search

uci-cbcl / DeepLung

WACV18 paper "DeepLung: Deep 3D Dual Path Nets for Automated Pulmonary Nodule Detection and Classification"
GNU General Public License v3.0
153 stars 54 forks source link

About the test results #8

Closed anruoxi123 closed 5 years ago

anruoxi123 commented 5 years ago

Hello, teacher.

I run before the test, in part due to insufficient memory, I use a total of three figures, run frocwrtdetpepchluna16. Py and noduleCADEvaluationLUNA16 py document gets a result;

Recently, I added all the data, and ran the detection part with ten data again. Running the two files above, I got a result.

Oddly, the results of these two runs are exactly the same. What's causing this?

Thanks for comment!

anruoxi123 commented 5 years ago

The results of both tests did not produce an image, but the.txt file did. 9fc2222f155b50ab355f6d71f94cfce 88c0785e0d426836ba66fe963cbee01

wentaozhu commented 5 years ago

Is it the new data or from LUNA16?

anruoxi123 commented 5 years ago

It is LUNA16. I just deleted part of the data folder of the test data set on the basis of the original program. I found that running the frocwrtdetpepchluna16.py file generated the.txt.png.csv file was incorrect,

But there is no error in the running process. What is the problem?

wentaozhu commented 5 years ago

You must use all the test data. Otherwise, it considers the nodules in other part of test data as missing nodule. Maybe it is the reason for low FROC.

anruoxi123 commented 5 years ago

When I run the test part of the code, if I test all the data, the program will be stuck in some data, but the GPU display is running all the time.

So after I deleted the data of the test section, I did not have this problem.

So, is the program stuck because of GPU memory?

I only used two gpus.

wentaozhu commented 5 years ago

Try to use bigger detection threshold

anruoxi123 commented 5 years ago

Ok, then I am training again. Thank you very much!

anruoxi123 commented 5 years ago

Is this threshold size limit?

I went from negative 3 to negative 8, and it's still stuck,

Can I change it to -20?

wentaozhu commented 5 years ago

-2, -1, or 0. Please read our paper for details. Deeplung: Deep 3d dual path nets for automated pulmonary nodule detection and classification, WACV 2018.

anruoxi123 commented 5 years ago

Ok, thank you very much!