ValueError: logits and labels must have the same shape

[dpoiesz]

I receive the above error when attempting to train the vanilla_unet. The pipeline works for a unet I wrote that uses "same" padding but was hoping to try the vanilla_unet for it's "valid" padding. I'm not sure what else in my code could have caused this error and would have expected these conditions to have been handled inside the vanilla_unet. Any Recommendations?

initial stack trace below:

Traceback (most recent call last):
File "C:\Users\dpoiesz\Repos\bespin\venv\bespin\lib\site-packages\tensorflow_core\python\framework\tensor_shape.py", line 926, in merge_with
new_dims.append(dim.merge_with(other[i]))
File "C:\Users\dpoiesz\Repos\bespin\venv\bespin\lib\site-packages\tensorflow_core\python\framework\tensor_shape.py", line 309, in merge_with
self.assert_is_compatible_with(other)
File "C:\Users\dpoiesz\Repos\bespin\venv\bespin\lib\site-packages\tensorflow_core\python\framework\tensor_shape.py", line 276, in assert_is_compatible_with
(self, other))
ValueError: Dimensions 256 and 68 are not compatible

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "C:\Users\dpoiesz\Repos\bespin\venv\bespin\lib\site-packages\tensorflow_core\python\ops\nn_impl.py", line 167, in sigmoid_cross_entropy_with_logits
labels.get_shape().merge_with(logits.get_shape())
File "C:\Users\dpoiesz\Repos\bespin\venv\bespin\lib\site-packages\tensorflow_core\python\framework\tensor_shape.py", line 929, in merge_with
raise ValueError("Shapes %s and %s are not compatible" % (self, other))
ValueError: Shapes (None, 256, 256, 1) and (None, 68, 68, 1) are not compatible

[karolzak]

Hi @dpoiesz !
The answer to your problem is in this single line:

ValueError: Shapes (None, 256, 256, 1) and (None, 68, 68, 1) are not compatible

I assume that your input images and masks are of size 256x256?
Since vanilla_unet is using valid padding (as described in the original U-Net research paper the actual output mask size is smaller than the input size (should be 68x68 in your case).
I covered more specifically why this is happening and how to handle that in this issue: #40
Good luck!

[dpoiesz]

Thank you very much for responding, I'll try larger chips and let you know how it works out!

[dpoiesz]

Unfortunately, increasing the input size did not help. A similar error is still received. As I stated before this is very confusing since all I did was swap out my function for yours. The code works fine for the code that uses "same" padding.

Any ideas on what I should look at next?

1/1052 [..............................] - ETA: 4:14:15Traceback (most recent call last):
File "/home/jordan/anaconda3/envs/bespin/lib/python3.7/site-packages/tensorflow_core/python/framework/tensor_shape.py", line 926, in merge_with
new_dims.append(dim.merge_with(other[i]))
File "/home/jordan/anaconda3/envs/bespin/lib/python3.7/site-packages/tensorflow_core/python/framework/tensor_shape.py", line 309, in merge_with
self.assert_is_compatible_with(other)
File "/home/jordan/anaconda3/envs/bespin/lib/python3.7/site-packages/tensorflow_core/python/framework/tensor_shape.py", line 276, in assert_is_compatible_with
(self, other))
ValueError: Dimensions 512 and 324 are not compatible

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "/home/jordan/anaconda3/envs/bespin/lib/python3.7/site-packages/tensorflow_core/python/ops/nn_impl.py", line 167, in sigmoid_cross_entropy_with_logits
labels.get_shape().merge_with(logits.get_shape())
File "/home/jordan/anaconda3/envs/bespin/lib/python3.7/site-packages/tensorflow_core/python/framework/tensor_shape.py", line 929, in merge_with
raise ValueError("Shapes %s and %s are not compatible" % (self, other))
ValueError: Shapes (None, 512, 512, 1) and (None, 324, 324, 1) are not compatible

[karolzak]

512x512 is now both your input size and output mask size, right? That's incorrect for vanilla unet. valid padding is losing border pixels information on each consecutive conv layer so your input size of 512x512 outputs a 324x324 output mask. In the original research paper behind U-Net they used edge mirroring technic as a preprocessing step to turn original image from size 512x512 to 572x572 (to address border pixels loss) and it resulted with 388x388 output mask.
It all makes sense - read the paper! 😃