Clallbacks: Cannot feed value of shape (16, 128, 128, 1) for Tensor 'conv2d/truediv:0', which has shape '(?, 128, 128, 3)' about multiplanarunet HOT 7 CLOSED

And the shape of X_ins is (?, 128, 128, 1), the shape of y_ins is (?, 128, 128, 3), and the shape of tensor is (?, 128, 128, 3).

from multiplanarunet.

The model is

from multiplanarunet.

Hi!

Your y_ins should be of shape (?, 128, 128, 1) with each entry giving an integer-valued label. That is, values in {0, 1, 2} in your case for a 3-class problem. I would expect that you have your label maps stored in one-hot encoding (which would be normal, but not what is used here).

I am, however, surprised that you did not get a more appropriate error message - so perhaps this is not the error. I will look into it later this week as soon as I have more time.

If this does solve your issue let me know.

from multiplanarunet.

Hi! Thank you for your reply! One thing I did not mention. I changed slightly your code in the callbacks.py
(y_ins = {in_tens: in_ for in_tens, in_ in zip(self.model.targets, y)} --> y_ins = {in_tens: in_ for in_tens, in_ in zip(self.model.output, y)}),
due to the error "Unet object has no attribute 'targets' "

from multiplanarunet.

Hi! I think I understand it. The problem is the code of callbacks.py:
'y_ins = {in_tens: in_ for in_tens, in_ in zip(self.model.targets, y)}', because I changed 'targets'(,,,1) to 'outputs'(,,,2) for binary segmentation.
However, I still do not know the placeholder variable name of y_ins due to the error "Unet object has no attribute 'targets' ".

Best Wishes,
Tao H

from multiplanarunet.

Hi! Maybe I solved the problem. I replaced 'targets' with '_targets' and it works. However, I do not known if it is true.

Best Wishes,
Tao H

from multiplanarunet.

Hi,

Please update to the latest version (0.2.4) with TensorFlow >=2.0. If you still receive this error, please raise a new issue.

Cheers,
Mathias

from multiplanarunet.