic[0], fake_im[0]]
titles = ['Your picture', 'Celebrity picture', 'Synthesized picture']
fig, axes = plt.subplots(1, len(titles))
for i in range(len(images)):
im = images[i]
im = im.data.cpu().numpy().transpose(1, 2, 0)
im = (im + 1) / 2
axes[i].imshow(im)
axes[i].axis('off')
axes[i].set_title(titles[i])
plt.show()
def load_vgg16(model_dir):
""" Use the model from https://github.com/abhiskk/fast-neural-style/blob/master/neural_style/utils.py """
#from torch.utils.serialization import load_lua
import torchvision.models as models
vgg16 = models.vgg16(pretrained=True)
vgg = vgg16.features[:-1]
return vgg
import torchfile
if not os.path.exists(model_dir):
os.mkdir(model_dir)
if not os.path.exists(os.path.join(model_dir, 'vgg16.weight')):
if not os.path.exists(os.path.join(model_dir, 'vgg16.t7')):
os.system(
'wget https://www.dropbox.com/s/76l3rt4kyi3s8x7/vgg16.t7?dl=1 -O '
+ os.path.join(model_dir, 'vgg16.t7'))
vgglua = torchfile.load(os.path.join(model_dir, 'vgg16.t7'))
vgg = Vgg16()
for (src, dst) in zip(vgglua.parameters()[0], vgg.parameters()):
dst.data[:] = src
torch.save(vgg.state_dict(), os.path.join(model_dir, 'vgg16.weight'))
vgg = Vgg16()
vgg.load_state_dict(torch.load(os.path.join(model_dir, 'vgg16.weight')))
return vgg
import torch
import os
from torch.utils.serialization import load_lua
from torchvision.models import resnet18
from networks import Vgg16, ResNet
import torch.utils.model_zoo as model_zoo
model_dir = 'models/'
with open('lua.txt', 'w') as luaf:
vgglua = load_lua(os.path.join(model_dir, 'vgg16.t7'))
vgg = Vgg16()
for (src, dst) in zip(vgglua.parameters()[0], vgg.parameters()):
print(src, file=luaf)
input()
dst.data[:] = src
resnet = ResNet()
resnet.load_state_dict(
model_zoo.load_url(
'https://download.pytorch.org/models/resnet18-5c106cde.pth',
model_dir))
def main(trainfile,
validfile,
batch_size,
epochs,
valid_frequency=2,
write_frequency=1000,
checkpoint_path=None,
logdir='logs/',
model_root='checkpoints'):
# Datasets
trainset = DatasetFromImageLabelList(trainfile, transform['train'],
target_transform)
validset = DatasetFromImageLabelList(validfile, transform['valid'],
target_transform)
# Dataloader
trainloader = DataLoader(trainset,
batch_size=batch_size,
shuffle=True,
num_workers=4)
validloader = DataLoader(validset,
batch_size=batch_size * 2,
shuffle=False,
num_workers=4)
# # Test trainloader
# dataiter = iter(trainloader)
# images, targets = dataiter.next()
# samples_img = transforms.ToPILImage()(torchvision.utils.make_grid(images))
# samples_img.save('samples.png')
# SummaryWriter
writer = SummaryWriter(log_dir=logdir, flush_secs=5)
# Model
model = Vgg16(num_classes=10, pretrained=True)
# Writer model graph
writer.add_graph(model, torch.rand((1, 3, 224, 224), dtype=torch.float32))
# Loss function
criterion = nn.CrossEntropyLoss()
# Optimizer
optimizer = optim.Adam(model.parameters(), lr=1e-2)
# learning rate scheduler
scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer, 'min')
# Available device
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# Model to cuda
model.to(device)
# Global step counter
steps = 1
# accracy_list and loss_list
loss_record = []
accuracy_record = []
# Minist loss and Best accuracy
minist_loss = 1
best_accuracy = 0
# Start epoch
start_epoch = 0
# Loaing model
if checkpoint_path:
start_epoch = load_model_state_dict(checkpoint_path, model, optimizer)
# Train epochs
for epoch in range(start_epoch, epochs):
# Train mode
model.train()
# Clear record
loss_record.clear()
accuracy_record.clear()
# Train one epoch
for inputs, targets in trainloader:
# Data to device
inputs, targets = inputs.to(device), targets.to(device)
# Model outputs
outputs = model(inputs)
# Caculate batch loss
batch_loss = criterion(outputs, targets)
# Zero grad before backward
optimizer.zero_grad()
# Backward
batch_loss.backward()
# Optimizer weights
optimizer.step()
