How to calculate the input_size according to the data shape?

[Overflowu7]

I see the input_size is different in the different datasets. For example, the input_size in adac is :
class UnetrPPEncoder(nn.Module):
def init(self, input_size=[16 * 40 * 40, 8 * 20 * 20, 4 * 10 * 10, 2 * 5 * 5],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=0.1 ,kwargs):
super().init()
and I check the shape of the acdc is [216,256,10].So i'm confused because the shape of data always miss match the embedding size like this:
File "/home/unetrpp/transformerblock.py", line 60, in forward
x = x + self.pos_embed
RuntimeError: The size of tensor a (27648) must match the size of tensor b (25600) at non-singleton dimension 1
Can you tell me how to solve this problem??

[Amshaker]

Hi @Overflowu7 ,

Yes, the shape of data for ACDC is 16 * 40 * 40 for the first stage.

How do you get the shape of the acdc is [216,256,10]?

[annacga]

I have the same problem @Overflowu7 if you managed to fix the problem, could you provide some insights?

[q1556450920]

Change the patch_size in plan.pkl