When training with on your own dataset,The size of tensor a (76800) must match the size of tensor b (32768) at non-singleton dimension 1

[Ohwang123]

When training with on your own dataset, the following error occurs:unetr_pp_trainer_synapse
Traceback (most recent call last):
File "C:\Project\unetr_pp\unetr_pp\run\run_training.py", line 169, in
main()
File "C:\Project\unetr_pp\unetr_pp\run\run_training.py", line 153, in main
trainer.run_training()
File "C:\Project\unetr_pp\unetr_pp\training\network_training\unetr_pp_trainer_synapse.py", line 472, in run_training
ret = super().run_training()
File "C:\Project\unetr_pp\unetr_pp\training\network_training\Trainer_synapse.py", line 320, in run_training
super(Trainer_synapse, self).run_training()
File "C:\Project\unetr_pp\unetr_pp\training\network_training\network_trainer_synapse.py", line 482, in run_training
l = self.run_iteration(self.tr_gen, True)
File "C:\Project\unetr_pp\unetr_pp\training\network_training\unetr_pp_trainer_synapse.py", line 285, in run_iteration
output = self.network(data)
File "C:\Monai\envs\unetr_pp\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "C:\Project\unetr_pp\unetr_pp\network_architecture\synapse\unetr_pp_synapse.py", line 134, in forward
x_output, hidden_states = self.unetr_pp_encoder(x_in)
File "C:\Monai\envs\unetr_pp\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "C:\Project\unetr_pp\unetr_pp\network_architecture\synapse\model_components.py", line 69, in forward
x, hidden_states = self.forward_features(x)
File "C:\Project\unetr_pp\unetr_pp\network_architecture\synapse\model_components.py", line 56, in forward_features
x = self.stages0
File "C:\Monai\envs\unetr_pp\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "C:\Monai\envs\unetr_pp\lib\site-packages\torch\nn\modules\container.py", line 141, in forward
input = module(input)
File "C:\Monai\envs\unetr_pp\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "C:\Project\unetr_pp\unetr_pp\network_architecture\synapse\transformerblock.py", line 58, in forward
x = x + self.pos_embed
RuntimeError: The size of tensor a (76800) must match the size of tensor b (32768) at non-singleton dimension 1

I am running the following nnFormer: 3d_fullres
My trainer class is: <class 'unetr_pp.training.network_training.unetr_pp_trainer_synapse.unetr_pp_trainer_synapse'>
For that I will be using the following configuration:
num_classes: 1
modalities: {0: 'CT'}
use_mask_for_norm OrderedDict([(0, False)])
keep_only_largest_region None
min_region_size_per_class None
min_size_per_class None
normalization_schemes OrderedDict([(0, 'CT')])
stages...

stage: 0
{'batch_size': 2, 'num_pool_per_axis': [4, 5, 5], 'patch_size': array([ 96, 160, 160]), 'median_patient_size_in_voxels': array([148, 258, 258]), 'current_spacing': array([1.98689442, 1.41643912, 1.41643912]), 'original_spacing': array([1. , 0.71289098, 0.71289098]), 'do_dummy_2D_data_aug': False, 'pool_op_kernel_sizes': [[2, 2, 2], [2, 2, 2], [2, 2, 2], [2, 2, 2], [1, 2, 2]], 'conv_kernel_sizes': [[3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3]]}

stage: 1
{'batch_size': 2, 'num_pool_per_axis': [4, 5, 5], 'patch_size': array([ 96, 160, 160]), 'median_patient_size_in_voxels': array([294, 512, 512]), 'current_spacing': array([1. , 0.71289098, 0.71289098]), 'original_spacing': array([1. , 0.71289098, 0.71289098]), 'do_dummy_2D_data_aug': False, 'pool_op_kernel_sizes': [[2, 2, 2], [2, 2, 2], [2, 2, 2], [2, 2, 2], [1, 2, 2]], 'conv_kernel_sizes': [[3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3]]}

pleas!! how to solve this problem?

[Amshaker]

Could you please print the shape of the data?
The patches of Synapse should be in the shape of 128 × 128 × 64. Please review our paper to see the correct input size for each dataset.

