nvalidArgumentError: Computed input depth 384 doesn't match filter input depth 192 [Op:Conv2DBackpropInput]

General Discussion
#1

Hi everyone,

I am currently writing a training loop from scratch using GradientTape and Tensorflow 2.6.0, and I encounter the following error when trying to compute the gradients using tape.gradient().

InvalidArgumentError: Computed input depth 384 doesn’t match filter input depth 192 [Op:Conv2DBackpropInput]

Any ideas how to solve this? I am using two customs loss functions and the model is composed of AlexNet sequential plus several other layers using FunctionalAPI.

#3

Please share standalone code to reproduce the issue reported above? Thank you