Please Help: Transfer Learning using pretrained segmentation.fcn_resnet101

1

Please i have been trying to train this model (segmentaion_fcn_resnet101) but i kept on getting this error:

Capturep
Capturep740×327 40.5 KB

And my train_loader and val_loader
if name == ‘main’:

train_dataset = ListDataset('/content/gdrive/My Drive/dataset/Train', 'Train')

train_loader = DataLoader(train_dataset, batch_size=5, shuffle=True, sampler=None,

           batch_sampler=None, num_workers=0, collate_fn=None,

           pin_memory=False, drop_last=False, timeout=0,

           worker_init_fn=None)

val_dataset = ListDataset('/content/gdrive/My Drive/dataset/Validate', 'Validate')

val_loader = DataLoader(dataset=val_dataset, batch_size=5, shuffle=True, sampler=None,

           batch_sampler=None, num_workers=0, collate_fn=None,

           pin_memory=False, drop_last=False, timeout=0,

           worker_init_fn=None)

model = models.segmentation.fcn_resnet101(pretrained= True, aux_loss=None)

for param in model.parameters():

param.requires_grad=False
model.classifier[4] = nn.Conv2d(512, 2, kernel_size=(1, 1), stride=(1, 1))

model.aux_classifier = None

def train_model(model, criterion, optimizer, scheduler, num_epochs=5):

since = time.time()

val_acc_history = []

best_model_wts = copy.deepcopy(model.state_dict())

best_acc = 0.0

for epoch in range(num_epochs):

    print('Epoch {}/{}'.format(epoch, num_epochs - 1))

    print('-' * 10)

    model.train()

    # Each epoch has a training and validation phase

    

    running_loss = 0.0

    running_corrects = 0

        # Iterate over data.

    for inputs, labels in train_loader:

        

        inputs = inputs.to(device)

        labels = labels.to(device)

            # zero the parameter gradients

        optimizer.zero_grad()

        

            # forward

            # track history if only in train

        with torch.set_grad_enabled(True):

          outputs = model(inputs)

          loss = criterion(outputs, labels)

          _, preds = torch.max(outputs, 1)

                    

        # backward + optimize 

          

        loss.backward()

        optimizer.step()

            # statistics

        running_loss += loss.item() * inputs.size(0)

        running_corrects += torch.sum(preds == labels.data)

        

        scheduler.step()

    epoch_loss = running_loss / len(train_loader.dataset)

    epoch_acc = running_corrects.double() / len(train_loader.dataset)

    print('{} Loss: {:.4f} Acc: {:.4f}'.format(epoch_loss, epoch_acc))

    ## A Loop for Validation data

    model.eval()

    for val_inputs, val_labels in val_loader:

                  

        val_inputs = val_inputs.to(device)

        val_labels = val_labels.to(device)

            # zero the parameter gradients

        optimizer.zero_grad()

        with torch.no_grad():

          val_outputs = model(val_inputs)

          loss = criterion(val_outputs, val_labels)  

          _, preds = torch.max(val_outputs, 1)

        

        # deep copy the model

        if epoch_acc > best_acc:

          best_acc = epoch_acc

          best_model_wts = copy.deepcopy(model.state_dict())

    print()

time_elapsed = time.time() - since

print('Training complete in {:.0f}m {:.0f}s'.format(

    time_elapsed // 60, time_elapsed % 60))

print('Best val Acc: {:4f}'.format(best_acc))

# load best model weights

model.load_state_dict(best_model_wts)

return model

##Training
for param in model.parameters():

param.requires_grad=False

device = torch.device('cuda’if torch.cuda.is_available() else ‘cpu’)

model = model.to(device)

criterion = nn.CrossEntropyLoss()

optimizer = optim.SGD(model.parameters(), lr = 0.001, momentum=0.9)

exp_lr_scheduler = lr_scheduler.StepLR(optimizer, step_size=7, gamma=0.1)

num_epochs = 5

model = train_model(model, criterion, optimizer, exp_lr_scheduler, num_epochs=num_epochs)

3

fcn_resnet101 will return an OrderedDict instead of a single tensor, which explains your error.
You should be able to get the output tensor via:

output = model(input)
out = output['out']

and could pass it to the criterion.

4

Thanks for your reply, i imported the model as
model = models.segmentation.fcn_resnet101(pretrained= True, aux_loss=False)
for param in model.parameters():
param.requires_grad=False

model.classifier[4] = nn.Conv2d(512, 2, kernel_size=(1, 1), stride=(1, 1))
model.aux_classifier = None

… i have changed the line u suggested to
with torch.set_grad_enabled(True):
outputs = model(inputs)
print(type(outputs))
labels = labels.long()
loss = criterion(outputs[‘out’], labels)
_, preds = torch.max(outputs, 1)

but getting this error :
TypeError: max() received an invalid combination of arguments - got (collections.OrderedDict, int), but expected one of:

  • (Tensor input)
  • (Tensor input, name dim, bool keepdim, *, tuple of Tensors out)
  • (Tensor input, Tensor other, *, Tensor out)
  • (Tensor input, int dim, bool keepdim, *, tuple of Tensors out)
5

The torch.max operation raises now the error, since you are again passing the OrderedDict to it instead of the tensor. :wink:

6

I solved that with this but could not call:
loss.backward()

during training and ended up with this
RuntimeError: cuDNN error: CUDNN_STATUS_NOT_INITIALIZED (createCuDNNHandle at /pytorch/aten/src/ATen/cudnn/Handle.cpp:9)

7

This might be raised, if you are running out of memory.
Check the memory usage via nvidia-smi and try to lower the batch size.
If that’s not the case, disable cudnn via torch.backends.cudnn.enabled = False and rerun the code.
If the code is working fine without cudnn and is not running OOM with cudnn, could you post a code snippet to reproduce this issue as well as your current setup (PyTorch, CUDA, cudnn version and the used hardware)?

8

Alright i will do just that but i am running the training on Google Colab

9

I think i have seen where the error is coming from, i changed the runtime type to cpu …
This is the size of my dataset from the getitem (5, 3, 256, 256) and i added this line of code (#Edit label tensor to remove channel since nn.CrossEntropy Loss() receives (B, H, W)
mask_array = torch.cat((mask_array[:1], mask_array[256:]))
mask_array = torch.squeeze(mask_array, 0))
to squeeze only the mask(label) to (5, 256,256)

for my training i have this:
outputs = model(inputs)
print(type(outputs))
labels = labels.long()
loss = criterion(outputs[‘out’], labels) ## the error is coming from here
_, preds = torch.max(outputs[‘out’], 1)
# backward + optimize
loss.backward()
optimizer.step()

still ended up with this error : IndexError: Target -2 is out of bounds

Thanks for your help