[Ohwang123]

hey，Amshaker
my own data is：
num_classes: 1
modalities: {0: 'CT'}
stage: 0
{'batch_size': 2, 'num_pool_per_axis': [4, 5, 5], 'patch_size': array([ 96, 160, 160]), 'median_patient_size_in_voxels': array([167, 240, 240]), 'current_spacing': array([2.13021975, 1.49781076, 1.49781076]), 'original_spacing': array([1. , 0.703125, 0.703125]), 'do_dummy_2D_data_aug': False, 'pool_op_kernel_sizes': [[2, 2, 2], [2, 2, 2], [2, 2, 2], [2, 2, 2], [1, 2, 2]], 'conv_kernel_sizes': [[3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3]]}

stage: 1
{'batch_size': 2, 'num_pool_per_axis': [4, 5, 5], 'patch_size': array([ 96, 160, 160]), 'median_patient_size_in_voxels': array([356, 512, 512]), 'current_spacing': array([1. , 0.703125, 0.703125]), 'original_spacing': array([1. , 0.703125, 0.703125]), 'do_dummy_2D_data_aug': False, 'pool_op_kernel_sizes': [[2, 2, 2], [2, 2, 2], [2, 2, 2], [2, 2, 2], [1, 2, 2]], 'conv_kernel_sizes': [[3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3]]}

my enviroment is Windows，python==3.8，use 3d_fullres unetr_pp_trainer_synapse 1 4

This is shape of data：

def forward(self, x_in):
        print(f"x_in.shape------>{x_in.shape}")
        x_output, hidden_states = self.unetr_pp_encoder(x_in)
        print(f"x_out.shape----->{x_output.shape}")
        convBlock = self.encoder1(x_in)

        # Four encoders
        enc1 = hidden_states[0]
        enc2 = hidden_states[1]
        enc3 = hidden_states[2]
        enc4 = hidden_states[3]

x_in.shape------>torch.Size([1, 1, 64, 128, 128])
UnetrPPEncodertorch.Size([1, 1, 64, 128, 128])
x_out.shape----->torch.Size([1, 64, 256])
epoch: 0
x_in.shape------>torch.Size([2, 1, 96, 160, 160])
UnetrPPEncodertorch.Size([2, 1, 96, 160, 160])

Then make a mistake：

Traceback (most recent call last):
File "C:\Project\unetr_plus_plus-main\unetr_pp\run\run_training.py", line 169, in
main()
File "C:\Project\unetr_plus_plus-main\unetr_pp\run\run_training.py", line 153, in main
trainer.run_training()
File "C:\Project\unetr_plus_plus-main\unetr_pp\training\network_training\unetr_pp_trainer_synapse.py", line 474, in run_training
ret = super().run_training()
File "C:\Project\unetr_plus_plus-main\unetr_pp\training\network_training\Trainer_synapse.py", line 320, in run_training
super(Trainer_synapse, self).run_training()
File "C:\Project\unetr_plus_plus-main\unetr_pp\training\network_training\network_trainer_synapse.py", line 481, in run_training
l = self.run_iteration(self.tr_gen, True)
File "C:\Project\unetr_plus_plus-main\unetr_pp\training\network_training\unetr_pp_trainer_synapse.py", line 284, in run_iteration
output = self.network(data)
File "C:\Monai\envs\unetr_pp\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "C:\Project\unetr_plus_plus-main\unetr_pp\network_architecture\synapse\unetr_pp_synapse.py", line 135, in forward
x_output, hidden_states = self.unetr_pp_encoder(x_in)
File "C:\Monai\envs\unetr_pp\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "C:\Project\unetr_plus_plus-main\unetr_pp\network_architecture\synapse\model_components.py", line 70, in forward
x, hidden_states = self.forward_features(x)
File "C:\Project\unetr_plus_plus-main\unetr_pp\network_architecture\synapse\model_components.py", line 56, in forward_features
x = self.stages0
File "C:\Monai\envs\unetr_pp\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "C:\Monai\envs\unetr_pp\lib\site-packages\torch\nn\modules\container.py", line 141, in forward
input = module(input)
File "C:\Monai\envs\unetr_pp\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl
return forward_call(*input, **kwargs)
File "C:\Project\unetr_plus_plus-main\unetr_pp\network_architecture\synapse\transformerblock.py", line 57, in forward
x = x + self.pos_embed
RuntimeError: The size of tensor a (76800) must match the size of tensor b (32768) at non-singleton dimension 1

