def validate(self, epoch):
self.elog.print('VALIDATE')
self.model.eval()
data = None
loss_list = []
acc_list = []
metrics = defaultdict(float)
with torch.no_grad():
for batch_idx, (images, masks) in enumerate(self.val_data_loader):
data, target = images.to(self.device), masks.to(self.device)
pred = self.model(data)
# pred_softmax = F.softmax(pred, dim=1)
# We calculate a softmax, because our SoftDiceLoss expects that as an input. The CE-Loss does the softmax internally.
# Ramesh check if soft max is needed
# loss = self.dice_loss(pred_softmax, target.squeeze()) + self.ce_loss(pred, target.squeeze())
# loss = F.binary_cross_entropy(pred, masks)
#loss,dice = calc_loss(pred, target, metrics)
acc = soft_dice(pred, target)
acc_list.append(acc.item())
loss = F.binary_cross_entropy(pred, target) + soft_dice(
pred, target)
loss_list.append(loss.item())
assert data is not None, 'data is None. Please check if your dataloader works properly'
self.scheduler.step(np.mean(loss_list))
self.elog.print(
'Epoch: %d Mean Loss: %.4f Mean Dice :' %
(self._epoch_idx, np.mean(loss_list)), np.mean(acc_list))
self.add_result(value=np.mean(loss_list),
name='Val_Loss',
tag='Loss',
counter=epoch + 1)
self.add_result(value=np.mean(acc_list),
name='Val_Mean_Accuracy',
tag='Accuracy',
counter=epoch + 1)
self.clog.show_image_grid(data.float().cpu(),
name="data_val",
normalize=True,
scale_each=True,
n_iter=epoch)
self.clog.show_image_grid(target.float().cpu(),
name="mask_val",
title="Mask",
n_iter=epoch)
self.clog.show_image_grid(torch.argmax(pred.data.cpu(),
dim=1,
keepdim=True),
name="unt_argmax_val",
title="Unet",
n_iter=epoch)
def train(self, epoch):
self.elog.print('=====TRAIN=====')
self.model.train()
data = None
batch_counter = 0
metrics = defaultdict(float)
#running_loss = 0.0
for batch_idx, (images, masks) in enumerate(self.train_data_loader):
data, target = images.to(self.device), masks.to(self.device)
self.optimizer.zero_grad()
#print("data shape :",data.shape, "target shape :",target.shape)
pred = self.model(data)
#pred_softmax = F.softmax(pred, dim=1)
#We calculate a softmax, because our SoftDiceLoss expects that as an input. The CE-Loss does the softmax internally.
#print("pred_softmax shape :",pred_softmax.shape, "target shape :",target.shape)
#loss = self.dice_loss(pred_softmax, target.squeeze()) + self.ce_loss(pred, target.squeeze())
loss = F.binary_cross_entropy(pred, target) + soft_dice(
pred, target)
#loss,_ = calc_loss(pred, target, metrics)
loss.backward()
self.optimizer.step()
#running_loss+=loss.item()
#epoch_loss = running_loss/len(train_data_loader)
# Some logging and plotting
if (batch_counter % self.config.plot_freq) == 0:
self.elog.print('Epoch: {0} Loss: {1:.4f}'.format(
self._epoch_idx, loss.item()))
#self.add_result(value=loss.item(), name='Train_Loss', tag='Loss', counter=epoch + (batch_counter / self.train_data_loader.num_batches))
self.add_result(value=loss.item(),
name='Train_Loss',
tag='Loss',
counter=epoch)
self.clog.show_image_grid(data.float().cpu(),
name="data",
normalize=True,
scale_each=True,
n_iter=epoch)
self.clog.show_image_grid(target.float().cpu(),
name="mask",
title="Mask",
n_iter=epoch)
self.clog.show_image_grid(torch.argmax(pred.cpu(),
dim=1,
keepdim=True),
name="unt_argmax",
title="Unet",
n_iter=epoch)
#self.clog.show_image_grid(pred.cpu()[:, 1:2, ], name="unt", normalize=True, scale_each=True, n_iter=epoch)
batch_counter += 1
assert data is not None, 'data is None. Please check if your dataloader works properly'
def test(self):
# TODO
print(' In test() method here')
self.elog.print('----------Test-------------')
self.model.eval()
trial = 10
data = None
loss_list = []
acc_list = []
metrics = defaultdict(float)
with torch.no_grad():
for batch_idx, (images, masks) in enumerate(self.test_data_loader):
data, target = images.to(self.device), masks.to(self.device)
pred = self.model(data)
pred = torch.sigmoid(pred)
#pred = F.softmax(pred, dim=1)
# We calculate a softmax, because our SoftDiceLoss expects that as an input. The CE-Loss does the softmax internally.
# Ramesh check if soft max is needed
# loss = self.dice_loss(pred_softmax, target.squeeze()) + self.ce_loss(pred, target.squeeze())
# loss = F.binary_cross_entropy(pred, masks)
#loss,dice = calc_loss(pred, target, metrics)
acc = (-1)*soft_dice(pred,target)
acc_list.append(acc.item())
#loss = F.binary_cross_entropy(pred, target) + soft_dice(pred,target)
loss = self.bce_weight*F.binary_cross_entropy(pred, target) + (1-self.bce_weight)*soft_dice(pred,target)
loss_list.append(loss.item())
assert data is not None, 'data is None. Please check if your dataloader works properly'
#self.scheduler.step(np.mean(loss_list))
self.add_result(value=loss.item(), name='Test_Loss', tag='Test_Loss', counter=trial+1)
self.add_result(value=acc.item(), name='Test_Mean_Accuracy', tag='Test_Accuracy', counter=trial+1)
self.clog.show_image_grid(data.float().cpu(), name="data_test", normalize=True, scale_each=True, n_iter=trial)
self.clog.show_image_grid(target.float().cpu(), name="mask_test", title="Mask", n_iter=trial)
self.clog.show_image_grid(torch.argmax(pred.data.cpu(), dim=1, keepdim=True), name="unt_argmax_test", title="Unet", n_iter=trial)
self.clog.show_image_grid(pred.data.cpu(), name="unt_test", normalize=True, scale_each=True, n_iter=trial)
self.elog.print('Test Mean Loss: %.4f Test Mean Dice :' % (np.mean(loss_list)),np.mean(acc_list))
# %%########################################################################### ########################### HYPERPARAMETERS ################################### ############################################################################### LEARN_RATE_0 = 1e-3 LEARN_RATE_FINETUNE = 1e-4 LEARN_RATE_ALPHA = 0.25 LEARN_RATE_STEP = 3 N_EPOCH = 30 MB_SIZE = 10 KEEP_PROB = 0.8 ACTIVATION = 'selu' PADDING = 'SYMMETRIC' AUGMENTATION = True LOSS = [[categorical_cross_entropy(), soft_dice(logits_axis=1, label_axis=0, weight=2)]] #LOSS = [[categorical_cross_entropy(), soft_dice(logits_axis=1, label_axis=0)], # [categorical_cross_entropy(onehot_convert=False), # soft_dice(logits_axis=1, label_axis=1, weight=2), # soft_dice(logits_axis=2, label_axis=2, weight=10)]] # LOSS = [None, # [categorical_cross_entropy(onehot_convert=False), # soft_dice(logits_axis=1, label_axis=1), # soft_dice(logits_axis=2, label_axis=2, weight=2)]] # %%########################################################################### ############################## LOADING DATASETS ############################### ############################################################################### # Display working/train/test directories.