How to freeze weights of deeplabv3+ for fine-tuning?

[aj96]

I know that normally you are suppose to filter out variables you don't want to train to a particular scope as per this stackoverflow: https://stackoverflow.com/questions/35298326/freeze-some-variables-scopes-in-tensorflow-stop-gradient-vs-passing-variables But I do not see where to do this in the deeplab folder as they are using tf.slim to do all of the training.

[renan-r-santos]

Hi @aj96
I want to freeze some layers too for fine tuning. Looking at slim docs, I found that:

Typically for fine-tuning one only want train a sub-set of layers, so the flag --trainable_scopes allows to specify which subsets of layers should trained, the rest would remain frozen.

I will try this and let you know. If you find any solution, please let me know too. Thanks

[ewen1024]

@scherzocrk
Hi, did you manage to find the way of freezing some layers?

Thx

[Damanpreet]

@scherzocrk
Were you able to freeze the layers and train your model?