-1

I have been given a trained U-Net (on tensorflow), but its performance is not great. For my research I would like to run this convolutional neural network in "reverse". I would like to generate its Input by using the Weights (checkpoints) and its Output. I found a topic that had the same goal: Using a learned Artificial Neural Network to solve inputs, but it hasn't been answered and has been inactive for a while, it also does not dealing with U-Net (not sure if this adds complications). Essentially the goal of this is to gain a better understanding of the U-Net's performance and how it learns. I am not able to retrain the networks as that would take up too much time and is beyond the scope of my research, so all I have is the architecture and the weights.

desertnaut
  • 57,590
  • 26
  • 140
  • 166
Aug.Rad
  • 1

1 Answers1

0

You can use model.get_layer() method to retrieve a layer based on either its name (unique) or index. I think using the index will be useful in your case. You can visit https://keras.io/api/models/model/#getlayer-method.