def test_model(model,optimizer,trial):
since = time.time()
model.eval()
# Iterate over data.
for index, (inputs, labels) in enumerate(dataloaders_test):
inputs = inputs.to(device)
labels = labels.to(device)
inputs = inputs.float()
labels = labels.float()
# zero the parameter gradients
optimizer.zero_grad()
# forward
# track history if only in train
with torch.set_grad_enabled(False):
outputs = model(inputs)
# backward + optimize only if in training phase
if index == 0:
outputs_test = outputs
labels_test = labels
else:
outputs_test = torch.cat((outputs_test, outputs), 0)
labels_test = torch.cat((labels_test, labels), 0)
cmap, emap = cemap_cal(outputs_test.to(torch.device("cpu")).numpy(), labels_test.to(torch.device("cpu")).numpy())
print('Test:')
print(cmap,emap)
p, r, f = prf_cal(outputs_test.to(torch.device("cpu")).numpy(), labels_test.to(torch.device("cpu")).numpy(), 3)
time_elapsed = time.time() - since
print('Training complete in {:.0f}m {:.0f}s'.format(time_elapsed // 60, time_elapsed % 60))
scipy.io.savemat('./results/resnet101_model1fc_results_'+str(trial)+'imgSize'+str(img_size)+'.mat', mdict={'cmap': cmap, 'emap': emap, 'p': p,'r': r, 'f': f, 'scores': outputs_test.to(torch.device("cpu")).numpy()})
def train_model(model, criterion, optimizer, scheduler, num_epochs=25):
since = time.time()
best_model_wts = copy.deepcopy(model.state_dict())
best_emap = 0
for epoch in range(num_epochs):
scheduler.step()
print('Epoch {}/{}'.format(epoch, num_epochs - 1))
print('-' * 10)
for param_group in optimizer.param_groups:
cur_lr = param_group['lr']
print('Current learning rate: ' + '%.5f' % cur_lr)
if cur_lr < 0.00001:
break
# Each epoch has a training and validation phase
for phase in ['train']:
if phase == 'train':
model.train() # Set model to training mode
else:
model.eval() # Set model to evaluate mode
running_loss = 0.0
running_corrects = 0
# Iterate over data.
for index, (inputs, labels) in enumerate(dataloaders[phase]):
inputs = inputs.to(device)
labels = labels.to(device)
# mixed up
inputs = inputs.float()
labels = labels.float()
labels_tr = labels.clone()
labels_tr[labels == 0] = 1
labels_tr[labels == -1] = 0
#if epoch % 2 == 0:
# inputs,labels_tr = mixup_data3(inputs,labels_tr,use_cuda=True)
# zero the parameter gradients
optimizer.zero_grad()
# forward
# track history if only in train
with torch.set_grad_enabled(phase == 'train'):
outputs = model(inputs)
loss = criterion(outputs, labels_tr)
# loss = newHinge_rank_loss(outputs,labels)
# backward + optimize only if in training phase
if phase == 'train':
loss.backward()
optimizer.step()
if phase == 'val':
if index == 0:
outputs_val = outputs
labels_val = labels
else:
outputs_val = torch.cat((outputs_val, outputs), 0)
labels_val = torch.cat((labels_val, labels), 0)
running_loss += loss.item() * inputs.size(0)
if phase == 'train':
epoch_loss = running_loss / dataset_sizes['trainval']
else:
epoch_loss = running_loss / dataset_sizes['val']
if phase == 'val':
cmap, emap = cemap_cal(
outputs_val.to(torch.device("cpu")).numpy(),
labels_val.to(torch.device("cpu")).numpy())
p, r, f = prf_cal(
outputs_val.to(torch.device("cpu")).numpy(),
labels_val.to(torch.device("cpu")).numpy(), 3)
val_loss = epoch_loss
print('{} Loss: {:.4f}'.format(phase, epoch_loss))
# deep copy the model
if phase == 'val' and cmap > best_emap:
best_emap = cmap
print(cmap, emap)
print(p, r, f)
test_model(model, optimizer, trial)
#best_model_wts = copy.deepcopy(model.state_dict())
#torch.save(model_ft.state_dict(),'./resnet101_model1_trial0.pt')
print()
time_elapsed = time.time() - since
print('Training complete in {:.0f}m {:.0f}s'.format(
time_elapsed // 60, time_elapsed % 60))
# load best model weights
model.load_state_dict(best_model_wts)
return model