koopa31 / segsrgan Goto Github PK

Hi!
Thanks for your interesting work!
May I know your discriminator parameter number and total model FLOPs? (including generator & discriminator)?

Hello, I am very interested in your research direction, but I have encountered some problems while running your code. I hope you can help me solve them:

1. May I ask what model of host you are using to run the program on? Have you tried it on different models of hosts? I encountered a problem when running the program on a 4090GPU, and I am not sure if the GPU model I am using is not suitable.
2. I noticed that the models of the TensorFlow and other packages displayed in your requests.txt file are relatively old, and some packages in the Pull requests section show that they have been updated. The updated version is not consistent with the requests.txt file. Which one should I run?
   Looking forward to your answer!

There's a path problem in the python file when using pip
https://github.com/koopa31/SegSRGAN/blob/master/SegSRGAN/SegSRGAN/Function_for_application_test_python3.py#L15

I think it'd be better to use pathlib instead of os.path.

Hello!
I have tested with the script seg.py the super resolution reconstruction and segmentation on the first patient of the dHCP dataset. The low resolution (LR) image has been obtained thanks to the module Resample Scalar Volume of 3DSlicer, setting linear interpolation and spacing equal to 0.5x3x0.5 in order to get LR image with axial orientation. I have used weights with no data augmentation, patch=128 and step=30. By comparing the obtained SR volume with the original high resolution volume i have noticed that the brain volume of the first one is bigger than the second one. How does it possible? I have repeated the test also with neonatal clinical images (with weights=Perso_with_constrast_0.5_and_noise_0.03_val_max) but the problem still exists.
I would like also to ask you if it's possible to have the email address of one of the author for further questions.
Thank you!
SR.nii.gz
LR.nii.gz