search

mohamed-ezz / saratan

Code for MICCAI 2016 paper : Automatic liver and lesions segmentation using Cascaded Fully Convolutional Neural Networks and 3D Conditional Random Fields
39 stars 18 forks source link

why did you use thresholding? #15

Closed lidaryani closed 4 years ago

lidaryani commented 5 years ago

Hi, I see your source and your paper, both of them were very useful. I saw you used thresholding in the last part of network training, but I couldn't find anything in the paper, about it. Could you please explain the matter of using thresholding and the type of thresholding you used?

mohamed-ezz commented 5 years ago

Hi @lidaryani , can you be more specific about which thresholding you are talking about ? point me to somewhere in the code or maybe in the paper?

lidaryani commented 5 years ago

Hi @lidaryani , can you be more specific about which thresholding you are talking about ? point me to somewhere in the code or maybe in the paper?

The "realtime_train.ipynb" file, the part labeled "Changing threshold (instead of 0.5)", actually is the end section of this code.

mohamed-ezz commented 5 years ago

The training portion of the notebook ends at the title "End of training notebook (the rest is one-off analysis)".

The thresholding part was a one-off analysis, i think i was just trying out and visualizing predictions different thresholds other than 0.5. But this is not part of the core training or inference code. This is more of a calibration for the trained network to see what threshold would produce the best dice score.

lidaryani commented 5 years ago

Since I used the same method in my research, I was wondering when I saw your code and I wanted to ask your reason and now I realized that your reason was similar to mine. Thanks for your clear answer.