def main():
# init random seed
init_random_seed(params.manual_seed)
# Load dataset
svhn_data_loader = get_svhn(split='train', download=True)
svhn_data_loader_eval = get_svhn(split='test', download=True)
mnist_data_loader = get_mnist(train=True, download=True)
mnist_data_loader_eval = get_mnist(train=False, download=True)
# Model init WDGRL
tgt_encoder = model_init(Encoder(), params.encoder_wdgrl_path)
critic = model_init(Discriminator(in_dims=params.d_in_dims,
h_dims=params.d_h_dims,
out_dims=params.d_out_dims),
params.disc_wdgrl_path)
clf = model_init(Classifier(), params.clf_wdgrl_path)
# Train critic to optimality
print("====== Training critic ======")
if not (critic.pretrained and params.model_trained):
critic = train_critic_wdgrl(tgt_encoder, critic, svhn_data_loader, mnist_data_loader)
# Train target encoder
print("====== Training encoder for both SVHN and MNIST domains ======")
if not (tgt_encoder.pretrained and clf.pretrained and params.model_trained):
tgt_encoder, clf = train_tgt_wdgrl(tgt_encoder, clf, critic,
svhn_data_loader, mnist_data_loader, robust=False)
# Eval target encoder on test set of target dataset
print("====== Evaluating classifier for encoded SVHN and MNIST domains ======")
print("-------- SVHN domain --------")
eval_tgt(tgt_encoder, clf, svhn_data_loader_eval)
print("-------- MNIST adaption --------")
eval_tgt(tgt_encoder, clf, mnist_data_loader_eval)
def main():
# init random seed
init_random_seed(params.manual_seed)
# Load dataset
mnist_data_loader = get_mnist(train=True, download=True)
mnist_data_loader_eval = get_mnist(train=False, download=True)
usps_data_loader = get_usps(train=True, download=True)
usps_data_loader_eval = get_usps(train=False, download=True)
# Model init Revgard
tgt_encoder = model_init(Encoder(), params.tgt_encoder_revgrad_path)
critic = model_init(Discriminator(), params.disc_revgard_path)
clf = model_init(Classifier(), params.clf_revgrad_path)
# Train models
print("====== Training source encoder and classifier in MNIST and USPS domains ======")
if not (tgt_encoder.pretrained and clf.pretrained and critic.pretrained and params.model_trained):
tgt_encoder, clf, critic = train_revgrad(tgt_encoder, clf, critic,
mnist_data_loader, usps_data_loader, robust=False)
# Eval target encoder on test set of target dataset
print("====== Evaluating classifier for encoded MNIST and USPS domain ======")
print("-------- MNIST domain --------")
eval_tgt(tgt_encoder, clf, mnist_data_loader_eval)
print("-------- USPS adaption --------")
eval_tgt(tgt_encoder, clf, usps_data_loader_eval)
def main():
# init random seed
init_random_seed(params.manual_seed)
# Load dataset
svhn_data_loader = get_svhn(split='train', download=True)
svhn_data_loader_eval = get_svhn(split='test', download=True)
mnist_data_loader = get_mnist(train=True, download=True)
mnist_data_loader_eval = get_mnist(train=False, download=True)
# Model init DANN
tgt_encoder = model_init(Encoder(), params.tgt_encoder_dann_rb_path)
critic = model_init(
Discriminator(in_dims=params.d_in_dims,
h_dims=params.d_h_dims,
out_dims=params.d_out_dims), params.disc_dann_path)
clf = model_init(Classifier(), params.clf_dann_path)
# Train models
print(
"====== Training source encoder and classifier in SVHN and MNIST domains ======"
)
if not (tgt_encoder.pretrained and clf.pretrained and critic.pretrained
and params.model_trained):
tgt_encoder, clf, critic = train_dann(tgt_encoder,
clf,
critic,
svhn_data_loader,
mnist_data_loader,
mnist_data_loader_eval,
robust=False)
# Eval target encoder on test set of target dataset
print(
"====== Evaluating classifier for encoded SVHN and MNIST domain ======"
)
print("-------- SVHN domain --------")
eval_tgt(tgt_encoder, clf, svhn_data_loader_eval)
print("-------- MNIST adaption --------")
eval_tgt(tgt_encoder, clf, mnist_data_loader_eval)
def main():
# init random seed
init_random_seed(params.manual_seed)
# Load dataset
mnist_data_loader = get_usps(train=True, download=True)
