def filter():
transform = transforms.Compose([
transforms.Resize((224, 224)),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
])
dataset = ImageFolder(root=args.icdar_patches, transform=transform)
loader = torch.utils.data.DataLoader(dataset, batch_size=10, shuffle=True)
model = Extractor()
for params in model.parameters():
params.requires_grad = False
acc = clustering(loader=loader, model=model)
# Part 2: train F1t, F2t with pseudo labels
logger.info("#################### Part2 ####################")
extractor.train()
s1_classifier.train()
s2_classifier.train()
s3_classifier.train()
t_pse_label = os.path.join(data_root, args.t,
"pseudo/pse_label_" + str(step) + ".txt")
t_pse_set = OfficeImage(t_root, t_pse_label, split="train")
t_pse_loader_raw = torch.utils.data.DataLoader(t_pse_set,
batch_size=batch_size,
shuffle=shuffle,
num_workers=num_workers)
logger.info("Length of pseudo-label dataset:{}".format(len(t_pse_set)))
optim_extract = optim.Adam(extractor.parameters(),
lr=lr,
betas=(beta1, beta2))
optim_s1_cls = optim.Adam(s1_classifier.parameters(),
lr=lr,
betas=(beta1, beta2))
optim_s2_cls = optim.Adam(s2_classifier.parameters(),
lr=lr,
betas=(beta1, beta2))
optim_s3_cls = optim.Adam(s3_classifier.parameters(),
lr=lr,
betas=(beta1, beta2))
for cls_epoch in range(cls_epoches): #cls_epoches
s1_loader, s2_loader, s3_loader, t_pse_loader = iter(
s1_loader_raw), iter(s2_loader_raw), iter(s3_loader_raw), iter(
#Parameters
Batch_size_s = int(length_source_training / 25) + 1
Batch_size_t = int(length_target / 25) + 1
num_epochs = 200
m = 1
n = 10**(-7)
E = Extractor()
D = Discriminator()
R = Regressor()
E.apply(weights_init)
D.apply(weights_init)
R.apply(weights_init)
e_learning_rate = 0.00003
d_learning_rate = 0.00015
r_learning_rate = 0.0000001
e_optimizer = optim.RMSprop(E.parameters(), lr=e_learning_rate, alpha=0.9)
d_optimizer = optim.RMSprop(D.parameters(), lr=d_learning_rate, alpha=0.9)
r_optimizer = optim.RMSprop(R.parameters(), lr=r_learning_rate, alpha=0.9)
e_steps = 1
d_steps = 1
r_steps = 1
#SAMPLING
source_training_dataset = Data.TensorDataset(source_training_features,
source_training_liquid_labels)
source_validation_dataset = Data.TensorDataset(
source_validation_features, source_validation_liquid_labels)
target_dataset = Data.TensorDataset(target_features, target_liquid_labels)
source_training_loader_d = Data.DataLoader(dataset=source_training_dataset,
batch_size=Batch_size_s,
pbar.update(1)
im, ann, im_fns = data[0], data[1], data[2]
im = im.cuda()
feat = model(im)
x = feat.cpu().numpy()
y = ann.numpy()
precision = precision + clf.score(x, y)
pbar.close()
print("\nAverage Precision is {}".format(precision / len(loader)))
if __name__ == '__main__':
icdar_patches = "./data/icdar2015/patches"
dataset = ImageFolder(root=icdar_patches, transform=transforms)
loader = torch.utils.data.DataLoader(dataset, batch_size=20, shuffle=True)
# extractor for deep features
model = Extractor()
model = model.cuda()
for params in model.parameters():
params.requires_grad = False
# vanilla svm
clf = svm.SVC(kernel="rbf", gamma=10)
train(loader=loader, model=model, clf=clf)
if t_pred[j, ids[j]] >= threshold:
fout.write(data[0] + " " + str(ids[j]) + "\n")
fin.close()
fout.close()
# Part 2: train F1t, F2t with pseudo labels
print ("#################### Part2 ####################")
extractor.train()
s1_classifier.train()
s2_classifier.train()
t_pse_label = os.path.join(data_root, args.t, "pseudo/pse_label_" + str(step) + ".txt")
t_pse_set = OfficeImage(t_root, t_pse_label, split="train")
t_pse_loader_raw = torch.utils.data.DataLoader(t_pse_set, batch_size=batch_size,
shuffle=shuffle, num_workers=num_workers)
print ("Length of pseudo-label dataset: ", len(t_pse_set))
optim_extract = optim.Adam(extractor.parameters(), lr=lr, betas=(beta1, beta2))
optim_s1_cls = optim.Adam(s1_classifier.parameters(), lr=lr, betas=(beta1, beta2))
optim_s2_cls = optim.Adam(s2_classifier.parameters(), lr=lr, betas=(beta1, beta2))
for cls_epoch in range(cls_epoches):
s1_loader, s2_loader, t_pse_loader = iter(s1_loader_raw), iter(s2_loader_raw), iter(t_pse_loader_raw)
for i, (t_pse_imgs, t_pse_labels) in tqdm.tqdm(enumerate(t_pse_loader)):
try:
s1_imgs, s1_labels = s1_loader.next()
except StopIteration:
s1_loader = iter(s1_loader_raw)
s1_imgs, s1_labels = s1_loader.next()
try:
s2_imgs, s2_labels = s2_loader.next()
except StopIteration:
s2_loader = iter(s2_loader_raw)