Stucked in zombie detector : fine tuning does not work

I’ve been checking the code and I can’t find the mistake. In the fine tuning process I always get results like this:
Start fine-tuning!
batch 0 of 100, loss=1.3546208
batch 10 of 100, loss=1.3543563
batch 20 of 100, loss=1.3540937
batch 30 of 100, loss=1.3538327
batch 40 of 100, loss=1.3535731
batch 50 of 100, loss=1.3533152
batch 60 of 100, loss=1.3530587
batch 70 of 100, loss=1.3528037
batch 80 of 100, loss=1.35255
batch 90 of 100, loss=1.3522975
Done fine-tuning!

I’ve cheked the chekpoint exercises (6.2 and 6.3) and everyhing seems to be ok with the notebook instructions. After a lot of hours I really can’t find the problem without more instructions in this notebook ( which is really meesy !!)


In excercise 6.3, check if you are restoring the weights using:
checkpoint.restore(checkpoint_path) and not with the tmp_model_checkpoint.restore(checkpoint_path)

Thanks so much Hrishabh Tiwari.

Altought I din’t use: tmp_model_checkpoint.restore(checkpoint_path)

in this exercise I forgot to use tf.train.Checkpoint() to create a new check point!!!..

Thanks for your help!!