Process finished with exit code 1

The error also occurred before：

C:\Monai\envs\unetr_pp\lib\site-packages\torch\nn\functional.py:2498: UserWarning: floordiv is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').
verify_batch_size([input.size(0) * input.size(1) // num_groups, num_groups] + list(input.size()[2:]))
C:\Project\unetr_plus_plus-main\unetr_pp\network_architecture\synapse\transformerblock.py:95: UserWarning: floordiv is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').
qkvv = self.qkvv(x).reshape(B, N, 4, self.num_heads, C // self.num_heads)
C:\Monai\envs\unetr_pp\lib\site-packages\einops\einops.py:204: UserWarning: floordiv is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').
inferred_length: int = length // known_product
Unsupported operator aten::mul encountered 108 time(s)
Unsupported operator aten::add encountered 67 time(s)
Unsupported operator aten::div encountered 68 time(s)
Unsupported operator aten::norm encountered 42 time(s)
Unsupported operator aten::clamp_min encountered 42 time(s)
Unsupported operator aten::expand_as encountered 42 time(s)
Unsupported operator aten::softmax encountered 42 time(s)
Unsupported operator aten::add encountered 65 time(s)
Unsupported operator aten::leaky_relu_ encountered 46 time(s)
Unsupported operator aten::feature_dropout encountered 21 time(s)
Unsupported operator aten::mul_ encountered 2 time(s)

[LimxRabbit]

hey，Amshaker my own data is： num_classes: 1 modalities: {0: 'CT'} stage: 0 {'batch_size': 2, 'num_pool_per_axis': [4, 5, 5], 'patch_size': array([ 96, 160, 160]), 'median_patient_size_in_voxels': array([167, 240, 240]), 'current_spacing': array([2.13021975, 1.49781076, 1.49781076]), 'original_spacing': array([1. , 0.703125, 0.703125]), 'do_dummy_2D_data_aug': False, 'pool_op_kernel_sizes': [[2, 2, 2], [2, 2, 2], [2, 2, 2], [2, 2, 2], [1, 2, 2]], 'conv_kernel_sizes': [[3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3]]}

stage: 1 {'batch_size': 2, 'num_pool_per_axis': [4, 5, 5], 'patch_size': array([ 96, 160, 160]), 'median_patient_size_in_voxels': array([356, 512, 512]), 'current_spacing': array([1. , 0.703125, 0.703125]), 'original_spacing': array([1. , 0.703125, 0.703125]), 'do_dummy_2D_data_aug': False, 'pool_op_kernel_sizes': [[2, 2, 2], [2, 2, 2], [2, 2, 2], [2, 2, 2], [1, 2, 2]], 'conv_kernel_sizes': [[3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3], [3, 3, 3]]}

my enviroment is Windows，python==3.8，use 3d_fullres unetr_pp_trainer_synapse 1 4

This is shape of data：

def forward(self, x_in):
        print(f"x_in.shape------>{x_in.shape}")
        x_output, hidden_states = self.unetr_pp_encoder(x_in)
        print(f"x_out.shape----->{x_output.shape}")
        convBlock = self.encoder1(x_in)

        # Four encoders
        enc1 = hidden_states[0]
        enc2 = hidden_states[1]
        enc3 = hidden_states[2]
        enc4 = hidden_states[3]

x_in.shape------>torch.Size([1, 1, 64, 128, 128]) UnetrPPEncodertorch.Size([1, 1, 64, 128, 128]) x_out.shape----->torch.Size([1, 64, 256]) epoch: 0 x_in.shape------>torch.Size([2, 1, 96, 160, 160]) UnetrPPEncodertorch.Size([2, 1, 96, 160, 160])