# Adjust learning rate by loss
scheduler.step(batch_loss)
# Caculate accuracy
batch_accuracy = caculate_accuracy(outputs, targets)
# Record loss and accuracy
loss_record.append(batch_loss.item())
accuracy_record.append(batch_accuracy)
# print('epoch: {}, steps: {}, loss: {:.5f}, accuracy: {:.5f}'.format(
# epoch, steps, np.array(loss_record).mean(), np.array(accuracy_record).mean()))
if steps % write_frequency == 0:
loss = np.array(loss_record).mean()
accuracy = np.array(accuracy_record).mean()
# writer.add_images('train sample', inputs, global_step=steps)
writer.add_scalar('train loss', loss, global_step=steps)
writer.add_scalar('train accuracy',
accuracy,
global_step=steps)
# Add steps
steps += 1
if epoch % valid_frequency == 0:
# Inference mode
model.eval()
with torch.no_grad():
for inputs, targets in validloader:
inputs, targets = inputs.to(device), targets.to(device)
outputs = model(inputs)
batch_loss = criterion(outputs, targets)
batch_accuracy = caculate_accuracy(outputs, targets)
loss_record.append(batch_loss.item())
accuracy_record.append(batch_accuracy)
loss = np.array(loss_record).mean()
accuracy = np.array(accuracy_record).mean()
writer.add_scalar('valid loss', loss, global_step=steps)
writer.add_scalar('valid accuracy',
accuracy,
global_step=steps)
print('valid:', epoch, steps, loss, accuracy)
# 检查点保存条件
if loss <= minist_loss and accuracy >= best_accuracy:
model.to('cpu')
save_model_state_dict(model_root, model, optimizer, epoch,
loss, accuracy, 'vgg16')
model.to(device)
def main(config):
model = load_model(config)
train_loader, val_loader = get_loaders(model, config)
# Make dirs
if not os.path.exists(config.checkpoints):
os.makedirs(config.checkpoints, exist_ok=True)
if not os.path.exists(config.save_path):
os.makedirs(config.save_path, exist_ok=True)
# Loss Functions
criterion_GAN = mse_loss
# Calculate output of image discriminator (PatchGAN)
patch = (1, config.image_size // 2**4, config.image_size // 2**4)
# Initialize
vgg = Vgg16().to(config.device)
resnet = ResNet18(requires_grad=True, pretrained=True).to(config.device)
generator = GeneratorUNet().to(config.device)
discriminator = Discriminator().to(config.device)
if config.epoch != 0:
# Load pretrained models
resnet.load_state_dict(
torch.load(
os.path.join(config.checkpoints, 'epoch_%d_%s.pth' %
(config.epoch - 1, 'resnet'))))
generator.load_state_dict(
torch.load(
os.path.join(
config.checkpoints,
'epoch_%d_%s.pth' % (config.epoch - 1, 'generator'))))
discriminator.load_state_dict(
torch.load(
os.path.join(
config.checkpoints,
'epoch_%d_%s.pth' % (config.epoch - 1, 'discriminator'))))
else:
# Initialize weights
# resnet.apply(weights_init_normal)
generator.apply(weights_init_normal)
discriminator.apply(weights_init_normal)
# Optimizers
optimizer_resnet = torch.optim.Adam(resnet.parameters(),
lr=config.lr,
betas=(config.b1, config.b2))
optimizer_G = torch.optim.Adam(generator.parameters(),
lr=config.lr,
betas=(config.b1, config.b2))
optimizer_D = torch.optim.Adam(discriminator.parameters(),
lr=config.lr,
betas=(config.b1, config.b2))
# ----------
# Training
# ----------
resnet.train()
generator.train()
discriminator.train()
for epoch in range(config.epoch, config.n_epochs):
for i, (im1, m1, im2, m2) in enumerate(train_loader):
assert im1.size(0) == im2.size(0)
valid = Variable(torch.Tensor(np.ones(
(im1.size(0), *patch))).to(config.device),
requires_grad=False)
fake = Variable(torch.Tensor(np.ones(
(im1.size(0), *patch))).to(config.device),
requires_grad=False)
# ------------------
# Train Generators
# ------------------
optimizer_resnet.zero_grad()
optimizer_G.zero_grad()
# GAN loss
z = resnet(im2 * m2)
if epoch < config.gan_epochs:
fake_im = generator(im1 * (1 - m1), im2 * m2, z)
else:
fake_im = generator(im1, im2, z)
if epoch < config.gan_epochs:
pred_fake = discriminator(fake_im, im2)