mnist_data_loader_eval = get_usps(train=False, download=True)
usps_data_loader = get_usps(train=True, download=True)
usps_data_loader_eval = get_usps(train=False, download=True)
# Model init ADDA
src_encoder = model_init(Encoder(), params.src_encoder_adda_rb_path)
tgt_encoder = model_init(Encoder(), params.tgt_encoder_adda_rb_path)
critic = model_init(Discriminator(), params.disc_adda_rb_path)
clf = model_init(Classifier(), params.clf_adda_rb_path)
# Train source model for adda
print(
"====== Robust training source encoder and classifier in MNIST domain ======"
)
if not (src_encoder.pretrained and clf.pretrained
and params.model_trained):
src_encoder, clf = train_src_robust(src_encoder, clf,
mnist_data_loader)
# Eval source model
print("====== Evaluating classifier for MNIST domain ======")
eval_tgt(src_encoder, clf, mnist_data_loader_eval)
# Train target encoder
print("====== Robust training encoder for USPS domain ======")
# Initialize target encoder's weights with those of the source encoder
if not tgt_encoder.pretrained:
tgt_encoder.load_state_dict(src_encoder.state_dict())
if not (tgt_encoder.pretrained and critic.pretrained
and params.model_trained):
tgt_encoder = train_tgt_adda(src_encoder,
tgt_encoder,
clf,
critic,
mnist_data_loader,
usps_data_loader,
usps_data_loader_eval,
robust=True)
# Eval target encoder on test set of target dataset
print("====== Ealuating classifier for encoded USPS domain ======")
print("-------- Source only --------")
eval_tgt(src_encoder, clf, usps_data_loader_eval)
print("-------- Domain adaption --------")
eval_tgt(tgt_encoder, clf, usps_data_loader_eval)
def train_src_robust(encoder, classifier, data_loader, mode='ADDA'):
"""Train classifier for source domain with robust training for ADDA"""
# Step 1: Network setup
# Set train state for both Dropout and BN layers
encoder.train()
classifier.train()
# Set up optimizer and criterion
optimizer = optim.Adam(list(encoder.parameters()) +
list(classifier.parameters()),
lr=params.learning_rate,
weight_decay=params.weight_decay)
criterion = nn.CrossEntropyLoss()
num_epochs = params.num_epochs_pre if mode == 'ADDA' else params.num_epochs
# Step 2: Pretrain the source model
for epoch in range(num_epochs):
# Init accuracy and loss
start_time = time.time()
train_loss, train_acc, train_n = 0, 0, 0
train_robust_loss, train_robust_acc = 0, 0
for step, (images, labels) in enumerate(data_loader):
# Make images and labels variable
images = make_variable(images)
labels = make_variable(labels)
# Zero gradients for optimizer
optimizer.zero_grad()
delta = attack_pgd(encoder, classifier, images, labels)
# Compute loss for critic with attack img
robust_images = normalize(
torch.clamp(images + delta[:images.size(0)],
min=params.lower_limit,
max=params.upper_limit))
robust_preds = classifier(encoder(robust_images))
robust_loss = criterion(robust_preds, labels)
# Optimize source classifier
robust_loss.backward()
optimizer.step()
# Compute loss for critic with original image
preds = classifier(encoder(images))
loss = criterion(preds, labels)
train_robust_loss += robust_loss.item() * labels.size(0)
train_robust_acc += torch.sum(
robust_preds.max(1)[1] == labels).double()
train_loss += loss.item() * labels.size(0)
train_acc += torch.sum(preds.max(1)[1] == labels.data).double()
train_n += labels.size(0)
# Print step info
if (step + 1) % params.log_step_pre == 0:
print(
"Epoch [{}/{}] Step [{}/{}]: Avg Training loss: {:.4f} Avg Training Accuracy: {:.4%}"
" Avg Robust Training Loss: {:.4f} Avg Robust Training Accuracy: {:.4%}"
.format(epoch + 1, num_epochs, step + 1, len(data_loader),
train_loss / train_n, train_acc / train_n,
train_robust_loss / train_n,
train_robust_acc / train_n))
time_elapsed = time.time() - start_time
# Eval model on test set
if (epoch + 1) % params.eval_step_pre == 0:
eval_tgt(encoder, classifier, data_loader)
# Save model parameters
if (epoch + 1) % params.save_step_pre == 0:
