About helpers/nn_mri.py file and import lrelu

[zaccharieramzi]

About helpers/nn_mri.py file and import lrelu

[zaccharieramzi]

Hi Fatma,

Yes this is to be expected in those notebooks as they are pretty old now and not maintained. They should still provide a good basis for testing the networks trained with the functional API, so you could try to just prune them to what you need.

When I have time I will try to correct them in order to have them work in the current state, or even better provide evaluation scripts for the functional networks.

[fharman]

etty old now and not maintained. They should still provide a good basis for testing the networks trained with the functional API, so you could try to just prune them to what you need.

thank you, i am waiting for your information

[fharman]

Hello again,

when i prune the code for reconstruction (comment all lines that included helpers folder), i encountered another problem that included helpers folder.

(#from fastmri_recon.helpers.nn_mri import lrelu
#from fastmri_recon.helpers.reconstruction import reco_and_gt_unet_from_val_file, reco_and_gt_net_from_val_file, reco_and_gt_zfilled_from_val_file)

NameError:name 'reco_and_gt_unet_from_val_file' is not defined

from fastmri_recon.helpers.reconstruction import reco_and_gt_unet_from_val_file, reco_and_gt_net_from_val_file, reco_and_gt_zfilled_from_val_file

As it is seen, in reconstruction.py , it is needed to import reco_and_gt_unet_from_val_file. In all_net_params section,

'val_gen': val_gen_zero,
'run_id': 'unet_af4_1569210349',
'reco_function': reco_and_gt_unet_from_val_file,
},

these lines calls the reco_and_gt_unet_from_val_file for the reconstruction. So helpers folder contains all these reconstruction.py and nn_mri.py files.

best regards,

[zaccharieramzi]

Sorry about that.

• You can find reco_and_gt_unet_from_val_file in fastmri_recon.evaluate.reconstruction.unet_reconstruction.py.
• You can find reco_and_gt_net_from_val_file in fastmri_recon.evaluate.reconstruction.cross_domain_reconstruction.py.
• You can find reco_and_gt_zfilled_from_val_file in fastmri_recon.evaluate.reconstruction.zero_filled_reconstruction.py

[fharman]

Sorry about that.

• You can find reco_and_gt_unet_from_val_file in fastmri_recon.evaluate.reconstruction.unet_reconstruction.py.
• You can find reco_and_gt_net_from_val_file in fastmri_recon.evaluate.reconstruction.cross_domain_reconstruction.py.
• You can find reco_and_gt_zfilled_from_val_file in fastmri_recon.evaluate.reconstruction.zero_filled_reconstruction.py

ok i will revise it. thank you:)

[fharman]

Hi again,

Sorry for disturbance. But i encountered the problem in all_net_params section named NameError: name 'lrelu' is not defined.

This function is derived from nonexisting nn_mri.py. Because lrelu is tried to import from there. Consequently, testing is not run .

Best regards,

[zaccharieramzi]

Hmm but did you use lrelu in your unet model ?

If so then the simplest I can do is just to add in an evaluation notebook to cover the old models.

[fharman]

Hi, i just only used unet from ../models/.functional_models/unet. Zaccharie Ramzi <[email protected]>, 19 Eki 2020 Pzt, 11:02 tarihinde şunu yazdı:

…

Hmm but did you use lrelu in your unet model ? If so then the simplest I can do is just to add in an evaluation notebook to cover the old models. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#101 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AO4IQ2XWDHJD3NE6IPFLYRLSLPW73ANCNFSM4STHUGEA> .

[zaccharieramzi]

If you used the current parameters listed in unet_approach_af4.py you don't need lrelu.

[fharman]

I changed before you mailed me. Thank you for your support, Best regards 19 Eki 2020 Pzt 12:00 tarihinde Zaccharie Ramzi <[email protected]> şunu yazdı:

…

If you used the current parameters listed in unet_approach_af4.py you don't need lrelu. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#101 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AO4IQ2RIJPJLKEKZTEERSLTSLP525ANCNFSM4STHUGEA> .

[fharman]

Hi everyone again,

Sorry for disturbance. But while running the qualitative_validation_for_net for unet , i encountered the error named as

ValueError: You are trying to load a weight file containing 19 layers into a model with 60 layers.

But As you know, anyway unet has 19 layers. I don't understand why i am always facing this error.

Do you have any idea for this?

Best regards,

[zaccharieramzi]

It probably means that you haven't used the same parameters for training and for evaluation. Make sure that the parameters match.

Btw, a U-net doesn't always have 19 layers, it depends on how you define layer and also on how you parametrize the U-net instance.

[fharman]

Sorry i was confused with vgg16 because of training fastmri both VGG16 and UNET at the same time. I changed the path as your folder

chkpt_path = f'/content/fastmri_reproducible_benchmark_master/checkpoints/{run_id}-{epoch}.hdf5' . I think this error is derived because of the model that i trained more less fastmri data.

But after changing the path as you added in fastmri_reproducible_benchmark_master/checkpoints ended with 300 epoch, i encountered another error. I am really sorry about this error disturbance.

Error is

TypeError: To be compatible with tf.contrib.eager.defun, Python functions must return zero or more Tensors; in compilation of <function unpack_model at 0x7f8ace097400>, found return value of type <class 'tensorflow.python.keras.engine.training.Model'>, which is not a Tensor.

Do you have any idea to overcome this error? i searched and tested all way but not overcome.

Best regards,

[zaccharieramzi]

Can you show me the entire stack trace ?

[fharman]

Thank you! 23 Kas 2020 Pzt 14:53 tarihinde Zaccharie Ramzi <[email protected]> şunu yazdı:

…

Closed #101 <#101> via #113 <#113> . — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#101 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AO4IQ2UEE7GPKCSIMUY5DRTSRJEKFANCNFSM4STHUGEA> .