mrkolarik / 3d-brain-segmentation Goto Github PK

It's a fake model.
Full of error and I think it does not run.

Dear authors:
Recently I have looking at your this project and feel interested in it ! After reading the github introduction, the URL is https://github.com/mrkolarik/3D-brain-segmentation, I face the question. That is, how can I find the dataset? In the project code, I could find the subfolds, such as train, test,mask, and I only find the txt file , like below of the train.txt:

This folder should contain only subfolders with train images which should be in the same order as in ../masks. Original code is written in a way that train images are contained in 21 subfolders each containing 256 images. Data.py lists all images in subfolders and creates train images matrix.

After reading, I could assume there might be 21 subfolders for training, and 1 subfolder for testing? And what is the meaning of the 21 subfolders for masking, and why is the 21 subfolders? Could I obtain the all the needed dataset needed in this project?

Thank you. I am looking forward to your reply Sincerely!

Best regards.

Hello, may I ask what kind of database you use, you can share your database, I have not found a suitable database. I’m sorry to bother you, but I need a dataset now. my email is [email protected]

Hi, the spineweb dataset2 contains vertebrae level labels for vertebrae segmentation. Any suggestions on how to modify/extend 3d dense UNET for vertebrae segmentation?

I have a question about the code of data3D.py. On line 42, img should be a three-dimensional array. Why assign a three-dimensional array to a two-dimensional array? (I'm sorry if this is a stupid question, I'm still a beginner)

could you provided the full code ? about denoising and others ...

Hey, I am trying to apply your network on a new dicom data set. It first has to be manually segmented, I guess different software programms can be used for this task. My question is in which data format should I store the ground truth labels, once they are labelled?
Thanks for your help in advance!

Hi, I doing my project have some problems. when I fit my data to 3D Unet model, I only could use 16 or 32 slices for training.
But.. In my project, my data numpy.npy shape was [20,192,192,1], how can I fix these problems in this model?
And this is my error log.
Sorry..I don't have much experience with keras..

on the paper ,table2, strides is (0,0,0)? that is mean stride =1? and do you use bn or other normalization?

Thank you so much for sharing this fantastic source code.
I have a couple of inquiries:

Could you please clarify the meaning of "total" in both line 21 and line 129 of the "data3D.py" file?
Specifically, I'm seeking an explanation for the relevance of the values 16, 2, and 18.

Additionally, the paper highlights the utilization of 8-bit PNG images.
Hence, I'm curious about the rationale behind setting the "image_depth" in the code to 16.

I am aware of your busy schedule and sincerely value your guidance.

Can you give me the spine dataset?
Thank you !

Where can I download the brain dataset you used in your work?
thanks

I am trying to calculate the dice coefficient using 2D-unet, however, I am getting the following error:

Using TensorFlow backend.
Traceback (most recent call last):
File "E:/2.U_Net_Humerus_Segmentation/U-Net/3D-brain-segmentation-master/2D-unet.py", line 235, in
compare()
File "E:/2.U_Net_Humerus_Segmentation/U-Net/3D-brain-segmentation-master/2D-unet.py", line 226, in compare
K.theano.shared(y_true),
AttributeError: module 'keras.backend' has no attribute 'theano'