problem getting saved_model.pb about models HOT 6 CLOSED

Hi,
please refer to this closed issue:
#11
This should do it.

from models.

Thanks, but that's not the same issue. He is deploying the downloaded neuronet, whereas I have trained on my own data using the neuronet architecture. So i'm not trying to find the path to saved_model.pb, but rather to create it as it doesn't exist once I do train.py

from models.

This is where the model is exported: https://github.com/DLTK/models/blob/master/ukbb_neuronet_brain_segmentation/train.py#L181

If you break the training loop with CTRL+C, it will be triggered too. Is that what you mean? The downloaded models are produced as is from train.py

from models.

Yes, the files produced by this line of code are : checkpoint, events.out.tfevents, model.ckpt.index, model.ckpt.data, model.ckpt.meta, and graph.pbtxt. However, deploy.py requires 'saved_model.pb' which is not created inside train.py. It is however included in the downloaded models. I am trying to create it.

from models.

No, these files are created during training with the estimator train func (c.f. https://github.com/DLTK/models/blob/master/ukbb_neuronet_brain_segmentation/train.py#L168). The above line should export the model AFAIR. In order to trigger it, you will either need to:
a) Stop the iterations with CTRL+C, triggering the line here: https://github.com/DLTK/models/blob/master/ukbb_neuronet_brain_segmentation/train.py#L177
or
b) let the training run to maximum iterations.

I am not entirely certain that you have done this. Let me know if that worked or visit us in our gitter chat (https://gitter.im/DLTK/DLTK).

from models.

Thanks a lot, I wrongly ended the training indeed.

from models.