search

codedecde / Luna2016-Lung-Nodule-Detection

Course Project for Bio Medical Imaging: Detecting Lung Nodules from CT Scans
MIT License
48 stars 29 forks source link

Regarding the predicted output... #6

Closed bundelesneha05 closed 6 years ago

bundelesneha05 commented 6 years ago

I trained the given U-net model on 1459 samples of LUNA16 database with dice coefficient 0.92 with tensorflow(1.3.0) as well as theano(1.0.1) backend. But it gives me output as below,

predicted_image1

predicted_image

It is not same as the results given in this documentation. How should I solve this problem??? Thanks in advance!

sushmakulkani commented 6 years ago

I am also facing same problem. I am using same database and annotation as in the documentation. Not getting how to proceed for getting results as updated by @shreyaspadhy in the documentation ? Please reply Thanks in advance!!

mrudangpandya commented 6 years ago

i am getting " No such file or directory: 'data.npz'". what data.npz file contents ?

bundelesneha05 commented 6 years ago

@mrudangpandya , have you trained the model by using tensorflow or theano? about your question in result.py file I think, data.npz might contain the numpy files of test and predicted images.

bundelesneha05 commented 6 years ago

My humble request to the @shreyaspadhy and @codedecde, please let us answered about our doubt.

codedecde commented 6 years ago

Try visualising the train data once ? Because if you are getting a dice coefficient of 0.92, that means the model is training, and the predicted mask and the gold mask have a significant overlap. I can take a better look at it after the 7th of May.

shreyaspadhy commented 6 years ago

Hi, unfortunately right now our LUNA_Unet.py doesn't save the results as data.npz by default. I believe I had saved some results by manually modifying the LUNA_Unet.py code to save results into a data.npz file. Here is what I suggest -

In LUNA_UNet.py, after calling model.fit, call model.predict() and save the results as a Python dictionary data, where it has the following keys imgs_test_X (The original images as numpy arrays), imgs_test_Y(the gold label masks), imgs_test_Pred (the output from UNet).

Then you could just use Results_Plot.py (the code for plotting such a dictionary is commented out later, you can just uncomment it).

mrudangpandya commented 6 years ago

ok thank you

I will do this changes..

Thank you

On Wed, Apr 25, 2018 at 7:09 AM, Shreyas Padhy notifications@github.com wrote:

Hi, unfortunately right now our LUNA_Unet.py doesn't save the results as data.npz by default. I believe I had saved some results by manually modifying the LUNA_Unet.py code to save results into a data.npz file. Here is what I suggest -

In LUNA_UNet.py, after calling model.fit, call model.evaluate() and save the results as a Python dictionary data, where it has the following keys imgs_test_X (The original images as numpy arrays), imgs_test_Y(the gold label masks), imgs_test_Pred (the output from UNet).

Then you could just use Results_Plot.py (the code for plotting such a dictionary is commented out later, you can just uncomment it).

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/codedecde/Luna2016-Lung-Nodule-Detection/issues/6#issuecomment-384133921, or mute the thread https://github.com/notifications/unsubscribe-auth/AUWnddzC6Af2m1QpmV_xFcxp8NbGms-Aks5tr9PHgaJpZM4S2cnO .