Issue finding model

[matthieufaillot]

Hi, Thank you for making your great software available
I run into problems when trying to get thin slices from thick slices of Head CT data
I installed KevinSR in a conda python 3.6 env on Ubuntu 22.04
I cloned the repo in Pycharms and run the code on CPU
When running "thins_slices = SOUP_GAN(thicks_ori, Z_FAC, 1)" I get :

 Traceback (most recent call last):
  File "/home/dell/PycharmProjects/SOUP-GAN/Example_MF.py", line 25, in <module>
    thins_slices = SOUP_GAN(thicks_ori, Z_FAC, 1)
TypeError: SOUP_GAN() takes 2 positional arguments but 3 were given

When running thins_slices = SOUP_GAN(thicks_ori, Z_FAC) I get

File "/home/dell/PycharmProjects/SOUP-GAN/Example_MF.py", line 25, in
thins_slices = SOUP_GAN(thicks_ori, Z_FAC)
File "/home/dell/anaconda3/envs/KevinSR/lib/python3.6/site-packages/KevinSR/SOUP.py", line 21, in SOUP_GAN
new_model=keras.models.load_model('./Thin-to-thin')
File "/home/dell/anaconda3/envs/KevinSR/lib/python3.6/site-packages/keras/saving/save.py", line 205, in load_model
return saved_model_load.load(filepath, compile, options)
File "/home/dell/anaconda3/envs/KevinSR/lib/python3.6/site-packages/keras/saving/saved_model/load.py", line 108, in load
meta_graph_def = tf.internal.saved_model.parse_saved_model(path).meta_graphs[0]
File "/home/dell/anaconda3/envs/KevinSR/lib/python3.6/site-packages/tensorflow/python/saved_model/loader_impl.py", line 121, in parse_saved_model
constants.SAVED_MODEL_FILENAME_PB))
OSError: SavedModel file does not exist at: ./Thin-to-thin/{saved_model.pbtxt|saved_model.pb}

Process finished with exit code 1

This is the definition of the SOUP_GAN function in the SOUP.py file :

def SOUP_GAN(thicks_ori, Z_FAC):

thicks_ori = rescale_img(thicks_ori, max_val= 10000)

thins = zoom(thicks_ori, (1,1,Z_FAC))

new_model=keras.models.load_model('./Thin-to-thin')
thins_gen = thins.copy()

target = np.moveaxis(thins,-1,0)
target = target [...,np.newaxis]
target = target [np.newaxis,...]

index  = attention_coeff(target, Z_FAC)

target = new_model.predict([target,index])
 
target_small = target[0,...,0]
thins_gen = np.moveaxis(target_small, 0,-1)

return thins_gen

My understanding is that the SOUP_GAN function
-only takes two arguments ( not three as in example scripts)
-a Thin-to-Thin model by default and that this model is not installed with KevinSR

Is there any way to solve this ? Should I train a model on my own data ?

Best regards
Matthieu

[kevin-vmi]

Thanks for your questions. Below are my comments:

-only takes two arguments ( not three as in example scripts)
Actually, SOUP_GAN() takes three arguments, for example: CT1_sr = SOUP_GAN(CT1_thick, rb, 0)._

-a Thin-to-Thin model by default and that this model is not installed with KevinSR
The function with two types 0: thick-to-thin; 1: thin-to-thin are by default installed with KevnSR.

-Is there any way to solve this ? Should I train a model on my own data ?
Fine-tuning on user's data may improve the performance. Yet, the model was designed to be used directly. Please first check the version of KevinSR installed to see if it is up to date: https://pypi.org/project/KevinSR/

Print dimensions of variables: thicks_ori, Z_FAC could help better understand the issue.

[matthieufaillot]

Thanks for your quick response KevinSR version was 0.0.9, I think because the environment used Python3.6 and when typing "pip3 install KevinSR" it displayed error message for installation of matplotlib (could not find version > 3.4) I created new environment with python3.9 It is now working on CPU although I experience a lot of (memory ?) errors when working with too many slices (or Z_FAC >3) : Process finished with exit code 134 (interrupted by signal 6: SIGABRT) Is there any way to export the predicted images to dicom ? Le sam. 17 sept. 2022 à 22:30, Kevin ***@***.***> a écrit :

…

Thanks for your questions. Below are my comments: -only takes two arguments ( not three as in example scripts) Actually, SOUP_GAN() takes three arguments, for example: CT1_sr = SOUP_GAN(CT1_thick, rb, 0)._ -a Thin-to-Thin model by default and that this model is not installed with KevinSR The function with two types 0: thick-to-thin; 1: thin-to-thin are by default installed with KevnSR. -Is there any way to solve this ? Should I train a model on my own data ? Fine-tuning on user's data may improve the performance. Yet, the model was designed to be used directly. Please first check the version of KevinSR installed to see if it is up to date: https://pypi.org/project/KevinSR/ Print dimensions of variables: thicks_ori, Z_FAC could help better understand the issue. — Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHIU4FT5747T6ZDCGYRBYZTV6YS75ANCNFSM6AAAAAAQO7RYLI> . You are receiving this because you authored the thread.Message ID: ***@***.***>