print('Epoch [{}/{}] completed in {:.0f}m {:.0f}s'.format(
epoch + 1, num_epochs, time_elapsed // 60, time_elapsed % 60))
root = params.adda_root if mode == 'ADDA' else params.model_root
save_model(encoder, root,
"{}-source-encoder-rb-{}.pt".format(mode, epoch + 1))
save_model(classifier, root,
"{}-source-classifier-rb-{}.pt".format(mode, epoch + 1))
# Save final model
root = params.adda_root if mode == 'ADDA' else params.model_root
save_model(encoder, root, "{}-source-encoder-rb-final.pt".format(mode))
save_model(classifier, root,
"{}-source-classifier-rb-final.pt".format(mode))
return encoder, classifier
def train_tgt_adda(src_encoder,
tgt_encoder,
classifier,
critic,
src_data_loader,
tgt_data_loader,
tgt_data_loader_eval,
robust=False):
"""Train adda encoder for target domain """
# Step 1: Network Setup
# Set train state for Dropout and BN layers
src_encoder.eval()
tgt_encoder.train()
critic.train()
# Setup criterion and optimizer
criterion = nn.CrossEntropyLoss()
optimizer_tgt = optim.Adam(tgt_encoder.parameters(),
lr=params.learning_rate,
weight_decay=params.weight_decay)
optimizer_critic = optim.Adam(critic.parameters(),
lr=params.learning_rate,
weight_decay=params.weight_decay)
len_data_loader = min(len(src_data_loader), len(tgt_data_loader))
# Step 2 Train network
for epoch in range(params.num_epochs):
start_time = time.time()
train_disc_loss, train_disc_acc, train_n = 0, 0, 0
# Zip source and target data pair
data_zip = enumerate(zip(src_data_loader, tgt_data_loader))
for step, ((images_src, _), (images_tgt, _)) in data_zip:
# 2.1 train discriminator with fixed src_encoder
# Make images variable
images_src = make_variable(images_src)
images_tgt = make_variable(images_tgt)
# Prepare real and fake label (domain labels)
domain_src = make_variable(torch.ones(images_src.size(0)).long())
domain_tgt = make_variable(torch.zeros(images_tgt.size(0)).long())
domain_concat = torch.cat((domain_src, domain_tgt), 0)
if robust:
# Attack images with domain labels
delta_src = attack_pgd(src_encoder, critic, images_src,
domain_src)
delta_tgt = attack_pgd(tgt_encoder, critic, images_tgt,
domain_tgt)
robust_src = normalize(
torch.clamp(images_src + delta_src[:images_src.size(0)],
min=params.lower_limit,
max=params.upper_limit))
robust_tgt = normalize(
torch.clamp(images_tgt + delta_tgt[:images_tgt.size(0)],
min=params.lower_limit,
max=params.upper_limit))
# Zero gradients for optimizer for the discriminator
optimizer_critic.zero_grad()
# Extract and concat features
feat_src = src_encoder(images_src) if not robust else src_encoder(
robust_src)
feat_tgt = tgt_encoder(images_tgt) if not robust else tgt_encoder(
robust_tgt)
feat_concat = torch.cat((feat_src, feat_tgt), 0)
# Predict on discriminator
preds_src_domain = critic(feat_src)
preds_tgt_domain = critic(feat_tgt)
# pred_concat = critic(feat_concat)
# Compute loss for critic
l1 = criterion(preds_src_domain, domain_src)
l2 = criterion(preds_tgt_domain, domain_tgt)
# loss_critic = criterion((pred_concat, domain_concat)
loss_critic = l1 + l2
train_disc_loss += loss_critic.item() * domain_concat.size(0)
# train_disc_acc += torch.sum(pred_concat.max(1)[1] == domain_concat.data).double()
train_disc_acc += torch.sum(
preds_src_domain.max(1)[1] == domain_src.data).double()
train_disc_acc += torch.sum(
preds_tgt_domain.max(1)[1] == domain_tgt.data).double()
train_n += domain_concat.size(0)
loss_critic.backward()
# Optimize critic
optimizer_critic.step()
# 2.2 Train target encoder
# Zero gradients for optimizer
optimizer_critic.zero_grad()
optimizer_tgt.zero_grad()
# Prepare fake labels (flip labels)
domain_tgt = make_variable(torch.ones(feat_tgt.size(0)).long())
if robust:
# Attack the target images with domain labels
delta_tgt = attack_pgd(tgt_encoder, critic, images_tgt,
domain_tgt)
robust_tgt = normalize(
torch.clamp(images_tgt + delta_tgt[:images_tgt.size(0)],
min=params.lower_limit,
max=params.upper_limit))
