Issue type

Bug

Have you reproduced the bug with TensorFlow Nightly?

Yes

Source

source

TensorFlow version

tf-2.15.0, tf-2.16.0, tf-2.16.1

Custom code

Yes

OS platform and distribution

Windows 10

Mobile device

No response

Python version

3.9

Bazel version

No response

GCC/compiler version

No response

CUDA/cuDNN version

No response

GPU model and memory

No response

Current behavior?

I found an output shape inconsistent in Conv3D layer:

from tensorflow.keras.layers import Conv3D
import numpy as np

x=np.random.rand(3,5,5,5,4)
l=Conv3D(5,[2,2,3],[1,1,1],'valid','channels_first',[2,2,2],1)
print(l(x).shape)
print(l.compute_output_shape(x.shape))

The output is:

(3, 5, 5, 5, 4)
(3, 5, 3, 3, 0)

However, I failed to reproduce it on google colab. After checking the package and re-running it on a Linux machine and another Windows machine, I found this bug may come from the package 'tensorflow-intel', which exists only on Windows platform.

Could you please check it?

Standalone code to reproduce the issue

from tensorflow.keras.layers import Conv3D
import numpy as np

x=np.random.rand(3,5,5,5,4)
l=Conv3D(5,[2,2,3],[1,1,1],'valid','channels_first',[2,2,2],1)
print(l(x).shape)
print(l.compute_output_shape(x.shape))

### Relevant log output

```shell
(3, 5, 5, 5, 4)
(3, 5, 3, 3, 0)