if not os.path.exists('./output_img'):
os.mkdir('./output_img')
parser, metadata = get_parser_with_args()
opt = parser.parse_args()
dev = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
test_loader = get_test_loaders(opt, batch_size=1)
path = 'weights/snunet-32.pt' # the path of the model
model = torch.load(path)
model.eval()
index_img = 0
test_metrics = initialize_metrics()
with torch.no_grad():
tbar = tqdm(test_loader)
for batch_img1, batch_img2, labels in tbar:
batch_img1 = batch_img1.float().to(dev)
batch_img2 = batch_img2.float().to(dev)
labels = labels.long().to(dev)
cd_preds = model(batch_img1, batch_img2)
cd_preds = cd_preds[-1]
_, cd_preds = torch.max(cd_preds, 1)
cd_preds = cd_preds.data.cpu().numpy()
cd_preds = cd_preds.squeeze() * 255
model = load_model(opt, dev)
criterion = get_criterion(opt)
# optimizer = optim.AdamW(model.parameters(), lr=opt.learning_rate)
optimizer = AdamW(model.parameters(), lr=opt.learning_rate)
scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=8, gamma=0.5)
"""
Set starting values
"""
best_metrics = {'cd_f1scores': -1, 'cd_recalls': -1, 'cd_precisions': -1}
logging.info('STARTING training')
best_pre, best_recall, best_f1 = [-1, -1], [-1, -1], [-1, -1]
total_step = -1
for epoch in range(opt.epochs):
train_metrics = initialize_metrics()
val_metrics = initialize_metrics()
"""
Begin Training
"""
model.train()
logging.info('SET model mode to train!')
batch_iter = 0
tbar = tqdm(train_loader)
for batch_img1, batch_img2, labels in tbar:
tbar.set_description("epoch {} info ".format(epoch) + str(batch_iter) +
" - " + str(batch_iter + opt.batch_size))
batch_iter = batch_iter + opt.batch_size
total_step += 1
# Set variables for training
batch_img1 = autograd.Variable(batch_img1).float().to(dev)
from tqdm import tqdm
parser, metadata = get_parser_with_args()
opt = parser.parse_args()
dev = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
test_loader = get_test_loaders(opt)
path = 'tmp/checkpoint_epoch_64.pt'
model = torch.load(path)
criterion = get_criterion(opt)
model.eval()
val_metrics = initialize_metrics()
with torch.no_grad():
tbar = tqdm(test_loader)
for batch_img1, batch_img2, labels in tbar:
# Set variables for training
batch_img1 = autograd.Variable(batch_img1).float().to(dev)
batch_img2 = autograd.Variable(batch_img2).float().to(dev)
labels = autograd.Variable(labels).long().to(dev)
# Get predictions and calculate loss
cd_preds = model(batch_img1, batch_img2)
cd_preds = cd_preds[-1]
_, cd_preds = torch.max(cd_preds, 1)
# Calculate and log other batch metrics