# Extract target features
feat_tgt = tgt_encoder(images_tgt) if not robust else tgt_encoder(
robust_tgt)
# Predict on discriminator
pred_tgt = critic(feat_tgt)
# Compute loss for target encoder
loss_tgt = criterion(pred_tgt, domain_tgt)
loss_tgt.backward()
# Optimize target encoder
optimizer_tgt.step()
# 2.3 Print step info
if (step + 1) % params.log_step == 0:
print(
"Epoch [{}/{}] Step [{}/{}]: "
"Avg Discriminator Loss: {:.4f} Avg Discriminator Accuracy: {:.4%}"
.format(epoch + 1, params.num_epochs, step + 1,
len_data_loader, train_disc_loss / train_n,
train_disc_acc / train_n))
time_elapsed = time.time() - start_time
# Eval model
if (epoch + 1) % params.eval_step == 0:
if not robust:
eval_tgt(tgt_encoder, classifier, tgt_data_loader_eval)
else:
eval_tgt_robust(tgt_encoder, classifier, critic,
tgt_data_loader_eval)
# 2.4 Save model parameters #
if (epoch + 1) % params.save_step == 0:
print('Epoch [{}/{}] completec in {:.0f}m {:.0f}s'.format(
epoch + 1, params.num_epochs, time_elapsed // 60,
time_elapsed % 60))
filename = "ADDA-critic-{}.pt".format(epoch + 1) if not robust \
else "ADDA-critic-rb-{}.pt".format(epoch + 1)
save_model(critic, params.adda_root, filename)
filename = "ADDA-target-encoder-{}.pt".format(epoch + 1) if not robust \
else "ADDA-target-encoder-rb-{}.pt".format(epoch + 1)
save_model(tgt_encoder, params.adda_root, filename)
filename = "ADDA-critic-final.pt" if not robust else "ADDA-critic-rb-final.pt"
save_model(critic, params.adda_root, filename)
filename = "ADDA-target-encoder-final.pt" if not robust else "ADDA-target-encoder-rb-final.pt"
save_model(tgt_encoder, params.adda_root, filename)
return tgt_encoder
def main():
# init random seed
init_random_seed(params.manual_seed)
# Load dataset
mnist_data_loader = get_mnist(train=True, download=True)
mnist_data_loader_eval = get_mnist(train=False, download=True)
usps_data_loader = get_usps(train=True, download=True)
usps_data_loader_eval = get_usps(train=False, download=True)
# Model init DANN
tgt_encoder = model_init(Encoder(), params.tgt_encoder_dann_rb_path)
critic = model_init(Discriminator(), params.disc_dann_rb_path)
clf = model_init(Classifier(), params.clf_dann_rb_path)
# Train models
print(
"====== Robust Training source encoder and classifier in MNIST and USPS domains ======"
)
if not (tgt_encoder.pretrained and clf.pretrained and critic.pretrained
and params.model_trained):
tgt_encoder, clf, critic = train_dann(tgt_encoder,
clf,
critic,
mnist_data_loader,
usps_data_loader,
usps_data_loader_eval,
robust=False)
# Eval target encoder on test set of target dataset
print(
"====== Evaluating classifier for encoded MNIST and USPS domains ======"
)
print("-------- MNIST domain --------")
eval_tgt_robust(tgt_encoder, clf, critic, mnist_data_loader_eval)
print("-------- USPS adaption --------")
eval_tgt_robust(tgt_encoder, clf, critic, usps_data_loader_eval)
print("====== Pseudo labeling on USPS domain ======")
pseudo_label(tgt_encoder, clf, "usps_train_pseudo", usps_data_loader)
# Init a new model
tgt_encoder = model_init(Encoder(), params.tgt_encoder_path)
clf = model_init(Classifier(), params.clf_path)
# Load pseudo labeled dataset
usps_pseudo_loader = get_usps(train=True, download=True, get_pseudo=True)
print("====== Standard training on USPS domain with pseudo labels ======")
if not (tgt_encoder.pretrained and clf.pretrained):
train_src_adda(tgt_encoder, clf, usps_pseudo_loader, mode='ADV')
print("====== Evaluating on USPS domain with real labels ======")
eval_tgt(tgt_encoder, clf, usps_data_loader_eval)
tgt_encoder = model_init(Encoder(), params.tgt_encoder_rb_path)
clf = model_init(Classifier(), params.clf_rb_path)
print("====== Robust training on USPS domain with pseudo labels ======")
if not (tgt_encoder.pretrained and clf.pretrained):
train_src_robust(tgt_encoder, clf, usps_pseudo_loader, mode='ADV')
print("====== Evaluating on USPS domain with real labels ======")
eval_tgt(tgt_encoder, clf, usps_data_loader_eval)