gan_loss = config.lambda_gan * criterion_GAN(pred_fake, valid)
else:
gan_loss = torch.Tensor([0]).to(config.device)
# Hair, Face loss
fake_m2 = torch.argmax(model(fake_im),
1).unsqueeze(1).type(torch.uint8).repeat(
1, 3, 1, 1).to(config.device)
if 0.5 * torch.sum(m1) <= torch.sum(
fake_m2) <= 1.5 * torch.sum(m1):
hair_loss = config.lambda_style * calc_style_loss(
fake_im * fake_m2, im2 * m2, vgg) + calc_content_loss(
fake_im * fake_m2, im2 * m2, vgg)
face_loss = calc_content_loss(fake_im, im1, vgg)
else:
hair_loss = config.lambda_style * calc_style_loss(
fake_im * m1, im2 * m2, vgg) + calc_content_loss(
fake_im * m1, im2 * m2, vgg)
face_loss = calc_content_loss(fake_im, im1, vgg)
hair_loss *= config.lambda_hair
face_loss *= config.lambda_face
# Total loss
loss = gan_loss + hair_loss + face_loss
loss.backward()
optimizer_resnet.step()
optimizer_G.step()
# ---------------------
# Train Discriminator
# ---------------------
if epoch < config.gan_epochs:
optimizer_D.zero_grad()
# Real loss
pred_real = discriminator(im1 * (1 - m1) + im2 * m2, im2)
loss_real = criterion_GAN(pred_real, valid)
# Fake loss
pred_fake = discriminator(fake_im.detach(), im2)
loss_fake = criterion_GAN(pred_fake, fake)
# Total loss
loss_D = 0.5 * (loss_real + loss_fake)
loss_D.backward()
optimizer_D.step()
if i % config.sample_interval == 0:
msg = "Train || Gan loss: %.6f, hair loss: %.6f, face loss: %.6f, loss: %.6f\n" % \
(gan_loss.item(), hair_loss.item(), face_loss.item(), loss.item())
sys.stdout.write("Epoch: %d || Batch: %d\n" % (epoch, i))
sys.stdout.write(msg)
fname = os.path.join(
config.save_path,
"Train_Epoch:%d_Batch:%d.png" % (epoch, i))
sample_images([im1[0], im2[0], fake_im[0]],
["img1", "img2", "img1+img2"], fname)
for j, (im1, m1, im2, m2) in enumerate(val_loader):
with torch.no_grad():
valid = Variable(torch.Tensor(
np.ones((im1.size(0), *patch))).to(config.device),
requires_grad=False)
fake = Variable(torch.Tensor(
np.ones((im1.size(0), *patch))).to(config.device),
requires_grad=False)
# GAN loss
z = resnet(im2 * m2)
if epoch < config.gan_epochs:
fake_im = generator(im1 * (1 - m1), im2 * m2, z)
else:
fake_im = generator(im1, im2, z)
if epoch < config.gan_epochs:
pred_fake = discriminator(fake_im, im2)
gan_loss = config.lambda_gan * criterion_GAN(
pred_fake, valid)
else:
gan_loss = torch.Tensor([0]).to(config.device)
# Hair, Face loss
fake_m2 = torch.argmax(
model(fake_im),
1).unsqueeze(1).type(torch.uint8).repeat(
1, 3, 1, 1).to(config.device)
if 0.5 * torch.sum(m1) <= torch.sum(
fake_m2) <= 1.5 * torch.sum(m1):
hair_loss = config.lambda_style * calc_style_loss(
fake_im * fake_m2, im2 * m2,
vgg) + calc_content_loss(
fake_im * fake_m2, im2 * m2, vgg)
face_loss = calc_content_loss(fake_im, im1, vgg)
else:
hair_loss = config.lambda_style * calc_style_loss(
fake_im * m1, im2 * m2,
vgg) + calc_content_loss(
fake_im * m1, im2 * m2, vgg)
face_loss = calc_content_loss(fake_im, im1, vgg)
hair_loss *= config.lambda_hair
face_loss *= config.lambda_face
# Total loss
loss = gan_loss + hair_loss + face_loss
msg = "Validation || Gan loss: %.6f, hair loss: %.6f, face loss: %.6f, loss: %.6f\n" % \
(gan_loss.item(), hair_loss.item(), face_loss.item(), loss.item())
sys.stdout.write(msg)
fname = os.path.join(
config.save_path,
"Validation_Epoch:%d_Batch:%d.png" % (epoch, i))
sample_images([im1[0], im2[0], fake_im[0]],
["img1", "img2", "img1+img2"], fname)
break
if epoch % config.checkpoint_interval == 0:
if epoch < config.gan_epochs:
models = [resnet, generator, discriminator]
fnames = ['resnet', 'generator', 'discriminator']
else:
models = [resnet, generator]
fnames = ['resnet', 'generator']
fnames = [
os.path.join(config.checkpoints,
'epoch_%d_%s.pth' % (epoch, s)) for s in fnames
]
save_weights(models, fnames)
def load_vgg16(model_path):
vgg = Vgg16()
vgg.load_state_dict(torch.load(model_path))
return vgg