-
Notifications
You must be signed in to change notification settings - Fork 0
Description
Hello,
I'm trying to recreate a training pipeline originally written in tensorflow.
It is really simple and consists in a succession of transfer learnings and fine tunings.
It uses MobileNet, pre-trained on ImageNet, then performs transfer learning on another dataset, then fine tuning (training on more layers than the classification layers) on this same dataset.
I then need to perform the same 2 operations on another dataset, but i can't get past the first fine tuning.
The first transfer learning (using the retrain recipe on the last layer of the MobileNet downloaded with the provided macro) works, but i can't use another retrain recipe on the output model.
I don't get where the problem comes from as this is a really common practice and i only use the tools provided by the plugin.
Would you know what causes this problem and how to fix it ?
output model folder's content :

weights_no_top.h5 has 157 layers and model_info.json ('before_train') has 155 layers