Then make a mistake：

Traceback (most recent call last): File "C:\Project\unetr_plus_plus-main\unetr_pp\run\run_training.py", line 169, in main() File "C:\Project\unetr_plus_plus-main\unetr_pp\run\run_training.py", line 153, in main trainer.run_training() File "C:\Project\unetr_plus_plus-main\unetr_pp\training\network_training\unetr_pp_trainer_synapse.py", line 474, in run_training ret = super().run_training() File "C:\Project\unetr_plus_plus-main\unetr_pp\training\network_training\Trainer_synapse.py", line 320, in run_training super(Trainer_synapse, self).run_training() File "C:\Project\unetr_plus_plus-main\unetr_pp\training\network_training\network_trainer_synapse.py", line 481, in run_training l = self.run_iteration(self.tr_gen, True) File "C:\Project\unetr_plus_plus-main\unetr_pp\training\network_training\unetr_pp_trainer_synapse.py", line 284, in run_iteration output = self.network(data) File "C:\Monai\envs\unetr_pp\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl return forward_call(*input, **kwargs) File "C:\Project\unetr_plus_plus-main\unetr_pp\network_architecture\synapse\unetr_pp_synapse.py", line 135, in forward x_output, hidden_states = self.unetr_pp_encoder(x_in) File "C:\Monai\envs\unetr_pp\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl return forward_call(*input, **kwargs) File "C:\Project\unetr_plus_plus-main\unetr_pp\network_architecture\synapse\model_components.py", line 70, in forward x, hidden_states = self.forward_features(x) File "C:\Project\unetr_plus_plus-main\unetr_pp\network_architecture\synapse\model_components.py", line 56, in forward_features x = self.stages0 File "C:\Monai\envs\unetr_pp\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl return forward_call(*input, **kwargs) File "C:\Monai\envs\unetr_pp\lib\site-packages\torch\nn\modules\container.py", line 141, in forward input = module(input) File "C:\Monai\envs\unetr_pp\lib\site-packages\torch\nn\modules\module.py", line 1110, in _call_impl return forward_call(*input, **kwargs) File "C:\Project\unetr_plus_plus-main\unetr_pp\network_architecture\synapse\transformerblock.py", line 57, in forward x = x + self.pos_embed RuntimeError: The size of tensor a (76800) must match the size of tensor b (32768) at non-singleton dimension 1

Process finished with exit code 1

The error also occurred before：

C:\Monai\envs\unetr_pp\lib\site-packages\torch\nn\functional.py:2498: UserWarning: floordiv is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor'). verify_batch_size([input.size(0) * input.size(1) // num_groups, num_groups] + list(input.size()[2:])) C:\Project\unetr_plus_plus-main\unetr_pp\network_architecture\synapse\transformerblock.py:95: UserWarning: floordiv is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor'). qkvv = self.qkvv(x).reshape(B, N, 4, self.num_heads, C // self.num_heads) C:\Monai\envs\unetr_pp\lib\site-packages\einops\einops.py:204: UserWarning: floordiv is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor'). inferred_length: int = length // known_product Unsupported operator aten::mul encountered 108 time(s) Unsupported operator aten::add encountered 67 time(s) Unsupported operator aten::div encountered 68 time(s) Unsupported operator aten::norm encountered 42 time(s) Unsupported operator aten::clamp_min encountered 42 time(s) Unsupported operator aten::expand_as encountered 42 time(s) Unsupported operator aten::softmax encountered 42 time(s) Unsupported operator aten::add encountered 65 time(s) Unsupported operator aten::leaky_relu_ encountered 46 time(s) Unsupported operator aten::feature_dropout encountered 21 time(s) Unsupported operator aten::mul_ encountered 2 time(s)

i've met the same problem, do you solve it?

[qyuiiii123]

you have the wrong patch_size. It should be (64, 128, 128) for synapse, but yours data_shape is ([ 96, 160, 160]). you need to open the plan_file and change it.76800 = (96/2)(160/4)(160/4) and 32768 = (64/2)(128/2)(128/2), which you can in the file /.../unetr_plus_plus-main/unetr_pp/network_architecture/synapse/unetr_pp_synapse.py
class UnetrPPEncoder(nn.Module): def __init__(self, input_size=[32 * 32 * 32, 16 * 16 * 16, 8 * 8 * 8, 4 * 4 * 4],dims=[32, 64, 128, 256], proj_size =[64,64,64,32], depths=[3, 3, 3, 3], num_heads=4, spatial_dims=3, in_channels=1, dropout=0.0, transformer_dropout_rate=0.15 ,**kwargs): super().__init__()